issue_owner_repo listlengths 2 2 | issue_body stringlengths 0 261k ⌀ | issue_title stringlengths 1 925 | issue_comments_url stringlengths 56 81 | issue_comments_count int64 0 2.5k | issue_created_at stringlengths 20 20 | issue_updated_at stringlengths 20 20 | issue_html_url stringlengths 37 62 | issue_github_id int64 387k 2.46B | issue_number int64 1 127k |
|---|---|---|---|---|---|---|---|---|---|
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-19 02:20:29 +0100
From: relational <<relational>>
To: @njnes
Version: -- development
Last updated: 2007-02-12 09:14:04 +0100
## Comment 6777
Date: 2007-01-19 02:20:29 +0100
From: relational <<relational>>
this one is reproducible using both mserver4 and mserver5. test case is attached.
## Comment 6778
Date: 2007-02-12 21:14:04 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
The query at hand has outer references to the to be updated tables. And has an expression which has an implicite join between two tables. Conclusion we need a 3 way xproduct. This case wasn't detected before. Now thats solved. Reduce the test to a smaller input table, else the 3 way x-product take to much time.
## Comment 6779
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1639108 at http://sourceforge.net/support/tracker.php?aid=1639108
| mserver crashes on delete query | https://api.github.com/repos/MonetDB/MonetDB/issues/1210/comments | 0 | 2020-11-30T09:26:20Z | 2024-06-27T11:22:27Z | https://github.com/MonetDB/MonetDB/issues/1210 | 753,308,614 | 1,210 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-18 04:41:27 +0100
From: Jan Flokstra <<jflokstra>>
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2007-04-21 11:23:05 +0200
## Comment 6773
Date: 2007-01-18 16:41:27 +0100
From: Jan Flokstra <<jflokstra>>
The execution of the test:
$PATHFINDER/modules/pftijah/Tests/qmapi1
fails because it does not load pftijah correct. Some commands from the module can be executed anbd some cannot. For others it makes a difference if it is ADDHELP'ed or not. It looks as if some symbol tables are a mess.
Stefan advised me to create a qmapi.dbinit file which loads module(pftijah) at startuptime. Now the query works. When this file is removed the script will fail on the collName variable of module(pftijah). The function tj_setCollName(str name) is executed (with the help of ADDHELP, without ADDHELP it is also unknown) and it refers to the collName variable but because of this bug it cannot find the variable here.
Hope this is enough info for a bughunt. So to summarize:
- remove $PATHFINDER/modules/pftijah/Tests/qmapi1.dbinit
- run the tests in $PATHFINDER/modules/pftijah
and look what happens.
## Comment 6774
Date: 2007-02-19 13:35:41 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Lowered priority.
## Comment 6775
Date: 2007-04-21 23:23:04 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
With pftijah now being loaded by default whenever pathfinder is loaded,
the problem has disappeared.
All PFtijah tests now run fine without the *.dbinit files;
hence, I removed the latter.
## Comment 6776
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1638729 at http://sourceforge.net/support/tracker.php?aid=1638729
| Mtest/Mapi module load problem | https://api.github.com/repos/MonetDB/MonetDB/issues/1209/comments | 0 | 2020-11-30T09:26:17Z | 2024-06-27T11:22:26Z | https://github.com/MonetDB/MonetDB/issues/1209 | 753,308,585 | 1,209 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-17 07:17:33 +0100
From: @drstmane
To: Henning Rode <<hrode>>
Version: -- development
Last updated: 2007-01-18 08:09:23 +0100
## Comment 6770
Date: 2007-01-17 19:17:33 +0100
From: @drstmane
the *.stable.err files are missing for newly added tests in modules/pftijah/Tests/:
qmil0.milS
qmil2.milS
test_indexmerge.milS
test_nllr.milS
test_split2frag.milS
## Comment 6771
Date: 2007-01-18 20:09:23 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Henning added the missing files, hence, we can close this one.
## Comment 6772
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1637954 at http://sourceforge.net/support/tracker.php?aid=1637954
| PfTijah: *.stable.err missing for new tests | https://api.github.com/repos/MonetDB/MonetDB/issues/1208/comments | 0 | 2020-11-30T09:26:15Z | 2024-06-27T11:22:25Z | https://github.com/MonetDB/MonetDB/issues/1208 | 753,308,554 | 1,208 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-17 05:41:10 +0100
From: @drstmane
To: Jan Rittinger <<jan.rittinger>>
Version: -- development
Last updated: 2009-11-07 02:02:59 +0100
## Comment 6751
Date: 2007-01-17 17:41:10 +0100
From: @drstmane
Since somtime in November or December 2006 (sorry, I'm finally fed up with doing all the back-tracking),
tests
complex-XQuery-batbat_lng_add_inplace-error-OK.SF-1562868.xq
complex-XQuery-batbat_lng_add_inplace-error-KO.SF-1562868.xq
complex-XQuery-batbat_lng_add_inplace-error-KO2.SF-1562868.xq
in
tests/BugTracker/Tests/
fail with
ERROR = !type error: can't cast type 'xs:integer?' to type 'numeric'
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-OK.SF-1562868.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-KO.SF-1562868.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-KO2.SF-1562868.err.00.html
## Comment 6752
Date: 2007-01-17 17:51:34 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Since these tests do work in the XQuery_0-14 release branch, i.e., they have (recently) been broken only in the development trunk, I consider this bug report high priority.
## Comment 6753
Date: 2007-01-17 18:23:40 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Jan,
since you changed the respective tests on Dec 07 2006 (only in the development trunk; see below), you might want to have a look at this and comment.
Feel free to assign to someone else, in case you think this is not your bug.
Thanks!
===================================================================
2006/12/07 - tsheyar:
tests/BugTracker/Tests/complex-XQuery-batbat_lng_add_inplace-error-KO2.SF-1562868.xq,1.2
tests/BugTracker/Tests/complex-XQuery-batbat_lng_add_inplace-error-KO.SF-1562868.xq,1.2
tests/BugTracker/Tests/complex-XQuery-batbat_lng_add_inplace-error-OK.SF-1562868.xq,1.2
-- Fixed output/queries to cope with the last week introduced more restrictive
type tests in the cast operator
===================================================================
## Comment 6754
Date: 2007-01-18 15:15:18 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
This bug occurs because the static typing has been improved in the current branch while mps cannot cope with it.
(smaller example: doc("foo")[./@o cast as xs:integer?])
The reason is that a seqcast is treated like a normal cast in mps. In comparison to a normal cast a seqcast should be used only during typing to 'assert' a certain type. The algebra thus ignores all seqcasts. Mps in comparison however needs to interpret them as cast because it relies on the typing information to refer to the correct value containers.
The only hack to 'avoid' the error message is to introduce a special case for numeric where seqcasts are ignored. The introduced typeswitch up front of already ensures that the type is integer instead of integer?. This information however cannot be inferred as a generic type intersection is impossible.
## Comment 6755
Date: 2007-01-18 15:26:39 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
If the error message is the correct/expected behaviour after the static typing changes for this test, then the error message should be approved as such by who ever changed the static typing and hence triggered the new output --- the respective checkin message should include a hint that this actually renders the test more or less useless wrt. the original bug; even adding a comment to the original bug report about this should be considered.
## Comment 6756
Date: 2007-01-18 16:39:45 +0100
From: Maurice van Keulen <<mvankeulen>>
Logged In: YES
user_id=654674
Originator: NO
The problematic expressions are like "$seq[path cast as xs:integer]" which is not value anymore. What I have found to be accepted is "$seq[string(path) cast as xs:integer]" or "$seq[string(exactly-one(path)) cast as xs:integer]". Apparently, the requested type is numeric, so perhaps 'cast as numeric' will work as well.
I believe we should adapt these testcases, because otherwise they do not test the original problem anymore.
## Comment 6757
Date: 2007-01-18 16:48:27 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Maurice,
it's "your bug", so feel free to fix the tests ;-)
Thank you very much in advance!
Stefan
## Comment 6758
Date: 2007-01-18 17:19:06 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
Fixed complex... queries and introduced new query seqcast_opt_int_to_numeric.SF-1637867 that tests the observed problem.
## Comment 6759
Date: 2007-01-18 18:10:33 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
changed summary (for explanation read my first comment)
## Comment 6760
Date: 2007-01-18 20:08:32 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Jan,
Thank you very much for clearing this up fixing the old test and adding the new one!!
Stefan
## Comment 6761
Date: 2007-02-19 13:38:15 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
What is the status of this bug? Can it be closed? If so, please do so.
## Comment 6762
Date: 2007-02-19 13:53:18 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
The mps version still cannot cope with the predicate that initially triggered the bug. Thus the bug should remain open for now. See also:
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/seqcast_opt_int_to_numeric.SF-1637867.err.00.html
## Comment 6763
Date: 2007-02-19 14:58:46 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Who is responsible for getting this fixed? Please assign somebody.
## Comment 6764
Date: 2007-02-21 14:30:35 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
I checked in a fix that removes seqcast operators that cast to numeric (-- these seqcasts were only generated for predicates). In addition I introduced a core rewrite that replaces the typeswitches that test for the type numeric.
The problematic query now works for me. I'm however not 100% sure whether MPS does break now at a different place or not. Testing will hopefully tell us tomorrow.
## Comment 6765
Date: 2007-02-22 09:26:09 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
Testing did not produce more errors -- Thus I close the bug report.
## Comment 6766
Date: 2009-02-16 00:43:09 +0100
From: @drstmane
re-opened as the test output (still?) differs from the stable one:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d.1-Fedora8/tests_BugTracker/seqcast_opt_int_to_numeric.SF-1637867.out.00.html
## Comment 6767
Date: 2009-02-16 11:13:24 +0100
From: Jan Rittinger <<jan.rittinger>>
This one fails due to incorrect type disjointness test:
http://sourceforge.net/tracker2/?func=detail&aid=2211565&group_id=56967&atid=482468
## Comment 6768
Date: 2009-11-07 14:02:56 +0100
From: @drstmane
closing as all related test seem to work fine now/again:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d.0-Fedora10/tests_BugTracker/seqcast_opt_int_to_numeric.SF-1637867.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d.0-Fedora10/tests_BugTracker/seqcast_opt_int_to_numeric.SF-1637867.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora10/tests_BugTracker/seqcast_opt_int_to_numeric.SF-1637867.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora10/tests_BugTracker/seqcast_opt_int_to_numeric.SF-1637867.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora10/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-OK.SF-1562868.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora10/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-OK.SF-1562868.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora10/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-KO.SF-1562868.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora10/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-KO.SF-1562868.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora10/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-KO2.SF-1562868.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora10/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-KO2.SF-1562868.err.00.html
## Comment 6769
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1637867 at http://sourceforge.net/support/tracker.php?aid=1637867
| XQ: seqcast int? to numeric unsupported | https://api.github.com/repos/MonetDB/MonetDB/issues/1207/comments | 0 | 2020-11-30T09:26:11Z | 2024-06-27T11:22:24Z | https://github.com/MonetDB/MonetDB/issues/1207 | 753,308,511 | 1,207 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-17 05:26:13 +0100
From: @drstmane
To: Jan Rittinger <<jan.rittinger>>
Version: -- development
Last updated: 2007-01-18 09:40:15 +0100
## Comment 6745
Date: 2007-01-17 17:26:13 +0100
From: @drstmane
While working fine until Tuesday Nov 7 2006, the below tests fail (at least) since Saturday Nov 11 2006(!) (there were no pathfinder testing results in between these dates), because syntax/parse error messages now go to the server console as opposed to be set to the client to be handled/displayed there.
See
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XPathMark/Q14.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XPathMark/A3.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XPathMark/A4.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XPathMark/A6.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XPathMark/A7.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugDay_2005-11-09_0.9.3/strange_ampersand.SF-1208620.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugsViaSourgeforce/ID.1208620-2.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_i18n/bad-001.err.00.html
## Comment 6746
Date: 2007-01-17 17:51:31 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Since these tests do work in the XQuery_0-14 release branch, i.e., they have (recently) been broken only in the development trunk, I consider this bug report high priority.
## Comment 6747
Date: 2007-01-17 18:57:37 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
The new behaviour was most probabaly triggered by
===================================================================
2006/11/08 - tsheyar: ./compiler/oops.c,1.22
-- Split up the handling of PFoops, PFinfo and PFlog. Before the change
everything went into a fixed character array (err_buf) and was printed
to stderr after the query was finished (for the call of 'pf' and was
returned as error message for the server side compilation). This was
not feasible anymore as the generated timing information was already
to big.
Now only PFoops and PFinfo print to this array. (I'm however not sure
whether a call to PFinfo should trigger an error in the server side
compilation.) PFlog on the other side now prints directly to stderr.
(Here again I'm not sure whether this does harm server side compilation.)
===================================================================
In case the new behaviour is intended, *please* approve the respective new output as well;
in case the new behaviour is not intended, *please* fix the respective code.
Thanks!
## Comment 6748
Date: 2007-01-18 21:40:15 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
The errors are back in the error buffer (and thus not printed to stderr anymore).
The tests are adjusted and the bug thus is solved. (Thanks Stefan for hunting this one down!)
## Comment 6749
Date: 2007-01-19 01:13:02 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Jan, thank you very much for fixing also this one!
## Comment 6750
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1637853 at http://sourceforge.net/support/tracker.php?aid=1637853
| XQ: syntax/parse error go to server instead of client | https://api.github.com/repos/MonetDB/MonetDB/issues/1206/comments | 0 | 2020-11-30T09:26:08Z | 2024-06-27T11:22:23Z | https://github.com/MonetDB/MonetDB/issues/1206 | 753,308,483 | 1,206 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-17 04:53:29 +0100
From: @drstmane
To: Jan Rittinger <<jan.rittinger>>
Version: -- development
Last updated: 2007-01-19 10:41:25 +0100
## Comment 6738
Date: 2007-01-17 16:53:29 +0100
From: @drstmane
While working fine until Tuesday Jan 9 2007, all XRPC tests fail since (at least) Friday Jan 12 2007 (there were no pathfinder testing results in between these dates), i.e., after checkins on Tuesday Jan 9 2007 through Thursday Jan 11 2007 (which include propagations of changes from the XQuery_0-14 branch that were originally applied there between Friday Jan 05 2007 and Thursday Jan 11 2007).
fatal error: The Pathfinder compiler experienced
an internal problem in the MIL code generation.
You may want to report this problem to the Pathfinder
development team (pathfinder@pathfinder-xquery.org).
When reporting problems, please attach your XQuery input,
as well as the following information:
could not lookup variable 'pf:v_0219' in join recognition.
We apologize for the inconvenience...
See (e.g.):
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_XRpc/add_dbl.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XMark_XRpc/q01.err.00.html
## Comment 6739
Date: 2007-01-17 17:51:24 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Since these tests do work in the XQuery_0-14 release branch, i.e., they have (recently) been broken only in the development trunk, I consider this bug report high priority.
## Comment 6740
Date: 2007-01-17 18:20:40 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Jennie,
maybe you could have a look at this.
Feel free to assign it to someone else, in case you think it's not your bug.
Thanks!
## Comment 6741
Date: 2007-01-19 00:12:15 +0100
From: @yzchang
Logged In: YES
user_id=341633
Originator: NO
This problem is caused by Jan R's checkin on Jan. 09:
====================
From: Jan Rittinger <tsheyar@users.sourceforge.net>
Date: Tue, 09 Jan 2007 14:40:58 +0000
Update of /cvsroot/monetdb/pathfinder/compiler/core
In directory sc8-pr-cvs7.sourceforge.net:/tmp/cvs-serv4167/core
Modified Files:
coreopt.brg fs.brg
Log Message:
...
-- Introduced new core optimizations:
* casts are removed whose input type already matches the cast type
...
@@ -1067,6 +1085,16 @@
}
break;
+ /* CoreExpr: cast (SequenceType, CoreExpr) */
+ case 72:
+ if (PFty_subtype (TY(L(p)), TY(R(p))) &&
+ PFty_subtype (TY(R(p)), TY(L(p)))) {
+ *p = *R(p);
+ rewritten = true;
+ break;
+ }
+ break;
+
/* Atom: var */
case 73:
{
====================
I used the following query and module:
====================
$ cat add_dbl.xq
import module namespace xrpc = "xrpc-fun" at "/ufs/zhang/www/xrpc/xrpcfun.xq";
for $i in (40.5, 50.5)
return execute at {"localhost"} {xrpc:add($i, (10.5, 30.5))}
$ cat /ufs/zhang/www/xrpc/xrpcfun.xq
module namespace xrpcfun = "xrpc-fun";
declare function xrpcfun:add($v1 as xs:decimal,
$v2 as xs:decimal*) as xs:decimal*
{
let $res := $v1
for $i in $v2
return $res + $i
};
====================
The pathfinder compiler gives the following error:
$ pf add_dbl.xq
fatal error: The Pathfinder compiler experienced....
could not lookup variable 'pf:v_0041' in join recognition.
...
halted in ../../../compiler/mil/milprint_summer.c (mps_error), line 158
The variable v_0041 should contain the URI string of the destination of an XRPC call,
but its value is removed during core optimization.
Before the call to reduce() in PFcoreopt() (coreopt.brg line 1190), a fraction of the
XQuery Core tree looks like this:
...
flwr (nil (),
flwr
(let
(letbind (var (pf:v_0040),
apply (fn:data,
arg (lit_str (localhost), nil ()))),
let
(letbind (var (pf:v_0041),
cast (seqtype (string), var (pf:v_0040))),
nil ())),
xrpc (var (pf:v_0041),
...
After the call, the XQuery Core tree look like this:
...
flwr
(let
(letbind (var (pf:v_0022),
seq (lit_dec (40.5), lit_dec (50.5))),
for
(forbind (forvars (var (i), nil ()),
var (pf:v_0022)), nil ())),
xrpc (var (pf:v_0041),
flwr
(let
(letbind (var (pf:v_0032),
seq (lit_dec (10.5), lit_dec (30.5))),
nil ()),
apply (xrpcfun:add,
arg (var (i), arg (var (pf:v_0032), nil ())))))))
...
Obviously, the value of v_0041 was removed during the optimization,
by the new "case 72".
The simplest solution is to remove "case 72", and all XRPC tests work again,
but it might not be Jan's intension...
Jan, I reassign this bug to you. Would you please have a look at this to see
if the implementation of "case 72" can be improved? Otherwise, would you mind
to (temporarily) undo this change? Thank you very much!
Jennie
## Comment 6742
Date: 2007-01-19 10:41:25 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
The bug is now fixed in CVS.
Thanks Jennie for hunting this one down. The bug was hidden in coreopt.brg. It was however not related to my latest changes but only triggered by this rewrite.
The variable expansion that removed the binding did not fill in the replacement because the XRPC operator rule did use the wrong token (var instead of Atom).
## Comment 6743
Date: 2007-01-19 11:20:18 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Great team-work, well done!
Thank you very much!
## Comment 6744
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1637825 at http://sourceforge.net/support/tracker.php?aid=1637825
| PF: XRPC tests fails | https://api.github.com/repos/MonetDB/MonetDB/issues/1205/comments | 0 | 2020-11-30T09:26:05Z | 2024-06-27T11:22:23Z | https://github.com/MonetDB/MonetDB/issues/1205 | 753,308,449 | 1,205 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-17 04:47:50 +0100
From: @drstmane
To: Romulo Goncalves <<goncalve>>
Version: -- development
Last updated: 2007-01-17 10:10:23 +0100
## Comment 6732
Date: 2007-01-17 16:47:50 +0100
From: @drstmane
after recent changes, SQL fails to compile in case MonetDB4 is available:
gcc -m64 -DHAVE_CONFIG_H -I. -I/ufs/manegold/_/scratch0/Monet/Testing/Current/sql/src/backends/monet4 -I../../.. -I/ufs/manegold/_/scratch0/Monet/Testing/Current/sql/src/backends/monet4 -I../../include -I/ufs/manegold/_/scratch0/Monet/Testing/Current/sql/src/backends/monet4/../../include -I../../common -I/ufs/manegold/_/scratch0/Monet/Testing/Current/sql/src/backends/monet4/../../common -I../../storage -I/ufs/manegold/_/scratch0/Monet/Testing/Current/sql/src/backends/monet4/../../storage -I../../server -I/ufs/manegold/_/scratch0/Monet/Testing/Current/sql/src/backends/monet4/../../server -I/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/.GNU.64.64.d.--disable-optimize_--enable-debug_--enable-assert.PREFIX./MONETDB/include/MonetDB -I/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/.GNU.64.64.d.--disable-optimize_--enable-debug_--enable-assert.PREFIX./MONETDB/include/MonetDB/common -I/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/.GNU.64.64.d.--disable-optimize_--enable-debug_--enable-assert.PREFIX./MONETDB/include/MonetDB/gdk -I/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/.GNU.64.64.d.--disable-optimize_--enable-debug_--enable-assert.PREFIX./MONETDB4/include/MonetDB4 -I/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/.GNU.64.64.d.--disable-optimize_--enable-debug_--enable-assert.PREFIX./MONETDB4/include/MonetDB4/monet -I/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/.GNU.64.64.d.--disable-optimize_--enable-debug_--enable-assert.PREFIX./MONETDB4/include/MonetDB4/plain -I/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/.GNU.64.64.d.--disable-optimize_--enable-debug_--enable-assert.PREFIX./MONETDB4/include/MonetDB4/contrib -I/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/.GNU.64.64.d.--disable-optimize_--enable-debug_--enable-assert.PREFIX./MONETDB4/include/MonetDB4/mapi -I/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/.GNU.64.64.d.--disable-optimize_--enable-debug_--enable-assert.PREFIX./MONETDB4/include/MonetDB4/calibrator -DLIBSQL_SERVER -Wall -W -std=c99 -g -Werror-implicit-function-declaration -Werror -Wno-unused-function -Wno-unused-label -c sql_gencode.c -fPIC -DPIC -o .libs/lib_sql_server_la-sql_gencode.o
cc1: warnings being treated as errors
/ufs/manegold/_/scratch0/Monet/Testing/Current/sql/src/backends/monet4/sql_gencode.mx: In function 'backend_dumpstmt_':
/ufs/manegold/_/scratch0/Monet/Testing/Current/sql/src/backends/monet4/sql_gencode.mx:346: warning: enumeration value 'st_connection' not handled in switch
make[7]: *** [lib_sql_server_la-sql_gencode.lo] Error 1
## Comment 6733
Date: 2007-01-17 16:55:44 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: NO
Yes you are right.
In my pc I only have M5 and not M4. I had the idea that M4 was dead.
I will install M4 and correct the problem.
Should I run also the Mtest for M4?
Regards,
Romulo
## Comment 6734
Date: 2007-01-17 17:57:17 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
As long as SQL accepts M4, it should also compile with M4; otherwise, configure should already tell that M4 is not supported any longer, or simply ignore M4, in case M5 is available.
But until then, please make sql compile also if M4 is available.
In case it is too complicated to back-port your new functionality to M4, just make sure that it (a) compiles, and (b) gives a proper error messages in case a user tries to use with M4 the new functionality that is only supported with M5.
Thanks!
## Comment 6735
Date: 2007-01-17 22:10:23 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: NO
This problem was resolved and also checked in.
Regards,
Romulo
## Comment 6736
Date: 2007-01-17 23:09:15 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
The problem was (presumably) fixed; hence, we say so.
;-)
## Comment 6737
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1637820 at http://sourceforge.net/support/tracker.php?aid=1637820
| SQL fails to compile in case MonetDB4 is available | https://api.github.com/repos/MonetDB/MonetDB/issues/1204/comments | 0 | 2020-11-30T09:26:02Z | 2024-06-27T11:22:22Z | https://github.com/MonetDB/MonetDB/issues/1204 | 753,308,390 | 1,204 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-17 04:20:32 +0100
From: @drstmane
To: Jan Rittinger <<jan.rittinger>>
Version: -- development
Last updated: 2007-01-18 02:23:00 +0100
## Comment 6727
Date: 2007-01-17 16:20:32 +0100
From: @drstmane
While working fine until Tuesday Jan 9 2007, the following tests fails since (at least) Friday Jan 12 2007 (there were no pathfinder testing results in between these dates), i.e., after checkins on Tuesday Jan 9 2007 through Thursday Jan 11 2007 (which include propagations of changes from the XQuery_0-14 branch that were originally applied there between Friday Jan 05 2007 and Thursday Jan 11 2007).
Tests qa04alg, qj01alg, qj02alg, qj03alg, qj04alg,
in benchmarks/MBench/Tests/
fail with
!ERROR: assert_order(tmp_263): tail was not ordered.
See
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_MBench/qa04alg.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_MBench/qj01alg.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_MBench/qj02alg.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_MBench/qj03alg.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_MBench/qj04alg.out.00.html
## Comment 6728
Date: 2007-01-17 17:51:19 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Since these tests do work in the XQuery_0-14 release branch, i.e., they have (recently) been broken only in the development trunk, I consider this bug report high priority.
## Comment 6729
Date: 2007-01-17 18:24:30 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Jan,
algebra is your "baby" ;-)
## Comment 6730
Date: 2007-01-18 14:23:00 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
The problem is that assert_order (of course) cannot cope with reverse order. I introduced reverse order to support 'order by descdening'. A reverse order however is sometimes also planned for queries without order by. For now the solution is to leave out assert_order.
(Adding an assert_rev_order proc could help merged_union -- the command that requires the order in most cases -- but only if it also copes with reverse input orderings, which is as far as I know not the case.)
(A fix is already checked in.)
## Comment 6731
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1637798 at http://sourceforge.net/support/tracker.php?aid=1637798
| PF/algebra: tests fails in benchmarks/MBench | https://api.github.com/repos/MonetDB/MonetDB/issues/1203/comments | 0 | 2020-11-30T09:25:59Z | 2024-06-27T11:22:21Z | https://github.com/MonetDB/MonetDB/issues/1203 | 753,308,352 | 1,203 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-17 02:43:19 +0100
From: @drstmane
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2007-01-22 07:19:08 +0100
## Comment 6721
Date: 2007-01-17 14:43:19 +0100
From: @drstmane
Since Friday Jan 12 (i.e., after changes on Thursday Jan 13) test tests/BugDay_2005-10-06_4.9.3/Tests/rename_bat_hyphen.SF-1152369.milS produces
!ERROR: yytree: rename: 3.thash.
instead of the (formerly) expected
!ERROR: yytree: rename: x-.
This might actually be non-crucial or even a feature / intended, but before investing time in investigating it, I'd like to hear whether anybody happens to know what changed this error messages, and whether we can simply accept and approve this change....
## Comment 6722
Date: 2007-01-17 15:23:23 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
forgot:
for details, see
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB4/.mTests103/GNU.64.64.d-Fedora4/tests_BugDay_2005-10-06_4.9.3/rename_bat_hyphen.SF-1152369.out.00.html
## Comment 6723
Date: 2007-01-17 15:45:37 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
On Windows with the MS compiler, I get as output:
!ERROR: yytree: rename: ÛÛÛÛÛÛÛÛ.
## Comment 6724
Date: 2007-01-17 17:27:39 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Assigning to Peter. I have the distinct impression that the error is caused by the recently added line
*res = v->binding;
in interpret_var. The value of *res is assigned there basically a copy of what it already contains, but the string value is apparently freed before it is printed.
## Comment 6725
Date: 2007-01-22 18:24:21 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
fixed: the string should not be printed in the first place
## Comment 6726
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1637725 at http://sourceforge.net/support/tracker.php?aid=1637725
| test rename_bat_hyphen.SF-1152369 produces different output | https://api.github.com/repos/MonetDB/MonetDB/issues/1202/comments | 0 | 2020-11-30T09:25:57Z | 2024-06-27T11:22:20Z | https://github.com/MonetDB/MonetDB/issues/1202 | 753,308,310 | 1,202 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-17 12:58:30 +0100
From: Jan Rittinger <<jan.rittinger>>
To: Jens Teubner <<teubner>>
Version: -- development
Last updated: 2007-01-17 01:16:47 +0100
## Comment 6718
Date: 2007-01-17 12:58:30 +0100
From: Jan Rittinger <<jan.rittinger>>
Is the following warning/error a problem?
configure: creating ./config.status
config.status: creating Makefile
config.status: creating mel_config.h
config.status: mel_config.h is unchanged
config.status: executing depfiles commands
configure: creating ./config.status
./config.status: line 301: Buildtools config.status 1.0
configured by /local_tmp/buildtools/configure, generated by GNU Autoconf 2.59,
with options "'--enable-shared' '--disable-static' '--enable-bi1970-01-01 01:00:32 +0100' '--prefix=/local_tmp/buildtools/Linux'"
Copyright (C) 2003 Free Software Foundation, Inc.
This config.status script is free software; the Free Software Foundation
gives unlimited permission to copy, distribute and modify it.: No such file or directory
config.status: creating Makefile
config.status: creating autogen/Makefile
config.status: creating conf/Makefile
config.status: creating dummy.h
config.status: dummy.h is unchanged
configure: configuring in Mx
## Comment 6719
Date: 2007-01-17 13:13:39 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: YES
replacing AC_INIT("MonetDB Buildtools", 1.0, monet@cwi.nl)
by AC_INIT([MonetDB Buildtools], 1.0, monet@cwi.nl)
seems to help
## Comment 6720
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1637630 at http://sourceforge.net/support/tracker.php?aid=1637630
| buildtools: configure - file not found | https://api.github.com/repos/MonetDB/MonetDB/issues/1201/comments | 0 | 2020-11-30T09:25:54Z | 2024-06-27T11:22:19Z | https://github.com/MonetDB/MonetDB/issues/1201 | 753,308,282 | 1,201 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-17 11:36:08 +0100
From: Maurice van Keulen <<mvankeulen>>
To: GDK devs <<bugs-common>>
Version: -- development
Last updated: 2007-01-17 11:51:40 +0100
## Comment 6714
Date: 2007-01-17 11:36:08 +0100
From: Maurice van Keulen <<mvankeulen>>
(milprint_summer version on Suse 9)
The query below selects the first and third element from the sequence $all. On each iteration of $i, $all[$i] returns all four elements of $all instead of only one. If you replace the $num2 in the for with $num2, the result is correct.
<res>{
let $all:=(<a/>,<b/>,<c/>,<d/>)
,$nums1:=(1,3)
,$nums:=<nums><num i="1"/><num i="3"/></nums>
,$nums2:=$nums/num/@i
for $i in $nums2
return <elem num="{$i}">{$all[$i]}</elem>
}</res>
Intended result:
<res>
<elem num="1">
<a/>
</elem>
<elem num="3">
<c/>
</elem>
</res>
Computed result:
<res>
<elem num="1">
<a/>
<b/>
<c/>
<d/>
</elem>
<elem num="3">
<a/>
<b/>
<c/>
<d/>
</elem>
</res>
## Comment 6715
Date: 2007-01-17 11:51:40 +0100
From: @Teggy
Logged In: YES
user_id=642632
Originator: NO
Your result is correct in both cases. Note that $num2 is bound to a sequence of two *attribute nodes*, not to two xs:integer values.
As a result, in the for-loop $i will be bound to a node and the effective Boolean value of $i will be 'true'. That's why $all[$i] returns all nodes in the $all sequence.
Cheers,
--Teggy
## Comment 6716
Date: 2007-01-17 14:09:36 +0100
From: Maurice van Keulen <<mvankeulen>>
Logged In: YES
user_id=654674
Originator: YES
Hmmm ... I should have noticed this. This behaviour of predicates is awkward. Even "$all[data($i)]" does not work (apparently the effective Boolean value of a string is also true). Only "$all[$i cast as xs:integer]" seems to work. Any other (nicer) way to retrieve the integer value of an attribute?
## Comment 6717
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1637554 at http://sourceforge.net/support/tracker.php?aid=1637554
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-common@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| Duplicates in sequence lookup | https://api.github.com/repos/MonetDB/MonetDB/issues/1200/comments | 0 | 2020-11-30T09:25:51Z | 2024-06-28T07:36:52Z | https://github.com/MonetDB/MonetDB/issues/1200 | 753,308,250 | 1,200 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-16 03:23:28 +0100
From: @MarcinZukowski
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2007-02-19 09:18:06 +0100
## Comment 6710
Date: 2007-01-16 15:23:28 +0100
From: @MarcinZukowski
I have some problems with fork, for example:
$ cat y.mil:
fork(print(1));
$ Mserver y.mil
MonetDB>!ERROR: interpret: no matching MIL operator to 'ÛÛÛÛÛÛÛÛ (int)'.
But sometimes it does:
$ cat x.mil
module(unix);
module(alarm);
PROC blah() : void {
bat("blah").insert(1,1);
commit();
}
destroy("blah"); commit();
var x := new(int, int).rename("blah").persists(true);
commit();
fork(blah());
sleep(1);
bat("blah").print();
$ Mserver x.mil
-----------------
h t name
int int type
-----------------
[ 1, 1 ]
$ Mserver < x.mil
!ERROR: interpret: no matching MIL operator to 'sleep()'.
!MAYBE YOU MEAN:
! sleep(int) : void
-----------------
h t name
int int type
-----------------
Any ideas?
## Comment 6711
Date: 2007-02-15 16:59:54 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Peter, I'm assigning this to you. I have investigated what the problem is, but I don't have a solution. And it's your code. Please read the below.
This report is likely related to http://sourceforge.net/support/tracker.php?aid=1640248: "PF: ~500 tests randomly fail on Solaris x86".
This is also related to the recent problem in the test web where lots of Pathfinder tests fail on various 32-bit systems with the message "!ERROR: interpret: no matching MIL operator to 'pn-realaudio()'."
This is what is happening (in the Pathfinder case):
In pathfinder.mil (generated from pathfinder.mx) there is code
fork(pf_logmanager());
When this is executed, the CMDfork() implementation in builtin.mx calls forkClient() and then makes a copy of the parse tree with Myycopy() for the argument and queues that copy with QMenqueue in the queue manager. The copy is a shallow copy: the structure of the tree is copied to a new structure, but no copy is made of the data that the yyval structure contained in there points to. In particular, the sval/pval values are shared with the original. The copy does get the bit iscopy set.
After this is done it so happens that the pathfinder module continues to be executed until the end.
Once the end of the module is reached, the parse tree for the module is freed. This includes the original for the argument to the fork() call and also (and this is the important bit) the buffers that the sval/pval pointers in that original point to. In the case here, the parse tree is very small, consisting only of a TOK_FUNCTION node with a string valued argument with value "pf_logmanager".
Once this is done, the queued events get a chance to run. Before the call to fork(pf_manager()) there was a call to rpcd_start() which starts the httpd server. This server makes a copy of MIME types, among others "audio/x-pn-realaudio". It so happens that this copy ends up in the memory that was previously occupied by the "pf_logmanager" string. Note that "pf_logmanager" was allocated with GDKmalloc which has an overhead (at the start of the memory) of 8 (on 32 bit systems) bytes, and note that when you skip 8 bytes in "audio/x-pn-realaudio" you end up at "pn-realaudio".
The next event which is handled is the fork(pf_logmanager()) argument. By now the yyval.val.sval buffer was GDKfree()ed and malloc()ed to contain "pn-realaudio" and so the interpreter tries to execute the function pn-realaudio which of course fails.
To summarize, the problem is with the memory management during fork(). We need to make a copy of the parse tree because of the yychunking that happens these days. We cannot (I think) set the yysons pointer in the original parse tree to NULL since then that bit of parse tree would never get freed. We could perhaps allocate memory for the yyval.val.sval/pval values (if that is what the yyval structs represent), but that would incur a lot of malloc/free costs which we may not want to pay, plus, we may not know how large the data is that would need to be copied.
Since this also happens in the stable release I changed the Group.
## Comment 6712
Date: 2007-02-19 09:18:06 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
It looks like Peter's change did indeed fix the problem.
No test since 1) this is hard to reproduce in a test, and 2) it seems that importing pathfinder on FC6 happened to trigger the bug reliably.
## Comment 6713
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1636828 at http://sourceforge.net/support/tracker.php?aid=1636828
| fork() behaves strangely | https://api.github.com/repos/MonetDB/MonetDB/issues/1199/comments | 0 | 2020-11-30T09:25:49Z | 2024-06-27T11:22:17Z | https://github.com/MonetDB/MonetDB/issues/1199 | 753,308,219 | 1,199 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-16 01:50:25 +0100
From: Jens Teubner <<teubner>>
To: @sjoerdmullender
Version: -- development
Last updated: 2007-03-01 05:50:02 +0100
## Comment 6706
Date: 2007-01-16 13:50:25 +0100
From: Jens Teubner <<teubner>>
I am building Pathfinder directly in the source tree. Whenever I do a `make clean', however, both HowToStart-* files are being removed from the tree. A subsequent `make' then fails with ``no rule to make target...''.
## Comment 6707
Date: 2007-01-17 17:00:21 +0100
From: Jens Teubner <<teubner>>
Logged In: YES
user_id=731390
Originator: YES
Similar problems also arise in other sub-projects. In the
`MonetDB' sub-project, e.g., the `MonetDB.m4' file is
deleted during `make clean'.
## Comment 6708
Date: 2007-03-01 17:50:02 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
This is now fixed in the stable branch with a fix to autogen.
## Comment 6709
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1636756 at http://sourceforge.net/support/tracker.php?aid=1636756
| Too many files removed during `make clean' | https://api.github.com/repos/MonetDB/MonetDB/issues/1198/comments | 0 | 2020-11-30T09:25:46Z | 2024-06-27T11:22:16Z | https://github.com/MonetDB/MonetDB/issues/1198 | 753,308,194 | 1,198 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-16 10:03:23 +0100
From: Arthur van Bunningen <<arthurvb>>
To: Jan Rittinger <<jan.rittinger>>
Version: -- development
Last updated: 2008-06-02 12:05:23 +0200
## Comment 6692
Date: 2007-01-16 10:03:23 +0100
From: Arthur van Bunningen <<arthurvb>>
Introduction:
The attached file p.xml contains different "possible worlds" about movies together with their probability. The goal is to return the movies that are thriller movies together with the probability that they are thriller movies. We do this by for each movie that is a thriller movie (in some world) summing the probabilities of all the worlds in which it actually is a thriller movie.
The problem:
If we do the following query (attached as sfquery.xq) in pathfinder 0.15.0 (HEAD version, checked out last Friday).
let $worlds := exactly-one(doc("/home/db/bunninge/phd/implementation/xquery/maurice/p.xml")/worlds)
let $ws := for $w in $worlds/world
return <r>{$w/@prob}{$w//genre[.="Thriller"]/ancestor::movie/title}</r>
for $v in distinct-values($ws/title)
let $rank := sum($ws[//title=$v]/@prob)
return
<res rank="{$rank}">{$v}</res>
This results in the following results (attached as sfresults_pf.xml):
<res rank="0.012197519800210664">King Kong</res>,
<res rank="0.012197519800210664">Die Hard: With a Vengeance</res>,
<res rank="0.0067562687537343277">Die Hard</res>,
<res rank="0.0054412510464763482">Die Hard 2</res>
These results seem to be wrong since if we just return the nodes in $ws (using sfquery_ws.xq as attached) we can see that there are "worlds" in which King Kong is a thriller with a probability higher than 0.35
Testing in Qizx/open studio seems to give more reasonable results for sfquery.xq (see sfresults_qizx.xml):
<?xml version='1.0' encoding='UTF-8'?>
<res rank="1.000000000000013">King Kong</res>
<res rank="1.000000000000013">Die Hard: With a Vengeance</res>
<res rank="0.7941176470588303">Die Hard 2</res>
<res rank="0.7941176470588311">Die Hard</res>
I did some tests and it seemed that even doing only a name-test predicate on title in the sum, does not give all results, whereas it does work when the test is done in the where-clause. However, since I am not able to test this extensively I will just report the original problem.
Thank you in advance,
Arthur
## Comment 6693
Date: 2007-01-16 10:04:24 +0100
From: Arthur van Bunningen <<arthurvb>>
Logged In: YES
user_id=1045637
Originator: YES
File Added: sfquery.xq
## Comment 6694
Date: 2007-01-16 10:06:38 +0100
From: Arthur van Bunningen <<arthurvb>>
Logged In: YES
user_id=1045637
Originator: YES
File Added: sfquery_ws.xq
## Comment 6695
Date: 2007-01-16 10:26:04 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
out of "pure curiosity":
which version of MonetDB/XQuery are you using / refering to?
(Please see / set / use the "Group:" in the header.)
Thanks!
## Comment 6696
Date: 2007-01-16 12:07:20 +0100
From: Arthur van Bunningen <<arthurvb>>
Logged In: YES
user_id=1045637
Originator: YES
I added the group Pathfinder CVS Head (checked out last Friday).
## Comment 6697
Date: 2007-01-16 12:32:33 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
Hi Arthur,
The query gives the expected result, if you would use
sum($ws[.//title=$v]/@prob)
instead of
sum($ws[//title=$v]/@prob)
That is, the context node for //title is not the $ws elementm, but fn:root($ws).
However, Qizx is right about the answer, because in this case, each $ws is in fact the root of a newly created fragment.
There was a bug in the element construction of mps (and algebra!). FRAG_ROOT contains:
head => the OPEN_DOCID, which allows to find documents within a collection.
tail => the root NID
In case of the temporary doc collection, the heads are all oid_nil (we cannot find tmp fragments with the doc() function) and the tails are PREs (PRE=NID=RID for temporary docs).
So basically, the format of FRAG_ROOT had been reversed in 4.14, and we forgot to update the node construction code.
fixed now
## Comment 6698
Date: 2007-01-16 17:30:23 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
added tests in
pathfinder/tests/BugTracker/Tests/predicate_selects_too_few_nodes.SF-1636588.*
pathfinder/tests/BugTracker/Tests/predicate_selects_too_few_nodes.SF-1636588_ws.*
seem to work fine.
Peter, Arthur,
could please check/verify whether the provided stable output are indeed correct / as expected?
(XQuery_0-14 branch; propagation to development trunk will follow as soon
as the recently introduced bugs/problems in the Stable branch are fixed)
## Comment 6699
Date: 2007-01-16 22:32:25 +0100
From: Arthur van Bunningen <<arthurvb>>
Logged In: YES
user_id=1045637
Originator: YES
I do not know how to test the test scripts, but at least I can confirm that the result of my original query is correct in the latest stable build.
Thank you.
Regards,
Arthur
## Comment 6700
Date: 2007-01-17 18:30:16 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Arthur,
see `Mtest.py --help`, pathfinder/HowToStart-PF, and/or http://monetdb.cwi.nl/Development/TestWeb/Background/index.html & http://monetdb.cwi.nl/monet/src/testing/README for details how to run tests.
Further, simply looking at pathfinder/tests/BugTracker/Tests/predicate_selects_too_few_nodes.SF-1636588*.stable.out probably tells you, whether this is the output that you expect.
Given that it seems to work for you, I feel free to colse this bug report.
Stefan
## Comment 6701
Date: 2008-06-01 23:39:18 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Re-opening:
While working fine with the MPS back-end, the tests fails with the ALG back-end, mainly as it seems to calculate different ranks (via fn:sum()); cf.
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTestsG103/GNU.64.64.d-Fedora8/tests_BugTracker/predicate_selects_too_few_nodes.SF-1636588.out.00.html
## Comment 6702
Date: 2008-06-02 00:02:20 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
See also
[ 1751684 ] XQ: merged_union error with fn:sum()
https://sourceforge.net/tracker/index.php?func=detail&aid=1751684&group_id=56967&atid=482468
## Comment 6703
Date: 2008-06-02 12:05:19 +0200
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
It turned out that this was a problem introduced on December 6.th, 2007. I changed back the implementation of unordered positions and approved the algebra specific output.
## Comment 6704
Date: 2008-06-02 12:05:22 +0200
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
It turned out that this was a problem introduced on December 6.th, 2007. I changed back the implementation of unordered positions and approved the algebra specific output.
## Comment 6705
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1636588 at http://sourceforge.net/support/tracker.php?aid=1636588
| XQ: Predicate selects too few nodes | https://api.github.com/repos/MonetDB/MonetDB/issues/1197/comments | 0 | 2020-11-30T09:25:43Z | 2024-06-27T11:22:15Z | https://github.com/MonetDB/MonetDB/issues/1197 | 753,308,155 | 1,197 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-15 03:53:07 +0100
From: Jens Teubner <<teubner>>
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2007-01-16 05:27:32 +0100
## Comment 6687
Date: 2007-01-15 15:53:07 +0100
From: Jens Teubner <<teubner>>
Sorry, Peter. But you asked for it...
The predicate rewrite magic in normalize.brg incorrectly rewrites the query
declare namespace xx = 'http://www.w3.org/2005/xpath-functions';
let $d :=
<a>
<b>
<c x='3'>no hit</c>
</b>
<b>
<c x='2'>no hit</c>
</b>
<b>
<c x='1'>hit!</c>
</b>
</a>
return
$d/b/c[@x = xx:position()]
such that the predicate is bound to the entire path expression, not just the final step. The effect is a wrong query result.
## Comment 6688
Date: 2007-01-15 17:37:33 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
fixed
I removed the check as I assume that at this stage of parsing we do not yet have access to bound namespaces..
## Comment 6689
Date: 2007-01-15 18:39:02 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
re-opened as no test has been provided, yet ...
## Comment 6690
Date: 2007-01-16 17:27:32 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
added test in pathfinder/tests/BugTracker/Tests/Attic/predicate_rewriting.SF-1635936.*
seems to work fine
(XQuery_0-14 branch; propagation to development trunk will follow as soon as the recently introduced bugs/problems in the Stable branch are fixed)
## Comment 6691
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1635936 at http://sourceforge.net/support/tracker.php?aid=1635936
| PF: Predicate rewriting incorrect | https://api.github.com/repos/MonetDB/MonetDB/issues/1196/comments | 0 | 2020-11-30T09:25:41Z | 2024-06-27T11:22:14Z | https://github.com/MonetDB/MonetDB/issues/1196 | 753,308,119 | 1,196 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-15 11:38:28 +0100
From: Riham A. K. <<rihamabdelkader>>
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2007-01-17 06:38:14 +0100
## Comment 6682
Date: 2007-01-15 11:38:28 +0100
From: Riham A. K. <<rihamabdelkader>>
Hi all,
I'm having a problem with installing MonetDB. I'm doing all the
installation with clean directories. I have three directories: src,
build and monetdb.
buildtools seems to work, but when bootstrapping monetDB I am getting
the following message:
./bootstrap: line 24: exec: Mbootstrap: not found
Any idea what's wrong?
Thanks,
Riham Abdel Kader
## Comment 6683
Date: 2007-01-15 11:41:54 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Do you have <[buildtools_]prefix>/bin in you PATH?
If now, add it and try again.
## Comment 6684
Date: 2007-01-17 18:25:34 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Riham,
has your problem been solved?
I.e., can we close this bug report?
## Comment 6685
Date: 2007-01-17 18:38:14 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Riham confirmed "fix" via private email.
## Comment 6686
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1635748 at http://sourceforge.net/support/tracker.php?aid=1635748
| Mbootstrap Problem | https://api.github.com/repos/MonetDB/MonetDB/issues/1195/comments | 0 | 2020-11-30T09:25:38Z | 2024-06-27T11:22:13Z | https://github.com/MonetDB/MonetDB/issues/1195 | 753,308,092 | 1,195 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-12 09:26:58 +0100
From: @drstmane
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2007-01-12 03:13:51 +0100
## Comment 6679
Date: 2007-01-12 09:26:58 +0100
From: @drstmane
After yesterday's (Thursday Jan 11 2007, 01:27:30 CET) changes,
dropping module salgebra hangs during salgebra_epilogue(),
most probably looping endlessly in src/gdk/gdk_bbp.mx's incref() (lines 1241-1247).
See attached log for details.
See also
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/MonetDB/.mTests103/GNU.64.64.d-Fedora4/tests_sigs/none.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/MonetDB/.mTests103/GNU.64.64.d-Fedora4/tests_sigs/none.err.00.html
## Comment 6680
Date: 2007-01-12 13:50:44 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
The respective changes also seem to cause/trigger MonetDB5 test src/mal/Tests/tst232.mal to hang:
http://monetdb.cwi.nl/testing/intern/monetdb/Current/_intern_/MonetDB5/.mTests103/GNU.64.64.d-Fedora4/src_mal/tst232.out.00.html
http://monetdb.cwi.nl/testing/intern/monetdb/Current/_intern_/MonetDB5/.mTests103/GNU.64.64.d-Fedora4/src_mal/tst232.err.00.html
## Comment 6681
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1633846 at http://sourceforge.net/support/tracker.php?aid=1633846
| dropping modules salgebra hangs during epilogue in incref | https://api.github.com/repos/MonetDB/MonetDB/issues/1194/comments | 0 | 2020-11-30T09:25:36Z | 2024-06-27T11:22:12Z | https://github.com/MonetDB/MonetDB/issues/1194 | 753,308,069 | 1,194 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-11 03:32:20 +0100
From: Jens Teubner <<teubner>>
To: XQuery devs <<bugs-xquery>>
Version: -- development
Last updated: 2007-05-25 12:27:32 +0200
## Comment 6673
Date: 2007-01-11 15:32:20 +0100
From: Jens Teubner <<teubner>>
The memory reallocation routine pa_realloc (compiler/mem/mem.c) currently does
char *r = pa_alloc( pa, new_size);
memcpy(r, old_mem, new_size);
(i.e., a new memory block of the requested size is allocated, then old contents are copied to the new memory area. As we don't know the size of the old area, the routine simply copies an area of the size of the *new* memory from old to new.)
This is problematic for two reasons:
-- memcpy must be invoked with *non-overlapping*
memory regions only. It is easily possible (in
fact, even the common case) that the two regions
do overlap.
-- There is no protection that memcpy() tries to
read memory that is beyond the memory area
allocated from the operating system. A
segmentation fault could be the consequence.
## Comment 6674
Date: 2007-01-11 15:39:56 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
A quick grep shows no usage of the pa_realloc, so why not simply remove it?
## Comment 6675
Date: 2007-01-11 15:45:50 +0100
From: Jens Teubner <<teubner>>
Logged In: YES
user_id=731390
Originator: YES
A quick grep shows no usage of the pa_realloc, because its
invocations have been elegantly hidden in a stack of macro
and function calls.
For example,
PFarray_at() -> PFrealloc -> PFrealloc_ -> PA_REALLOC -> pa_realloc
## Comment 6676
Date: 2007-01-11 16:01:47 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
The pa_alloc code wasn't made for realloc and free strategies, it simply allocates and grows. At the end of a statement, the allocator should be freed in one go.
So using pa_* for general (global) allocation is not correct.
PFrealloc is used in 4 places
0) compiler/main.c, simply not a good idea to use the pa_alloc code here. Use malloc/realloc,free (strdup) instead.
1) PFarray_at (there you have the correct old size)
2) mil_opt. (should also be replaced by malloc,realloc,free (possibly the GDK variants)
3) milprint_summer Also there, the pointer created lost, ie better to correctly use, malloc,realloc and free
## Comment 6677
Date: 2007-05-25 12:27:32 +0200
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
fixed
## Comment 6678
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1633232 at http://sourceforge.net/support/tracker.php?aid=1633232
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-xquery@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| PF: memory reallocation buggy | https://api.github.com/repos/MonetDB/MonetDB/issues/1193/comments | 0 | 2020-11-30T09:25:33Z | 2024-06-27T11:22:11Z | https://github.com/MonetDB/MonetDB/issues/1193 | 753,308,043 | 1,193 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-11 12:20:29 +0100
From: @drstmane
To: Jan Flokstra <<jflokstra>>
Version: -- development
Last updated: 2007-01-11 12:27:30 +0100
## Comment 6670
Date: 2007-01-11 12:20:29 +0100
From: @drstmane
pathfinder/modules/pftijah/termdb.mx
fails to compile with
/export/scratch0/monet/monet.GNU.64.64.d.4778/pathfinder/modules/pftijah/termdb.mx: In function 'tdb_insertTerm':
/export/scratch0/monet/monet.GNU.64.64.d.4778/pathfinder/modules/pftijah/termdb.mx:549: warning: overflow in implicit constant conversion
as function
int tdb_insertTerm(termDbCtx* tdb, char* term, oid o, int check)
tries to return an oid instead of an int:
return oid_nil;
## Comment 6671
Date: 2007-01-11 12:27:30 +0100
From: Jan Flokstra <<jflokstra>>
Logged In: YES
user_id=1054297
Originator: NO
Repaired mixed up return value. This should be 0.
## Comment 6672
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1633093 at http://sourceforge.net/support/tracker.php?aid=1633093
| PfTijah: modules/pftijah/termdb.mx fails to compile | https://api.github.com/repos/MonetDB/MonetDB/issues/1192/comments | 0 | 2020-11-30T09:25:31Z | 2024-06-27T11:22:10Z | https://github.com/MonetDB/MonetDB/issues/1192 | 753,308,013 | 1,192 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-11 11:34:58 +0100
From: Jan Rittinger <<jan.rittinger>>
To: GDK devs <<bugs-common>>
Version: -- development
Last updated: 2007-01-23 01:11:31 +0100
## Comment 6666
Date: 2007-01-11 11:34:58 +0100
From: Jan Rittinger <<jan.rittinger>>
Leftjoin in my script returns only a single row instead of a single hit for each tuple in the left relation.
As sf complains about the file size get the two files at
http://www-db.in.tum.de/~rittinge/query_gm7.mil (script)
http://www-db.in.tum.de/~rittinge/dblp.tar.gz (xml file & DTD)
(The path to the DTD in the XML file probably needs to be adjusted.)
-- Sorry for the inconvinience, but I couldn't reproduce the bug on a smaller example.
## Comment 6667
Date: 2007-01-23 10:39:36 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Jan,
did you run Mserver with the default [-d0], i.e., property checking disabled,
or (also) with property checking enabled (-d10).
If the latter, did the same problem occur?
If (only) the former could you re-run with -d10, and report the outcome?
Thanks!
## Comment 6668
Date: 2007-01-23 13:11:31 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: YES
Thanks for the advice, Stefan!
It turned out that the query contained some assert_order() calls on relations that were sorted descending. (This bugwas already fixed some days ago). The subsequent join that probably relied on the order thus returned the wrong result.
## Comment 6669
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1633060 at http://sourceforge.net/support/tracker.php?aid=1633060
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-common@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| MonetDB: leftjoin on strings returns wrong result | https://api.github.com/repos/MonetDB/MonetDB/issues/1191/comments | 0 | 2020-11-30T09:25:28Z | 2024-06-28T07:36:52Z | https://github.com/MonetDB/MonetDB/issues/1191 | 753,307,974 | 1,191 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-10 12:13:00 +0100
From: Jan Rittinger <<jan.rittinger>>
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2008-06-06 08:34:00 +0200
## Comment 6660
Date: 2007-01-10 12:13:00 +0100
From: Jan Rittinger <<jan.rittinger>>
No free() is called after a fatal PFoops call in pf which (after some errors) crashes the Mserver. (This bug affects only the embedded variant of pf as the standalone one quits anyway.)
The following small script reproduces the bug:
echo 'module(pathfinder);' > bug.mil;for a in `seq 0 10000`; do echo "xquery(\"var$a\");" >> bug.mil; done; Mserver bug.mil
## Comment 6661
Date: 2007-05-25 13:44:21 +0200
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
is this bug still open?
at least with MPS it seems not to crash.. and the PFmem_destroy() is called so which free() is being missed?
## Comment 6662
Date: 2007-06-04 10:38:50 +0200
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
Hi Jan,
any news on this one? Is this only happening when the algebra version is the 'embedded' compiler? I usually don't compile that way (yet).
Peter
## Comment 6663
Date: 2007-06-04 11:14:28 +0200
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: YES
If I run the above query -- assuming the xquery command uses mps -- the memory usage of Mserver still continues growing. (Note that tested version is 2-3 weeks old.)
After the 10000 buggy queries Mserver is still running but uses 683MB memory (+ 902MB virtual) --- If this behaviour is ok for you I don't object if you close the bug. With my (old) version I however can still see (a small) memory leak.
## Comment 6664
Date: 2008-06-06 20:34:00 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Running 10000 incorrect queries via `mclient -lx` (as opposed to using the non-recommended "xquery();" MIL command) does not show any groth in the Mserver at all, neither with MPS nor with ALG.
Hence, I consider this one "out of date" and close it.
## Comment 6665
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1632239 at http://sourceforge.net/support/tracker.php?aid=1632239
| MonetDB/XQuery leaks memory on erroneous XQuery query | https://api.github.com/repos/MonetDB/MonetDB/issues/1190/comments | 0 | 2020-11-30T09:25:26Z | 2024-06-27T11:22:08Z | https://github.com/MonetDB/MonetDB/issues/1190 | 753,307,947 | 1,190 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-09 10:36:04 +0100
From: Jan Rittinger <<jan.rittinger>>
To: GDK devs <<bugs-common>>
Version: 11.21.19 (Jul2015-SP4)
CC: duc
Last updated: 2018-04-24 20:47:35 +0200
## Comment 6656
Date: 2007-01-09 10:36:04 +0100
From: Jan Rittinger <<jan.rittinger>>
The support for NaN values seems to be broken (see below examples) as it results in random results. (To support NaN in XQuery NaN in C (MonetDB) seems the only feasible solution so it would be great if this could be fixed.)
MonetDB>var a := bat (void,dbl).seqbase(0@0).append(dbl(2)).append(dbl(nil)).append(dbl("nan"));
MonetDB>print(a.select (dbl(nil), dbl(nil)));
-----------------
h t name
oid dbl type
-----------------
[ 0@0, 2 ]
[ 2@0, nan ]
MonetDB>a := bat (void,dbl).seqbase(0@0).append(dbl("nan"));
MonetDB>print(a.select (dbl(nil), dbl(nil)));
-----------------
h t name
void dbl type
-----------------
MonetDB>a := bat (void,dbl).seqbase(0@0).append(dbl("nan")).append(dbl("nan"));
MonetDB>print(a.select (dbl(nil), dbl(nil)));
-----------------
h t name
void dbl type
-----------------
MonetDB>a := bat (void,dbl).seqbase(0@0).append(dbl(2)).append(dbl("nan"));
MonetDB>print(a.select (dbl(nil), dbl(nil)));
-----------------
h t name
void dbl type
-----------------
MonetDB>a := bat (void,dbl).seqbase(0@0).append(dbl(2)).append(dbl("nan")).append(dbl(nil));
MonetDB>print(a.select (dbl(nil), dbl(nil)));
-----------------
h t name
void dbl type
-----------------
## Comment 6657
Date: 2007-01-23 11:53:46 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
It seems as if NaN's do break both property management and property checking; or at least, property management/checking seem to ignore NaN's and then the select on supposely "sorted" tails (falling back to a slice) fails:
========
MonetDB>debugmask(10);
MonetDB>a := bat (void,dbl).seqbase(0@0).append(dbl(2)).append(dbl("nan")).append(dbl(3)); print(a); a.info().find("tsorted").print();
-----------------
h t name
void dbl type
-----------------
[ 0@0, 2 ]
[ 1@0, NaN ]
[ 2@0, 3 ]
[ "65" ]
MonetDB>a := bat (void,dbl).seqbase(0@0).append(dbl(2)).append(dbl(nil)).append(dbl("nan")); print(a); a.info().find("tsorted").print();
-----------------
h t name
void dbl type
-----------------
[ 0@0, 2 ]
[ 1@0, nil ]
[ 2@0, NaN ]
[ "0" ]
MonetDB>a := bat (void,dbl).seqbase(0@0).append(dbl(2)).append(dbl("nan")).append(dbl(nil)); print(a); a.info().find("tsorted").print();
-----------------
h t name
void dbl type
-----------------
[ 0@0, 2 ]
[ 1@0, NaN ]
[ 2@0, nil ]
[ "65" ]
MonetDB>debugmask(0);
MonetDB>a := bat (void,dbl).seqbase(0@0).append(dbl(2)).append(dbl("nan")).append(dbl(3)); print(a); a.info().find("tsorted").print();
-----------------
h t name
void dbl type
-----------------
[ 0@0, 2 ]
[ 1@0, NaN ]
[ 2@0, 3 ]
[ "65" ]
MonetDB>a := bat (void,dbl).seqbase(0@0).append(dbl(2)).append(dbl(nil)).append(dbl("nan")); print(a); a.info().find("tsorted").print();
-----------------
h t name
void dbl type
-----------------
[ 0@0, 2 ]
[ 1@0, nil ]
[ 2@0, NaN ]
[ "0" ]
MonetDB>a := bat (void,dbl).seqbase(0@0).append(dbl(2)).append(dbl("nan")).append(dbl(nil)); print(a); a.info().find("tsorted").print();
-----------------
h t name
void dbl type
-----------------
[ 0@0, 2 ]
[ 1@0, NaN ]
[ 2@0, nil ]
[ "65" ]
MonetDB>
=======
Open questions:
What are the semantics of NaN's wrt. properties like sortedness?
Should they destroy the properties, ust like NILs do?
Should select(nil,nil) (supposely returning all non-nil values) also return NaN's?
## Comment 6658
Date: 2007-02-19 14:03:59 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
So, what is it that we want to do here?
Do we want to support NaN's fully or do we want to map them to nil?
## Comment 6659
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1631298 at http://sourceforge.net/support/tracker.php?aid=1631298
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-common@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
## Comment 15662
Date: 2011-03-28 17:34:55 +0200
From: @sjoerdmullender
The Mar2011 version has been released.
## Comment 16020
Date: 2011-07-29 10:58:43 +0200
From: @sjoerdmullender
Apr2011-SP2 has been released.
## Comment 16259
Date: 2011-09-16 15:08:30 +0200
From: @sjoerdmullender
The Aug2011 version has been released.
## Comment 18140
Date: 2012-11-27 16:05:50 +0100
From: duc <<duc>>
Cannot create the test because MAL does not support NaN.
## Comment 26405
Date: 2018-04-24 20:47:35 +0200
From: @sjoerdmullender
Since the Mar2018 release, MonetDB uses NaN for its representation of NIL (or NULL in SQL).
| MonetDBs NaN support broken | https://api.github.com/repos/MonetDB/MonetDB/issues/1189/comments | 0 | 2020-11-30T09:25:23Z | 2024-06-27T11:22:07Z | https://github.com/MonetDB/MonetDB/issues/1189 | 753,307,908 | 1,189 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-03 11:09:24 +0100
From: @Teggy
To: @sjoerdmullender
Version: -- development
Last updated: 2007-01-10 03:39:30 +0100
## Comment 6650
Date: 2007-01-03 11:09:24 +0100
From: @Teggy
Dear all,
on case-preserving but case-insensitive file systems (like Mac OS X's VHFS+, Windows' HFS), the Pathfinder compilation process currently suffers from inclusion of (at least one) wrong C header files:
File src/tools/embeddedclient.mx, which includes "Mapi.h", is compiled with switches
-I.../include/MonetDB4/mapi
-I.../include/MonetDB/mapilib
in that order. The former directory contains the
header file "mapi.h" (lower case 'm') which is incorrectly included by src/tools/embeddedclient.mx on a case-insensitive file system -- only the latter directory contains the desired "Mapi.h" (upper case 'M').
Cheers,
--Teggy
## Comment 6651
Date: 2007-01-10 11:02:36 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
I have reordered the paths so that I now get a clean compile on Windows.
Can you check and close the bug if this fixed the issue?
## Comment 6652
Date: 2007-01-10 12:21:44 +0100
From: @Teggy
Logged In: YES
user_id=642632
Originator: YES
Hmm, after CVS update, bootstrap, configure, the path .../include/MonetDB/mapilib now seems to be missing from the compilation:
gcc -DHAVE_CONFIG_H -I. -I/Users/grust/sci/XQuery/pf/src/tools -I../.. -I/Users/grust/sci/XQuery/pf/src/tools -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB/common -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB/gdk -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB4 -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB4/common -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB4/gdk -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB4/monet -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB4/plain -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB4/contrib -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB4/mapi -I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB4/C -DLIBEMBEDDEDXQ -g -O2 -Wall -W -std=c99 -Werror-implicit-function-declaration -Werror -Wno-unused-function -Wno-unused-label -Wno-uninitialized -D_REENTRANT -c embeddedclient.c -fno-common -DPIC -o .libs/libembeddedxq_la-embeddedclient.o
/Users/grust/sci/XQuery/pf/src/tools/embeddedclient.mx:179: error: parse error before "embedded_xquery"
Cheers,
--Teggy
## Comment 6653
Date: 2007-01-10 12:40:21 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
There must be something with your settings. You have both
-I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB/common
-I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB/gdk
and
-I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB4/common
-I/Users/grust/sci/XQuery/Pathfinder/include/MonetDB4/gdk
The latter two make no sense. Instead you should have the mapilib include there.
Take a look at your CLIENTS_* variables.
I have the following paths:
gcc -m64 -DHAVE_CONFIG_H -I. -I/ufs/sjoerd/src/MonetDB/devel/pathfinder/src/tools -I../.. -I/ufs/sjoerd/src/MonetDB/devel/pathfinder/src/tools -I/ufs/sjoerd/Monet/include/MonetDB -I/ufs/sjoerd/Monet/include/MonetDB/common -I/ufs/sjoerd/Monet/include/MonetDB/gdk -I/ufs/sjoerd/Monet/include/MonetDB -I/ufs/sjoerd/Monet/include/MonetDB/mapilib -I/ufs/sjoerd/Monet/include/MonetDB4 -I/ufs/sjoerd/Monet/include/MonetDB4/monet -I/ufs/sjoerd/Monet/include/MonetDB4/plain -I/ufs/sjoerd/Monet/include/MonetDB4/contrib -I/ufs/sjoerd/Monet/include/MonetDB4/mapi -I/ufs/sjoerd/Monet/include/MonetDB4/calibrator -DODBCDEBUG -DDEBUG_YYSTREE -DLIBEMBEDDEDXQ -std=c99 -Wall -W -std=c99 -g -Werror-implicit-function-declaration -Werror -Wno-unused-function -Wno-unused-label -D_REENTRANT -c embeddedclient.c -fPIC
-DPIC -o .libs/libembeddedxq_la-embeddedclient.o
## Comment 6654
Date: 2007-01-10 15:39:30 +0100
From: @Teggy
Logged In: YES
user_id=642632
Originator: YES
A full wipe and rebuild did the trick. Thanks, Sjoerd.
Closed.
## Comment 6655
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1626775 at http://sourceforge.net/support/tracker.php?aid=1626775
| Header file confusion on case-insensitive file systems | https://api.github.com/repos/MonetDB/MonetDB/issues/1188/comments | 0 | 2020-11-30T09:25:20Z | 2024-06-27T11:22:06Z | https://github.com/MonetDB/MonetDB/issues/1188 | 753,307,874 | 1,188 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-02 05:50:40 +0100
From: @arjenpdevries
To: GDK devs <<bugs-common>>
Version: -- development
Last updated: 2007-01-02 06:20:02 +0100
## Comment 6643
Date: 2007-01-02 17:50:40 +0100
From: @arjenpdevries
Nodes inside comments cause the parser to fail
if they are not valid XML
Try e.g. query
(: <x> :)
MapiClient -lx comments.xq
MAPI = monetdb@localhost:50000
QUERY = (: <x> :)
ERROR = !parse error: XQuery parsing failed
! halted in /ufs/arjen/src/MonetDB/pathfinder/compiler/parser/parser.y (PFparse), line 2865
## Comment 6644
Date: 2007-01-02 18:07:37 +0100
From: Jens Teubner <<teubner>>
Logged In: YES
user_id=731390
Originator: NO
(: <x> :)
is not a valid query. It is only a comment, without the (required!) query body.
I don't have a running server-side XQuery setup here. With the standalone Pathfinder compiler, a query like
(: <x> :)
42
runs perfectly fine. If I leave out the `42', the query becomes illegal and Pathfinder complains. Let me know if server-side XQuery behaves differently.
## Comment 6645
Date: 2007-01-02 18:09:30 +0100
From: @arjenpdevries
Logged In: YES
user_id=572978
Originator: YES
My example was incorrect!
Not sure if this is a bug after all...
Sorry!
I did have a problem in a more complex query, but cannot reproduce it for now,
so maybe just close the bug.
## Comment 6646
Date: 2007-01-02 18:13:20 +0100
From: @arjenpdevries
Logged In: YES
user_id=572978
Originator: YES
Ah, got one after all
<x>
(: <y> :)
1
</x>
Fails:
MapiClient -lx comments1.xq
MAPI = monetdb@localhost:50000
QUERY = <x>
(: <y> :)
1
</x>
ERROR = !XML start/end tags do not match: at (2,5-4,5): <y> and </x>
!parse error: XQuery parsing failed
! halted in /ufs/arjen/src/MonetDB/pathfinder/compiler/parser/parser.y (PFparse), line 2865
Still not sure if it is valid XQuery though, is the comment allowed there?
Otherwise, with a closing </y> the result is:
<?xml version="1.0" encoding="utf-8"?>
<XQueryResult><x>
(: <y/> :)
1
</x></XQueryResult>
so the comment is included in the output - maybe this is expected behaviour?
A.
## Comment 6647
Date: 2007-01-02 18:20:02 +0100
From: Jens Teubner <<teubner>>
Logged In: YES
user_id=731390
Originator: NO
I guess Arjen unintentionally re-opened this bug.
## Comment 6648
Date: 2007-01-02 18:26:42 +0100
From: Jens Teubner <<teubner>>
Logged In: YES
user_id=731390
Originator: NO
What you observe is intended behavior. There are no comments within XML content. `(: ... :)' will be copied literally into the resulting XML document. `<!-- ... -->' will result in an XML comment node (I think this should work for your case, though I haven't tried).
If you want comment out things such that they will not appear in the final output, you need to switch into "XQuery mode" with `{' and use XQuery-style comment syntax there, e.g.,
<x>
{ (: a nice comment :) }
foo
</x>
Jens
## Comment 6649
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1626311 at http://sourceforge.net/support/tracker.php?aid=1626311
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-common@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| Nodes inside comments | https://api.github.com/repos/MonetDB/MonetDB/issues/1187/comments | 0 | 2020-11-30T09:25:17Z | 2024-06-28T07:36:52Z | https://github.com/MonetDB/MonetDB/issues/1187 | 753,307,846 | 1,187 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-02 05:36:37 +0100
From: @arjenpdevries
To: Jan Rittinger <<jan.rittinger>>
Version: -- development
Last updated: 2007-01-02 10:43:14 +0100
## Comment 6637
Date: 2007-01-02 17:36:37 +0100
From: @arjenpdevries
(:
Mergeunion bug
:)
for $r in (1, 2) return
(
<node>
<node2 />
<node2 test="-1">
<node3 id="1" />
</node2>
</node>
)
This gives
ERROR = !ERROR: merged_union: tail of second BAT must be sorted.
!ERROR: CMDmerged_union: operation failed.
Remove any of the nodes or attributes, and output is created.
This structure occurs in HTML output formatting
for a demo app I am working on, so it would be
great if an expert could give a hand and locate the
underlying problem!
Arjen
## Comment 6638
Date: 2007-01-02 18:08:54 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
As a quick fix/work-around, try to start Mserver with -d10 and let us know, whether that helps.
## Comment 6639
Date: 2007-01-02 18:18:18 +0100
From: @arjenpdevries
Logged In: YES
user_id=572978
Originator: YES
No, -d10 does not help.
My quick fix is for now to move the attribute to a CSS style-sheet (but would still like to see it fixed :-))
Arjen
## Comment 6640
Date: 2007-01-02 18:24:37 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
You guys are really in a hurry, or? :)
The algebra version (pf -A) of course works...
A quick look showed that combination/sequence construction of attributes in mps is done sometimes with inserts and sometimes with merged_union which messes up the order merged_union requires. (I think this is the first time where merged_union wasn't to picky :))
I will check in a fix in some minutes. For the ones that need it immediately here is the diff (for the CVS head):
RCS file: /cvsroot/monetdb/pathfinder/compiler/mil/milprint_summer.c,v
retrieving revision 1.314
diff -r1.314 milprint_summer.c
2480,2491c2480,2490
< "var seqb := oid(_attr_iter.count() + int(_attr_iter.seqbase()));\n"
< "_r_attr_iter := _r_attr_iter.tmark(seqb);\n"
< "_r_attr_qn := _r_attr_qn .tmark(seqb);\n"
< "_r_attr_prop := _r_attr_prop.tmark(seqb);\n"
< "_r_attr_cont := _r_attr_cont.tmark(seqb);\n"
< "attr_own := attr_own.tmark(seqb);\n"
< "seqb := nil;\n"
< "_attr_iter := _attr_iter.access(BAT_WRITE).insert(_r_attr_iter);\n"
< "_attr_qn := _attr_qn .access(BAT_WRITE).insert(_r_attr_qn);\n"
< "_attr_prop := _attr_prop.access(BAT_WRITE).insert(_r_attr_prop);\n"
< "_attr_cont := _attr_cont.access(BAT_WRITE).insert(_r_attr_cont);\n"
< "_attr_own := _attr_own .access(BAT_WRITE).insert(attr_own);\n"
---
> "var merged_result := merged_union ("
> "_attr_iter, _r_attr_iter, "
> "_attr_qn , _r_attr_qn, "
> "_attr_prop, _r_attr_prop, "
> "_attr_cont, _r_attr_cont, "
> "_attr_own , attr_own);\n"
> "_attr_iter := merged_result.fetch(0);\n" /* CONST? */
> "_attr_qn := merged_result.fetch(1);\n" /* CONST? */
> "_attr_prop := merged_result.fetch(2);\n" /* CONST? */
> "_attr_cont := merged_result.fetch(3);\n" /* CONST? */
> "_attr_own := merged_result.fetch(4);\n" /* CONST? */
## Comment 6641
Date: 2007-01-02 22:43:14 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
tests added in pathfinder/tests/BugTracker/Tests/merged_union-bug.SF-1626306.*
Tests and today fixes have alos been propagated to the development head.
However, for "mission-critical" usage, I'd highly recommend to use the stable branch instead of beta-testing the development head (in particular with the wounds of the recent major CVS surgery still being open ...).
## Comment 6642
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1626306 at http://sourceforge.net/support/tracker.php?aid=1626306
| merged_union bug | https://api.github.com/repos/MonetDB/MonetDB/issues/1186/comments | 0 | 2020-11-30T09:25:15Z | 2024-06-27T11:22:04Z | https://github.com/MonetDB/MonetDB/issues/1186 | 753,307,815 | 1,186 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-02 03:49:05 +0100
From: Jan Rittinger <<jan.rittinger>>
To: @sjoerdmullender
Version: -- development
Last updated: 2007-01-03 02:49:45 +0100
## Comment 6634
Date: 2007-01-02 15:49:05 +0100
From: Jan Rittinger <<jan.rittinger>>
In my (head) version of pathfinder the following query
let $a := <a><b/></a> return do insert <c/> into $a
results in a MIL error (!ERROR: CMDfind: value not found.).
I would expect either a real error message like 'Invalid updates on transient nodes' or the update has to be silently ignored.
## Comment 6635
Date: 2007-01-03 14:49:45 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Fixed.
Also added test tests/BugTracker/Tests/insert_into_transient.SF-1626208.xq.
## Comment 6636
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1626208 at http://sourceforge.net/support/tracker.php?aid=1626208
| XQ: updates on transient nodes | https://api.github.com/repos/MonetDB/MonetDB/issues/1185/comments | 0 | 2020-11-30T09:25:12Z | 2024-06-27T11:22:03Z | https://github.com/MonetDB/MonetDB/issues/1185 | 753,307,787 | 1,185 |
[
"MonetDB",
"MonetDB"
] | Date: 2007-01-01 10:20:36 +0100
From: @arjenpdevries
To: Jens Teubner <<teubner>>
Version: -- development
Last updated: 2007-01-02 05:19:48 +0100
## Comment 6630
Date: 2007-01-01 10:20:36 +0100
From: @arjenpdevries
Good morning, and a Happy New Year to all!
My pathfinder ($Revision: 1.74 $, $Date: 2006/11/13 17:16:28 $)
crashes on
declare variable $v as xs:string := "variable v";
but does process
declare variable $v := "variable v";
correctly.
The crash also kills the server,
MonetDB>Mserver: /ufs/arjen/src/MonetDB/pathfinder/compiler/core/core.c:123: PFcore_wire2: Assertion `n1 && n2' failed.
Aborted
Example script:
declare variable $query as xs:string := "QUERY";
Cheers,
Arjen
## Comment 6631
Date: 2007-01-02 10:28:58 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
added tests in
pathfinder/tests/BugTracker/Tests/declare_variable.SF-1625597.*
and
pathfinder/tests/BugTracker/Tests/declare_variable_as_TYPE.SF-1625597.*
(in the XQuery_0-14 release branch; not propagated to the development trunk, yet)
The problems occurs with both the latest CVS head and the XQuery_0-14 release branch.
## Comment 6632
Date: 2007-01-02 17:19:47 +0100
From: Jens Teubner <<teubner>>
Logged In: YES
user_id=731390
Originator: NO
Fixed in CVS (stable branch)
## Comment 6633
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1625597 at http://sourceforge.net/support/tracker.php?aid=1625597
| declare variable as TYPE := value; | https://api.github.com/repos/MonetDB/MonetDB/issues/1184/comments | 0 | 2020-11-30T09:25:10Z | 2024-06-27T11:22:02Z | https://github.com/MonetDB/MonetDB/issues/1184 | 753,307,754 | 1,184 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-21 01:56:35 +0100
From: @drstmane
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-12-22 06:53:21 +0100
## Comment 6626
Date: 2006-12-21 01:56:35 +0100
From: @drstmane
After running test
pathfinder/tests/BugTracker/Tests/pf_del-doc.SF-1617684*
(http://sourceforge.net/tracker/index.php?func=detail&aid=1617684&group_id=56967&atid=482468),
restarting Mserver (with -d10) fails with
!FATAL: BATpropcheck: BAT tmp_34(-28) has inconsistent descriptor 32768 (100000)
see test in
pathfinder/tests/BugTracker/Tests/pf_del-doc.SF-1617684_check.*
## Comment 6627
Date: 2006-12-21 13:09:04 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
See
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/pf_del-doc.SF-1617684.out.00.html
and
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/pf_del-doc.SF-1617684_check.err.00.html
for details.
## Comment 6628
Date: 2006-12-22 18:53:21 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
fixed bug in multiplex
## Comment 6629
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1619816 at http://sourceforge.net/support/tracker.php?aid=1619816
| XQ: Mserver fails to restart after pf:del-doc() | https://api.github.com/repos/MonetDB/MonetDB/issues/1183/comments | 0 | 2020-11-30T09:25:07Z | 2024-06-27T11:22:01Z | https://github.com/MonetDB/MonetDB/issues/1183 | 753,307,715 | 1,183 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-20 01:02:13 +0100
From: Jan Rittinger <<jan.rittinger>>
To: GDK devs <<bugs-common>>
Version: -- development
Last updated: 2007-05-25 02:33:03 +0200
## Comment 6623
Date: 2006-12-20 13:02:13 +0100
From: Jan Rittinger <<jan.rittinger>>
In the attached MIL script a query based on the 100 MB auction.xml file is evaluated. The script contains a while loop where the last (running) iteration starts with 7 million tuples. In this iteration (after about 10 minutes) Mserver just quits without an error message. The last (and the to be) processed line was/is an assignment to unused (which is again bound to nil).
I'd guess that freeing some memory mighht cause this crash. (A run with -d10 also did not change anything.)
## Comment 6624
Date: 2007-05-25 02:33:03 +0200
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
this appears to be just a case of a BAT becoming too big and Mserver running out of resources
## Comment 6625
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1619381 at http://sourceforge.net/support/tracker.php?aid=1619381
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-common@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| MonetDB crashes during nil assignment | https://api.github.com/repos/MonetDB/MonetDB/issues/1182/comments | 0 | 2020-11-30T09:25:05Z | 2024-06-28T07:36:52Z | https://github.com/MonetDB/MonetDB/issues/1182 | 753,307,674 | 1,182 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-19 03:38:29 +0100
From: Roel van Os <<roel_van_os>>
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-12-23 06:33:59 +0100
## Comment 6615
Date: 2006-12-19 15:38:29 +0100
From: Roel van Os <<roel_van_os>>
There seems to be a document handling bug in the development version of pathfinder.
### Steps to reproduce:
1. Extract the attached testfiles.tar.gz into a temporary dir
2. Execute the query attached as test_topics.xq *twice* (be sure to modify it to point to the correct directory). My method was:
pf test_topics.xq > queryplan.mil
Mserver queryplan.mil
Mserver queryplan.mil
This query loads the documents from the attached topics.tar.gz using fn:doc and performs a path step on them.
The first time the query is executed, it works as expected: the documents are indexed and a set of strings from the topic XML files is output.
The second time the query is executed, the following error is given:
!ERROR: _del_doc(/home/roel/Temp/topics/138.xml): document not found in database (2 such errors)!
!ERROR: interpret_params: leftfetchjoin(param 2): evaluation error.
This was reproducible for me on both 32 and 64-bit architectures (though with MonetDB configured with --enable-oid32 on the 64-bit).
## Comment 6616
Date: 2006-12-19 15:39:01 +0100
From: Roel van Os <<roel_van_os>>
Logged In: YES
user_id=1523350
Originator: YES
File Added: test_fndoc.xq
## Comment 6617
Date: 2006-12-20 14:47:32 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Attached testfiles.tar.gz seems to be broken:
========
$ l testfiles.tar.gz
--------
-rw-rw-r-- 1 manegold manegold 336 Dec 20 14:42 testfiles.tar.gz
========
$ file testfiles.tar.gz
--------
testfiles.tar.gz: gzip compressed data, from Unix, last modified: Tue Dec 19 15:38:16 2006
========
$ gzip -vd testfiles.tar.gz
--------
testfiles.tar.gz:
gzip: testfiles.tar.gz: decompression OK, trailing garbage ignored
98.4% -- replaced with testfiles.tar
========
$ l testfiles.tar
--------
-rw-rw-r-- 1 manegold manegold 20480 Dec 20 14:42 testfiles.tar
========
$ file testfiles.tar
--------
testfiles.tar: POSIX tar archive
========
$ tar -xvpf testfiles.tar
--------
testfiles/
testfiles/1.xml
testfiles/2.xml
testfiles/10.xml
testfiles/3.xml
testfiles/4.xml
testfiles/5.xml
testfiles/6.xml
testfiles/7.xml
testfiles/8.xml
testfiles/9.xml
========
$ l testfiles
--------
total 52
drwxr-xr-x 2 manegold manegold 4096 Dec 19 15:36 ./
drwxr-xr-x 5 manegold manegold 4096 Dec 20 14:42 ../
-rw-r--r-- 1 manegold manegold 41 Dec 19 15:36 10.xml
-rw-r--r-- 1 manegold manegold 41 Dec 19 15:35 1.xml
-rw-r--r-- 1 manegold manegold 41 Dec 19 15:36 2.xml
-rw-r--r-- 1 manegold manegold 41 Dec 19 15:36 3.xml
-rw-r--r-- 1 manegold manegold 41 Dec 19 15:36 4.xml
-rw-r--r-- 1 manegold manegold 41 Dec 19 15:36 5.xml
-rw-r--r-- 1 manegold manegold 41 Dec 19 15:36 6.xml
-rw-r--r-- 1 manegold manegold 41 Dec 19 15:36 7.xml
-rw-r--r-- 1 manegold manegold 41 Dec 19 15:36 8.xml
-rw-r--r-- 1 manegold manegold 41 Dec 19 15:36 9.xml
========
## Comment 6618
Date: 2006-12-20 15:02:17 +0100
From: Roel van Os <<roel_van_os>>
Logged In: YES
user_id=1523350
Originator: YES
The tar file is fine, it is just too small for GZIP compression (smaller than 10 kB or whatever), hence the trailing garbage. The tar should expand to ten identical XML files, each containing exactly this:
<?xml version="1.0"?>
<test>Hoi!</test>
I could have attached a bash script that generates these files, but this seemed more convenient.
## Comment 6619
Date: 2006-12-20 15:10:04 +0100
From: Roel van Os <<roel_van_os>>
Logged In: YES
user_id=1523350
Originator: YES
BTW, due to a misunderstanding of SF.net's web interface on my part, most of the filenames in the bug report are wrong (I submitted my description too soon, when uploading the example files).
The test XQuery is attached as test_fndoc.xq (test_topics.xq in the original bug report).
The _del_doc error should point to one of the test files (1.xml - 10.xml) from testfiles.tar.gz.
Hope this helps.
## Comment 6620
Date: 2006-12-20 18:33:39 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Added tests in pathfinder/tests/BugTracker/Tests/document_handling.SF-1618802.*
(on the XQuery_0-14 release branch; automa[tgn]ic propagation to development trunk pending)
A simple query loads 100 tiny document via fn:doc(<file_uri>).
Running the query a second time (both without and without stopping the Mserver in between) fn:doc() is supposed to re-use the cached document as opposed to re-shredding them.
This seems to work fine, in case the number of documents is small.
However, when the number of documents exceed a certain limit,
in my case (64-bit FC4 Linux, 64-bit OIDs, xquery_cacheMB=100) with more than 95 of these tiny documents (total serialized size is 404 KB, total size in dbfarm is ~500 KB),
MonetDB/XQuery seems to assume that the default 100 MB(!) cache is full, and hence apparently tries to clear it before proceeding
(although actually access only cached documents, i.e., no new documents have to be added to the cache).
Clearing the cache then fails (as reported in BUG #1618802) with
!ERROR: _del_doc(/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Stable/.GNU.64.64.d.--disable-debug_--enable-assert_--enable-optimize.PREFIX./PATHFINDER/mTests/tests/BugTracker/document_handling.SF-1618802/95.xml): document not found in database (1 such errors)!
In fact, I consider this three bugs:
1)
As originally reported, clearing the cache fails in _del_doc().
2)
With a total DB size of 500 KB and a cache size of 100 MB, clearing of the cache should never be triggered; i.e., some size calculation/estimation seems to be wrong.
3)
Since we only access cached documents who's files haven't changed, now new document need to be added, hence, a cache clearing should not be triggered, even if the cache was full.
Assigning to Peter, but I also plan to have a closer look once I'll have time --- don't expect that to be before next year.
## Comment 6621
Date: 2006-12-23 18:33:59 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
- made size bookkeeping the the xmlcache more consistent (i.e. monitor readl disk usage)
NOTE: due to bat headers and minimum heap sizes (512 bytes assumed), an *empty* XML collection is 72KB
that is the minimum overhead
- fixed a bug in the delete sequence of out-of-date cached documents
- keep per-document (not per-collection) document timestamps in fn:doc() generated collections
- removed verbose shredding output in fn:doc() invocations
## Comment 6622
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1618802 at http://sourceforge.net/support/tracker.php?aid=1618802
| XQ: document handling bug? | https://api.github.com/repos/MonetDB/MonetDB/issues/1181/comments | 0 | 2020-11-30T09:25:02Z | 2024-06-27T11:21:59Z | https://github.com/MonetDB/MonetDB/issues/1181 | 753,307,630 | 1,181 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-18 04:24:52 +0100
From: @drstmane
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-12-22 06:54:35 +0100
## Comment 6609
Date: 2006-12-18 04:24:52 +0100
From: @drstmane
========
$ echo 'xmldb_print();' | MapiClient
--------
---------------------------------
collection alias URI name
str str str type
---------------------------------
========
$ echo 'shred_doc("/tmp/x.xml","x.xml");' | MapiClient
--------
Elapsed time = 886us [295us/node]
Shredded 1 XML document (x.xml), total time after commit=0.070s
========
$ echo 'xmldb_print();' | MapiClient
--------
-------------------------------------------------
collection alias URI name
str str str type
-------------------------------------------------
[ "x.xml", "x.xml", "/tmp/x.xml" ]
========
$ echo 'pf:del-doc("x.xml")' | MapiClient -lx
--------
MAPI = monetdb@localhost:50000
QUERY = pf:del-doc("x.xml")
ERROR = !ERROR: merged_union: BATs 3 (ttype=7) & 4 (ttype=12) must have the same tail types.
!ERROR: CMDmerged_union: operation failed.
========
$ echo 'pf:del-doc("/tmp/x.xml")' | MapiClient -lx
--------
MAPI = monetdb@localhost:50000
QUERY = pf:del-doc("/tmp/x.xml")
ERROR = !ERROR: merged_union: BATs 3 (ttype=7) & 4 (ttype=12) must have the same tail types.
!ERROR: CMDmerged_union: operation failed.
========
## Comment 6610
Date: 2006-12-20 20:35:08 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
tests added as pathfinder/tests/BugTracker/Tests/pf_del-doc.SF-1617684*
## Comment 6611
Date: 2006-12-20 23:51:31 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
I'll have a look at this one.
## Comment 6612
Date: 2006-12-21 02:01:50 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
another (early) christmas present:
(almost) fixed in CVS:
runtime/pathfinder.mx:
fixed implementation on play_doc_tape() (wasn'\''t working at all)
compiler/mil/milprint_summer.c:
fixed code generation for pf:add-doc & pf:del-doc
tests/BugTracker/Tests/pf_add-doc.SF-1617681_s00.xq.in:
pf:add-doc expects at least two arguments: URI and name of the document
tests/BugTracker/Tests/pf_del-doc.SF-1617684_s00.xq.in:
pf:del-doc works only with document names, not with URIs
The tests for pf:add-doc & pf:del-doc now work fine;
at the expense of an ugly HACK in runtime/pathfinder.mx:
With _shredlock_set() being a void proc,
the multiplexed _shredlock_set() call in _shred_doc_base() triggers either a
!FATAL: BATpropcheck: BAT tmp_335(-221) has inconsistent descriptor 32768 (100000)
(when using pf:add-doc()) or a
!WARNING: [_shredlock_set]: empty result, hence return type unknown.
(when using pf:del-doc()).
This seems to be a bug in the multiplexer that I could not locate, yet.
Make _shredlock_set() "fake-return" a bit avoids the problem.
An alternative workarount to the fake-return would be to replace the
multiplexed _shredlock_set() call in _shred_doc_base() by a batloop-ed version.
TODO:
Someone needs to analyse and fix the multiplexer problem!
Another problem is that after the pf:del-doc test, restarting the Mserver (with -d10) fails with
!FATAL: BATpropcheck: BAT tmp_34(-28) has inconsistent descriptor 32768 (100000).
See new bug report at
http://sourceforge.net/tracker/index.php?func=detail&aid=1619816&group_id=56967&atid=482468
## Comment 6613
Date: 2006-12-22 18:54:35 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
fixed bug in multiplex
## Comment 6614
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1617684 at http://sourceforge.net/support/tracker.php?aid=1617684
| XQ: pf:del-doc does not work | https://api.github.com/repos/MonetDB/MonetDB/issues/1180/comments | 0 | 2020-11-30T09:24:59Z | 2024-06-27T11:21:58Z | https://github.com/MonetDB/MonetDB/issues/1180 | 753,307,591 | 1,180 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-18 04:18:56 +0100
From: @drstmane
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-12-22 06:54:54 +0100
## Comment 6604
Date: 2006-12-18 04:18:56 +0100
From: @drstmane
========
$ echo '<x/>' > /tmp/x.xml
========
$ echo 'pf:add-doc("/tmp/x.xml")' | MapiClient -lx
--------
MAPI = monetdb@localhost:50000
QUERY = pf:add-doc("/tmp/x.xml")
ERROR = !error in function application: at (1,1-1,24): wrong number of arguments for function `pf:add-doc' (expected 4, got 1)
! halted in /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/semantics/functions.c (check_fun_usage), line 367
========
$ echo 'pf:add-doc("/tmp/x.xml","/tmp/x.xml")' | --------
MapiClient -lx
MAPI = monetdb@localhost:50000
QUERY = pf:add-doc("/tmp/x.xml","/tmp/x.xml")
ERROR = !ERROR: interpret: no matching MIL operator to '[abs](oid)'.
!MAYBE YOU MEAN:
! abs(dbl) : dbl
! abs(lng) : lng
! abs(flt) : flt
! abs(wrd) : wrd
! abs(int) : int
! abs(sht) : sht
! abs(bte) : bte
! abs(chr) : chr
!ERROR: interpret_params: merged_union(param 4): evaluation error.
========
$ echo 'pf:add-doc("/tmp/x.xml","/tmp/x.xml","/tmp/x.xml")' | MapiClient -lx
--------
MAPI = monetdb@localhost:50000
QUERY = pf:add-doc("/tmp/x.xml","/tmp/x.xml","/tmp/x.xml")
ERROR = !ERROR: interpret: no matching MIL operator to '[abs](oid)'.
!MAYBE YOU MEAN:
! abs(dbl) : dbl
! abs(lng) : lng
! abs(flt) : flt
! abs(wrd) : wrd
! abs(int) : int
! abs(sht) : sht
! abs(bte) : bte
! abs(chr) : chr
!ERROR: interpret_params: merged_union(param 4): evaluation error.
========
$ echo 'pf:add-doc("/tmp/x.xml","/tmp/x.xml","/tmp/x.xml",0)' | MapiClient -lx
--------
MAPI = monetdb@localhost:50000
QUERY = pf:add-doc("/tmp/x.xml","/tmp/x.xml","/tmp/x.xml",0)
ERROR = !ERROR: interpret: no matching MIL operator to '[abs](oid)'.
!MAYBE YOU MEAN:
! abs(dbl) : dbl
! abs(lng) : lng
! abs(flt) : flt
! abs(wrd) : wrd
! abs(int) : int
! abs(sht) : sht
! abs(bte) : bte
! abs(chr) : chr
!ERROR: interpret_params: merged_union(param 4): evaluation error.
========
## Comment 6605
Date: 2006-12-20 20:35:01 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
tests added as pathfinder/tests/BugTracker/Tests/pf_add-doc.SF-1617681*
## Comment 6606
Date: 2006-12-20 23:51:39 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
I'll have a look at this one.
## Comment 6607
Date: 2006-12-21 02:01:40 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
another (early) christmas present:
(almost) fixed in CVS:
runtime/pathfinder.mx:
fixed implementation on play_doc_tape() (wasn'\''t working at all)
compiler/mil/milprint_summer.c:
fixed code generation for pf:add-doc & pf:del-doc
tests/BugTracker/Tests/pf_add-doc.SF-1617681_s00.xq.in:
pf:add-doc expects at least two arguments: URI and name of the document
tests/BugTracker/Tests/pf_del-doc.SF-1617684_s00.xq.in:
pf:del-doc works only with document names, not with URIs
The tests for pf:add-doc & pf:del-doc now work fine;
at the expense of an ugly HACK in runtime/pathfinder.mx:
With _shredlock_set() being a void proc,
the multiplexed _shredlock_set() call in _shred_doc_base() triggers either a
!FATAL: BATpropcheck: BAT tmp_335(-221) has inconsistent descriptor 32768 (100000)
(when using pf:add-doc()) or a
!WARNING: [_shredlock_set]: empty result, hence return type unknown.
(when using pf:del-doc()).
This seems to be a bug in the multiplexer that I could not locate, yet.
Make _shredlock_set() "fake-return" a bit avoids the problem.
An alternative workarount to the fake-return would be to replace the
multiplexed _shredlock_set() call in _shred_doc_base() by a batloop-ed version.
TODO:
Someone needs to analyse and fix the multiplexer problem!
Another problem is that after the pf:del-doc test, restarting the Mserver (with -d10) fails with
!FATAL: BATpropcheck: BAT tmp_34(-28) has inconsistent descriptor 32768 (100000).
See new bug report at
http://sourceforge.net/tracker/index.php?func=detail&aid=1619816&group_id=56967&atid=482468
## Comment 6608
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1617681 at http://sourceforge.net/support/tracker.php?aid=1617681
| XQ: pf:add-doc does not work | https://api.github.com/repos/MonetDB/MonetDB/issues/1179/comments | 0 | 2020-11-30T09:24:57Z | 2024-06-27T11:21:57Z | https://github.com/MonetDB/MonetDB/issues/1179 | 753,307,558 | 1,179 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-12 02:23:56 +0100
From: Ronald Oussoren <<ronaldoussoren>>
To: @njnes
Version: -- development
Last updated: 2007-01-15 09:46:51 +0100
## Comment 6601
Date: 2006-12-12 14:23:56 +0100
From: Ronald Oussoren <<ronaldoussoren>>
A (Python) database abstraction layer that I use generates statements where expressions make heavy use of parentheses. An example:
SELECT * FROM myTable
WHERE
key LIKE ("foo%")
This confuses the SQL parser in the current stable branch of monet:
sql> select * from myTable where key LIKE ('hel%');
MAPI = monetdb@localhost:50000
QUERY = select * from myTable where key LIKE ('hel%');
ERROR = !syntax error, unexpected '(' in: "select * from mytable where key like ("
The same query works correctly with PostgreSQL, and supposedly other database engines as well.
## Comment 6602
Date: 2007-01-15 21:46:51 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
Fixed parser and added a simple test in src/test/BugTracker/Tests/like_exp.SF-1613949
## Comment 6603
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1613949 at http://sourceforge.net/support/tracker.php?aid=1613949
| SQL parsing for LIKE expressions | https://api.github.com/repos/MonetDB/MonetDB/issues/1178/comments | 0 | 2020-11-30T09:24:54Z | 2024-06-27T11:21:56Z | https://github.com/MonetDB/MonetDB/issues/1178 | 753,307,527 | 1,178 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-12 02:06:24 +0100
From: Ronald Oussoren <<ronaldoussoren>>
To: GDK devs <<bugs-common>>
Version: -- development
Last updated: 2007-02-19 02:12:21 +0100
## Comment 6598
Date: 2006-12-12 14:06:24 +0100
From: Ronald Oussoren <<ronaldoussoren>>
(Tested with a 4.14 snapshot downloaded on december 12, 2006 on a SLES9 Linux system that's up-to-date w.r.t. patches).
The _MapiLib extension is build using automake/libtool instead of distutils. There are two problems with this:
1) The extension is build as _MapiLib.so.1.0.0 in a subdirectory, which means you have to move files around to make the extension available to the python interpreter
2) Python extensions aren't shared libraries on all Python platforms. The MachO file format used on Mac OS X has a seperate format for loadable bundles and that format is used for python extensions.
IMHO the build procedure for the python exensions should use distutils instead of the current mechanism.
cd $PREFIX; find ./lib/python2.4 says:
./lib/python2.4/site-packages
./lib/python2.4/site-packages/MonetSQLdb
./lib/python2.4/site-packages/MonetSQLdb/__init__.py
./lib/python2.4/site-packages/MonetSQLdb/converters.py
./lib/python2.4/site-packages/MonetSQLdb/cursors.py
./lib/python2.4/site-packages/MonetSQLdb/monetexceptions.py
./lib/python2.4/site-packages/MonetSQLdb/times.py
./lib/python2.4/site-packages/MapiLib.py
./lib/python2.4/site-packages/CMapi.py
./lib/python2.4/site-packages/lib
./lib/python2.4/site-packages/lib/_MapiLib.so.0.0.0
./lib/python2.4/site-packages/lib/_MapiLib.so.0
./lib/python2.4/site-packages/lib/_MapiLib.so
./lib/python2.4/site-packages/lib/_MapiLib.la
This shows that the _MapiLib extensions is installed in a subdirectory and with the wrong (although there is a symlink with the right name). My current workaround is moving lib/python2.4/site-packagdes/lib/* one level up in the directory tree.
## Comment 6599
Date: 2007-02-19 14:12:21 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
You are right that the extension should be built using distutils, but we're unlikely to make that happen.
I understand that you are working on a new implementation that presumably won't have this problem.
## Comment 6600
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1613934 at http://sourceforge.net/support/tracker.php?aid=1613934
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-common@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| Python extensions building | https://api.github.com/repos/MonetDB/MonetDB/issues/1177/comments | 0 | 2020-11-30T09:24:51Z | 2024-06-28T07:36:51Z | https://github.com/MonetDB/MonetDB/issues/1177 | 753,307,499 | 1,177 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-12 01:29:50 +0100
From: Ronald Oussoren <<ronaldoussoren>>
To: @sjoerdmullender
Version: -- development
Last updated: 2007-02-19 02:15:05 +0100
## Comment 6595
Date: 2006-12-12 13:29:50 +0100
From: Ronald Oussoren <<ronaldoussoren>>
When the configured prefix for MonetDB isn't the same as the prefix for the used python the python bindings are installed in ${monet_prefix}/${python_prefix}, where Python won't see them.
(I'm using a python with sys.prefix == "/opt/python2.4" and use "/opt/monetdb" as the --prefix for monet)
Solution: install the python bindings using distutils, that will install the files in the correct site-packages directory and makes them available to python scripts.
I'm currently using a workaround: install a "monet.pth" file in the site-packages directory for python that contains the path the site-packages directory in the monet tree.
## Comment 6596
Date: 2007-02-19 14:15:05 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
This is by design. It is also documented that you should set PYTHONPATH.
## Comment 6597
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1613913 at http://sourceforge.net/support/tracker.php?aid=1613913
| Python bindings installed in wrong location | https://api.github.com/repos/MonetDB/MonetDB/issues/1176/comments | 0 | 2020-11-30T09:24:49Z | 2024-06-28T07:36:51Z | https://github.com/MonetDB/MonetDB/issues/1176 | 753,307,467 | 1,176 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-12 01:08:50 +0100
From: Ronald Oussoren <<ronaldoussoren>>
To: @njnes
Version: -- development
Last updated: 2007-01-15 09:49:41 +0100
## Comment 6592
Date: 2006-12-12 13:08:50 +0100
From: Ronald Oussoren <<ronaldoussoren>>
This is probably bad SQL-fu on my part, but you never know :-). I'm
trying to insert a 0-byte blob into a table, but that doesn't work for
me. The SQL code below fails on the second insert, the database
complains about a bad blob syntax. Inserting an implicit NULL into a
BLOB colum also fails (the 3th insert), even though inserting an
explicit NULL (the fourth one) does succeed.
CREATE TABLE testtable (
id INTEGER
data BLOB
);
INSERT
INTO testtable (id, data)
VALUES (0, BLOB '00');
INSERT
INTO testtable (id, data)
VALUES (1, BLOB '');
INSERT
INTO testtable (id)
VALUES (2);
INSERT
INTO testtable (id, data)
VALUES (2, NULL);
## Comment 6593
Date: 2007-01-15 21:49:18 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
The BLOB with empty string, we concider an incorrect value, ie use NULL instead.
Added a test for this, the others now work correctly
See src/test/BugTracker/Tests/blob_corner_cases.SF-1613902
## Comment 6594
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1613902 at http://sourceforge.net/support/tracker.php?aid=1613902
| SQL BLOB edgecases | https://api.github.com/repos/MonetDB/MonetDB/issues/1175/comments | 0 | 2020-11-30T09:24:46Z | 2024-06-27T11:21:53Z | https://github.com/MonetDB/MonetDB/issues/1175 | 753,307,428 | 1,175 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-11 03:44:47 +0100
From: @drstmane
To: Jan Rittinger <<jan.rittinger>>
Version: -- development
Last updated: 2006-12-13 01:36:33 +0100
## Comment 6581
Date: 2006-12-11 15:44:47 +0100
From: @drstmane
After the latest checkin
===================================================================
2006/12/11 - teubner: compiler/mil/milprint_summer.c,1.298.2.13
compiler/semantics/xquery_fo.c,1.89.2.3
Fixes by Jan:
-- Fixed aggregates (fn:max(), fn:min(), ...) for the case of an
empty input.
===================================================================
the following Pathfinder tests fail as follows:
tests/BugTracker/Tests/if-condition_testing_empty_sequence.SF-1589820.xq:
!type error: body of fn:depthKO() evaluates to `integer?' which is not a subtype of `xs:integer'
tests/WebSite/Tests/voc3.xq:
produces different output:
stable/expected:
<XQueryResult>[...]<onboard>125</onboard>[...]<onboard>190</onboard>[...]<onboard>352</onboard>[...]<onboard>250</onboard>[...]<onboard>120</onboard>[...]</XQueryResult>
now:
<XQueryResult>[...]<onboard>125 0</onboard>[...]<onboard>190 0</onboard>[...]<onboard>352 0</onboard>[...]<onboard>250 0</onboard>[...]<onboard>120 0</onboard>[...]</XQueryResult>
tests/XQuery/Tests/fn_sum.xq:
produces different output:
stable/expected:
<XQueryResult>0 100 4 100 154.300000 100 -23 100 2 100 0 100 4.000000 100 154.300000</XQueryResult>
now:
<XQueryResult>100 4 100 154.300000 100 100 2 100 100 4.000000 100 154.300000</XQueryResult>
benchmarks/MBench/Tests/qa06.xq:
!type error: body of my:depth() evaluates to `integer?' which is not a subtype of `xs:integer'
benchmarks/XPathMark/Tests/Q45.xq:
produces different output:
(large) result nodes missing
## Comment 6582
Date: 2006-12-12 00:50:13 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
tests/BugTracker/Tests/if-condition_testing_empty_sequence.SF-1589820.xq
and
benchmarks/MBench/Tests/qa06.xq
can by "silenced" by the following patch of the test scripts:
Index: benchmarks/MBench/Tests/qa06.xq.in
===================================================================
RCS file: /cvsroot/monetdb/pathfinder/benchmarks/MBench/Tests/qa06.xq.in,v
retrieving revision 1.1
diff -u -0 -r1.1 qa06.xq.in
--- benchmarks/MBench/Tests/qa06.xq.in 14 Oct 2006 17:29:17 -0000 1.1
+++ benchmarks/MBench/Tests/qa06.xq.in 11 Dec 2006 23:40:31 -0000
@@ -2 +2 @@
-declare function my:depth($e as node()*) as xs:integer
+declare function my:depth($e as node()*) as xs:integer?
Index: tests/BugTracker/Tests/if-condition_testing_empty_sequence.SF-1589820.xq
===================================================================
RCS file: /cvsroot/monetdb/pathfinder/tests/BugTracker/Tests/if-condition_testing_empty_sequence.SF-1589820.xq,v
retrieving revision 1.1
diff -u -0 -r1.1 if-condition_testing_empty_sequence.SF-1589820.xq
--- tests/BugTracker/Tests/if-condition_testing_empty_sequence.SF-1589820.xq 3 Nov 2006 10:58:25 -0000 1.1
+++ tests/BugTracker/Tests/if-condition_testing_empty_sequence.SF-1589820.xq 11 Dec 2006 23:40:32 -0000
@@ -1 +1 @@
-declare function depthKO($x as node()) as xs:integer
+declare function depthKO($x as node()) as xs:integer?
@@ -9 +9 @@
-declare function depthOK($x as node()) as xs:integer
+declare function depthOK($x as node()) as xs:integer?
I'm not quite sure whether this is in line with the desired/correct typing.
Could some expert please check/confirm?
The extra "0" in tests/WebSite/Tests/voc3.xq are IMHO definitely wrong, just as the missing "-23" for tests/XQuery/Tests/fn_sum.xq, but JanR seems to have/be busy with a fix for these.
The missing "0" might actually be correct, in case sum(()) and sum((),()) are supposed to return an empty sequence instead of "0".
## Comment 6583
Date: 2006-12-12 10:45:26 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
Fixes are now checked in and all 5 queries return the expected result again.
@Stefan I did ignore your suggestions as I wanted to avoid changing the semantic of the query (-- both queries return a single integer value).
## Comment 6584
Date: 2006-12-12 11:49:47 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Jan,
thanks for the fix and the lesson about typing!
Unfortunately, while fixing the five tests listed here, you fix now breaks the following tests
$ pf tests/XQuery/Tests/fn_avg.xq
type error: can't cast type 'integer,integer,integer,integer,integer' to type 'double+'
halted in /net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Stable/pathfinder/compiler/mil/milprint_summer.c (translateCast), line 3607
$ pf tests/XQuery/Tests/fn_min.xq
type error: can't cast type 'string,string' to type 'string+'
halted in /net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Stable/pathfinder/compiler/mil/milprint_summer.c (translateCast), line 3607
$ pf tests/XQuery/Tests/fn_max.xq
type error: can't cast type 'string,string' to type 'string+'
halted in /net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Stable/pathfinder/compiler/mil/milprint_summer.c (translateCast), line 3607
## Comment 6585
Date: 2006-12-12 18:58:58 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
For benchmarks/MBench/Tests/qa06.xq we still get an error. I wrapped the generated {max} call by the following counts and my conclusion is that the implementation of {max} has to be screwed up.
print("item.count()");
print(item.slice(250,260));
print(item.leftfetchjoin(int_values).slice(250,260));
print(iter_grp.slice(250,260));
print(iter.slice(250,260));
if (item.materialize(iter).count() != iter.count()) ERROR("foo");
var iter_aggr := {max}(item.leftfetchjoin(int_values), iter_grp, iter).tmark(0@0);
[isnil](iter_aggr).select(true).print();
(I will check in a fix for the others problematic queries tomorrow morning.)
## Comment 6586
Date: 2006-12-13 03:58:16 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Jan,
benchmarks/MBench/Tests/qa06.xq, like the other 4 tests originally mentioned in this bug report, do work fine for me since your recent checkin that fixed the typing and sum issues.
However, I did fix some problems with [void,void] in MonetDB's aggrX3 in the MonetDB_4-14 release branch recently, that used to cause problems with benchmarks/MBench/Tests/qa06.xq before. They should have been propagated to the development trunk; hence, updating your MonetDB shoudl hopefully solve the problem. Let me know, whether it indeed does or not ...
## Comment 6587
Date: 2006-12-13 09:54:47 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
Ok - compiling MonetDB helped. Bug is now solved in CVS.
## Comment 6588
Date: 2006-12-13 11:56:27 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
We are getting there:
tests/XQuery/Tests/fn_avg.xq, tests/XQuery/Tests/fn_max.xq, tests/XQuery/Tests/fn_min.xq
do work fine, now.
However, now I get
$ pf tests/Update/Tests/update.xq type error: can't cast type 'string' to type 'QName'
halted in /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/mil/milprint_summer.c (translateCast), line 3570
???
## Comment 6589
Date: 2006-12-13 13:10:00 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
this one is now also fixed (any more problems :))
## Comment 6590
Date: 2006-12-13 13:36:33 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Indeed, this seems now fixed without further "damage".
Thanks!
## Comment 6591
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1613189 at http://sourceforge.net/support/tracker.php?aid=1613189
| XQ: tests fails after aggregates (fn:max(), ...) fixes | https://api.github.com/repos/MonetDB/MonetDB/issues/1174/comments | 0 | 2020-11-30T09:24:43Z | 2024-06-27T11:21:52Z | https://github.com/MonetDB/MonetDB/issues/1174 | 753,307,391 | 1,174 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-10 09:28:22 +0100
From: @peterboncz
To: @sjoerdmullender
Version: -- development
Last updated: 2007-01-05 06:27:16 +0100
## Comment 6576
Date: 2006-12-10 21:28:22 +0100
From: @peterboncz
voortbordurend op de setting beschreven in
1612739 XQ: attr_own not set to nil on delete
kun je het volgende doen op het 1Gb XMark document:
B.xq
C.xq
M.xq
C.xq
B.xq
C.xq
Ik heb twee dingen zien gebeuren:
- een ERROR in merge1 := merge_union()
- foute XML als resultaat. Dat komt uit bij de laatste C.xq, waarin de subtree van de tweede africa subtree veel te klein lijkt
de eerste fout is in milprint_summer code die de insert code bij elkaar raapt. Merk op dat op dat moment er al twee <africa> subtrees zijn.
## Comment 6577
Date: 2006-12-10 21:57:14 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: YES
Ik heb het volgende nog ontdekt:
als je een B.xq doet, insert je dus 14K elt nodes; dus maak je 1 nieuwe pagina aan, en vult een pagina uit (3K elt nodes in een bestaande page en de rest in een nieuwe).
In dezelfde sessie is het C.xq totaal correct; herstart je de server, dan is het totaal slechts 3K nodes.
Oftewel, de logger ignored waarschijnlijk alle updates die buiten het bat bereik vallen. Maar: log updates buiten het bestaande range dienen tot een BAT extend te leiden.
Misschien komt het omdat ik uit perf overwegingen het sorteren van de log_delta bats weggehaald heb?
Wat mij betreft hoeft de log niet gesorteerd te zijn. Je kunt evt ruimte in een void,X bat tussen waar de log action will replacen en het vroegere einden voorlopig met X(nil) uitvullen.
Het gevolg van het weggooien van de pagina is corrupte XML, en de merged_union fout in tweede instantie is daar dan hoogstwaarschijnlijk het gevolg van.
## Comment 6578
Date: 2006-12-10 22:19:37 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: YES
foutje=> ik had het over 14K nodes eerder, maar het zijn er 143K!
Hm, misschien ligt het niet aan de logger, want ook in dezelfde sessie gebeurt het:
B.xq
C.xq => 143K nodes
B.xq
C.xq => minder nodes
binnen dezelfde sessie sluit uit dat het aan de logger ligt eigenlijk. Dus waarschijnlijk ligt het aan de multi-page insert code in pf_support.
merk ok dat C.xq gebruik maakt van de nsloc index, dus het zou ook index corruption kunenn zijn.
Maar index corruption alleen niet, want na een restart (die de index herbouwd) is het ook nog fout. Dus waarschijnlijker ligt het aan de multi-page insert code in pf_support.
## Comment 6579
Date: 2007-01-05 18:27:16 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
The bug seems to be fixed.
No test since a test would be too big and slow.
## Comment 6580
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1612758 at http://sourceforge.net/support/tracker.php?aid=1612758
| XQ: repeated big insert causes problems | https://api.github.com/repos/MonetDB/MonetDB/issues/1173/comments | 0 | 2020-11-30T09:24:40Z | 2024-06-27T11:21:51Z | https://github.com/MonetDB/MonetDB/issues/1173 | 753,307,361 | 1,173 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-10 09:06:03 +0100
From: @peterboncz
To: @sjoerdmullender
Version: -- development
Last updated: 2007-01-04 09:57:34 +0100
## Comment 6570
Date: 2006-12-10 21:06:03 +0100
From: @peterboncz
Mijn test setting is het 1GB XMark document.
neem module /tmp/mod.xq
en de volgende queries:
G.xq - laat de node zien
C.xq - laat de groottes van de test nodes zien (na een B.xq kun je G,xq niet meer praktisch gebruiken)
M.xq - insert een test node
D.xq - delete alle test nodes
B.xq - BIG test node insert (bevat africa subtree 16K elements)
(deze files staan in /ufs/boncz/BUG/)
Er is in deletes nog een bug die betrekking heeft op de attributes. de ATTR-OWN waardes van gedelete nodes worden niet op nil gezet.
recept:
een paar keer M.xq doen en met G.xq het resultaat bekijken
dan D.xq doen (alle test nodes worden weggegooid)
server herstarten
een paar keer M.xq doen en met G.xq het resultaat bekijken
opeens zullen er duplicate attributen van de test node verschijnen. dwz de nids worden hebruikt, en nu matchen de oude ATTR_OWN entries ook met de nieuwe elements
De code in do_update_delete(), zie onder, lijkt er vanuit te gaan dat de attributes als nodes met type ATTR worden meegegeven, maar volgens mij worden er alleen "echte" nodes (NIDs) meegegeven. De attributen zou je met voorgenoemde get_attr_own() uit de element NIDs moeten ophalen, denk ik.
var conts := update_node_kind.get_container();
var types := update_node_kind.get_types();
var attrs := types.ord_uselect(ATTR);
var elems := types.ord_uselect(ELEM);
if (attrs.count() > 0) {
attrs := attrs.mirror(); [i,i]
var aconts := attrs.leftjoin(conts); [i,CONT]
{
var modified_attr := ws.fetch(MODIFIED_ATTR);
modified_attr.insert(aconts.reverse().join(update_node_item));
}
aconts.tunique()@batloop() {
var cont := $h;
var list := aconts.ord_uselect(cont).mirror(); [i,i]
var attr_prop_updates := ws.fetch(ATTR_PROP_UPDATE).find(cont);
var attr_own_updates := ws.fetch(ATTR_OWN_UPDATE).find(cont);
var attr_qn_updates := ws.fetch(ATTR_QN_UPDATE).find(cont);
var attrlist := list.leftjoin(update_node_item).reverse().project(oid_nil);
attr_prop_updates.insert(attrlist);
attr_own_updates.insert(attrlist);
attr_qn_updates.insert(attrlist);
}
}
## Comment 6571
Date: 2006-12-10 21:29:42 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: YES
o ja, ik shred het 1GB XMark document met:
shred_doc(url, "aaa.xml", "aaa.xml", 10LL);
waarbij op db2 je de doc url is:
/export/scratch1/manegold/ExpDB/Benchmarks/XMark/docs/XMark-f10.xml
## Comment 6572
Date: 2006-12-10 22:09:33 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: YES
foutje:
D.xq (de bulk insert) insert niet 16K nodes, maar 143K nodes. Veel groter dus.
## Comment 6573
Date: 2006-12-15 13:32:31 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
I think the problem is that when you delete a node, its attributes aren't deleted. The quoted code is needed for the case when you delete an attribute, but when you delete a whole node, all affected nodes' attributes should also be deleted.
Will look into this.
## Comment 6574
Date: 2007-01-04 09:57:34 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Attributes are now cleared on delete.
Also added a test: tests/BugTracker clear_attrs_on_delete.SF-1612739{,-b,-c}
## Comment 6575
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1612739 at http://sourceforge.net/support/tracker.php?aid=1612739
| XQ: attr_own not set to nil on delete | https://api.github.com/repos/MonetDB/MonetDB/issues/1172/comments | 0 | 2020-11-30T09:24:37Z | 2024-06-27T11:21:51Z | https://github.com/MonetDB/MonetDB/issues/1172 | 753,307,325 | 1,172 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-09 06:46:57 +0100
From: @drstmane
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-12-13 09:03:44 +0100
## Comment 6563
Date: 2006-12-09 18:46:57 +0100
From: @drstmane
While running fine before, I get the follwoing after the latest checkins (see below):
$ time Mtest_pathfinder -q -A0 runtime
.....
* (procs.test.out.FILTERED) significantly
.............
* (descendant.test.out.FILTERED) significantly
.
* (descendant-or-self.test.out.FILTERED) significantly
...
* (child.test.out.FILTERED) significantly
..........
!ERROR: Testing FAILED SIGNIFICANTLY !
4 out of 32 tests produced SIGNIFICANTLY different output
real 0m18.548s
user 0m13.969s
sys 0m3.900s
$ time Mtest_pathfinder -q -A0 tests/Update
........
* (symm-test-5.test.out.FILTERED) significantly
.
* (update.test.out.FILTERED) significantly
* (update.test.err.FILTERED) significantly
!ERROR: Testing FAILED SIGNIFICANTLY !
2 out of 9 tests produced SIGNIFICANTLY different output
real 0m18.989s
user 0m7.692s
sys 0m1.584s
===================================================================
2006/12/09 - boncz: compiler/mil/milprint_summer.c,1.298.2.12
- fix bug with constants in delete updates
(surfaced on empty deletes)
===================================================================
2006/12/09 - boncz: runtime/pathfinder.mx,1.269.2.15
runtime/pf_support.mx,1.172.2.12
various performance fixes; small inserts and deletes are now reasonably fast
- index new qnames in qnloc index
- better handling of the hash tables on the qn_loc index
- shortened the collection lock time by moving most of the logging
out of the critical period
===================================================================
## Comment 6564
Date: 2006-12-10 12:08:05 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Peter fixed tests descendant, descendant-or-self, & child in runtime/Tests/ .
I approved the new output of test runtime/Tests/procs.milS accoring to Peter's yesterday's changes.
Test tests/Update/Tests/symm-test-5.xq still fails, returning wrong result:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-5.out.00.html
Test tests/Update/Tests/update.xq still fails with "!ERROR: BATorder: access denied to tmp_441, aborting.":
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/update.err.00.html
## Comment 6565
Date: 2006-12-11 10:11:51 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
For the records:
After checkin
===================================================================
2006/12/10 - boncz: runtime/pathfinder.mx,1.269.2.17
runtime/pf_support.mx,1.172.2.14
- fix in xquery client handling
(sometimes enters in an endless loop because stream_read()
keeps returning two empty bytes)
- performance improvement in nsloc index maintenance
(aka hash maintenance is really slow in MonetDB)
- modified_nids: only administer the "NID-pages" where
modified nids are -- not all nids (1024 times less data)
- numerous fixes in insert code that does (multiple) page inserts
(it still does not work -- regrettably)
===================================================================
various update tests fail with
!ERROR: BATappend: access denied to tmp_755, aborting.
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/treat_as.SF-1586454.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/insert_multiple.SF-1590580.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/inserting_multiple_elements.SF-1590583.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/insert-2.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-1.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-2.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-3.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-4.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-5.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/update.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_AddressBook/Q1x.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4a.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4ax.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4b.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4bx.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4c.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4cx.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_R/Q7.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_R/Q7x.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_XRpc/insertNode.err.00.html
See also (e.g.)
Mtest_pathfinder -q -A0 -r tests/Update/ tests/W3C_use_cases/XQUF/
(takes less than a minute to run).
## Comment 6566
Date: 2006-12-12 11:57:06 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
While Peter's last night's checkins do fix most of the update tests (Thanks!),
tests
tests/BugTracker/Tests/treat_as.SF-1586454.xq &
tests/Update/Tests/update.xq
still fail (, now) with
Mserver: /net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Stable/MonetDB/src/monet/monet_interpreter.mx:1543: interpret_batcast: Assertion `tpe != 7' failed.
... but only with 64-bit OIDs:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/treat_as.SF-1586454.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/update.err.00.html
## Comment 6567
Date: 2006-12-13 12:02:41 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
I just fixed the assertion problem with 64-bit OID in
tests/BugTracker/Tests/treat_as.SF-1586454.xq &
tests/Update/Tests/update.xq
by making the tail-sorted-property propagation of MonetDB's interpret_batcast properly aware of (not that) "new" type wrd.
This leaves only tests/Update/Tests/symm-test-5.xq to be fixed.
## Comment 6568
Date: 2006-12-13 21:03:44 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
For the bookkeeping:
tests/Update/Tests/symm-test-5.xq seems to be fixed by Peter's latest checkins.
Thank you very much!
## Comment 6569
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1612267 at http://sourceforge.net/support/tracker.php?aid=1612267
| XQ: update tests fail after latest checkins | https://api.github.com/repos/MonetDB/MonetDB/issues/1171/comments | 0 | 2020-11-30T09:24:34Z | 2024-06-27T11:21:50Z | https://github.com/MonetDB/MonetDB/issues/1171 | 753,307,283 | 1,171 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-08 03:31:43 +0100
From: Jan Rittinger <<jan.rittinger>>
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2007-05-06 02:51:05 +0200
## Comment 6556
Date: 2006-12-08 15:31:43 +0100
From: Jan Rittinger <<jan.rittinger>>
When running query q02 from the testset benchmarks/XBench/TC/SD with the algebra version a duplicate <hw>knead</hw> magically appears.
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.32.32.d-Darwin8.8.0/benchmarks_XBench_TC_SD/q02.out.00.html
## Comment 6557
Date: 2006-12-08 15:55:52 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: YES
Evaluating the query http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.32.32.d-Darwin8.8.0/benchmarks_XBench_TC_MD/q17.out.00.html with the algebra variant also produces a large number of duplicates.
## Comment 6558
Date: 2007-02-01 14:56:51 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: YES
I attached the generated MIL script for q17 (with annotations where the error occurs). The ``profile'' path step does not return 26 results as required but 4862 nodes (-- the input contains 4862 nodes with 26 distinct values).
One can use Mtest to load a database with the files in benchmarks/XBench/TC/MD/Tests and only has to replace $TSTTRGDIR in q17.mil with the path to load_docs.xml.
File Added: q17.mil
## Comment 6559
Date: 2007-02-19 15:00:52 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Assigning to Mr. Algebra.
## Comment 6560
Date: 2007-02-19 16:48:10 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: YES
It's MIL code that calls path steps (like mps does). So perhaps Stefan might have a look...
## Comment 6561
Date: 2007-05-06 14:51:05 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Fixed in CVS:
loop-lifted child implementation did not properly cope with the case that the same item (context node) occurs more than once for the same iter, i.e., in case [iter,item] ([iter,ctx]) is not a set (i.e., contains duplicates).
Fix: avoid producing duplicates in such cases in ll_child result generation.
ToDo:
Can we do this more efficiently, e.g.,
by upfront ensuring that [iter,item] ([iter,ctx])
is a set, i.e., duplicate free??
Added tests in
benchmarks/XBench/TC/SD/Tests/q02alg.*
benchmarks/XBench/TC/MD/Tests/q17alg.*
## Comment 6562
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1611545 at http://sourceforge.net/support/tracker.php?aid=1611545
| XQ: wrong path step result (duplicates) | https://api.github.com/repos/MonetDB/MonetDB/issues/1170/comments | 0 | 2020-11-30T09:24:32Z | 2024-06-27T11:21:49Z | https://github.com/MonetDB/MonetDB/issues/1170 | 753,307,253 | 1,170 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-08 11:14:40 +0100
From: @drstmane
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2006-12-08 06:50:38 +0100
## Comment 6551
Date: 2006-12-08 11:14:40 +0100
From: @drstmane
Before this information get's lost:
While "somehow" being fixed on Thu Nov 9 2006 (cf.
http://sourceforge.net/tracker/index.php?func=detail&aid=1590066&group_id=56967&atid=482468
),
the tests in
pathfinder/tests/W3C_use_cases/XQUF/Parts/
(apparently unnoticedly) started failing, again, after checkins on Thu Nov 23 2006 and did so until Fri Dec 1 2006.
(Due to unattended compilation problems of Pathfinder and/or MonetDB, no pathfinder testing was performed between Fri Dec 1 2006 & Fri Dec 8 2006.)
During local testing on Thu Dec 7 2006 just before the below checkins, all tests worked fine, again.
But during testing on Fri Dec 8 2006 ust after the below checkins, all tests fail, again, now with
!ERROR: mposjoin: missing matches. The result requires the same number of tuples as the first two input arguments.
=!ERROR: CMDmposjoin: operation failed.
!ERROR: interpret_params: mposjoin(param 1): evaluation error.
!ERROR: interpret: no matching MIL operator to 'printf(void)'.
!MAYBE YOU MEAN:
! printf(str, ..any..) : void
(This *might* be related to
http://sourceforge.net/tracker/index.php?func=detail&aid=1611373&group_id=56967&atid=482468
)
For details, see:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q2.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q3a1.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q3a2.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4a.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4ax.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4b.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4bx.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4c.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q4cx.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q6.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_Parts/Q6x.out.00.html
The "problematic" checkin was:
===================================================================
2006/12/07 - boncz: compiler/mil/milprint_summer.c,1.298.2.11
runtime/ll_upwards.mx,1.15.2.3 runtime/pathfinder.mx,1.269.2.13
runtime/pf_support.mx,1.172.2.10 runtime/serialize.mx,1.79.2.1
- fix performance issue with ATTR_OWN
the access path often goes from nodes to their aatributes,
which needs a hash-table on OWN (i.e. the NIDs) in ATTR_OWN.
However, updatable collections get a isolated copy of ATTR_OWN,
thus each query starts without hash table.
Creating the hash table cots 1 sec on the 1GB XMark dataset.
Thus each trivial query took at least 1 sec.
The solution is to add a runtime index. It is an old copy of
ATTR_OWN with hash table on it. The access path uses both this
shared index (for the bulk of the work) and the latest private
copy of ATTR_OWN.
This holds both for query processing and serialization.
- fix a bug in the commit after a shred: first commit in the
WAL and then do the partial checkpoint.
plus some more per tuning:
- remove sorting from the logging phase.
- some tuning of the ll_upwards parameters
===================================================================
## Comment 6552
Date: 2006-12-08 16:48:15 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Reverting
===================================================================
@@ -1896,7 +1896,7 @@
"var oid_iter := iter;\n"
"var oid_item := item.materialize(ipik);\n"
"var oid_cont := kind.get_container();\n"
- "var temp1 := mvaljoin ( mposjoin(oid_item, oid_cont, ws.fetch(PRE_NID)), oid_cont, ws.fetch(ATTR_OWN));\n"
+ "var temp1 := get_attr_own (ws, mposjoin(oid_item, oid_cont, ws.fetch(PRE_NID)), oid_cont);\n"
"oid_item := nil;\n"
"oid_cont := temp1.hmark(0@0).leftfetchjoin(oid_cont);\n"
"var oid_attr := temp1.tmark(0@0);\n"
===================================================================
of Peter's yesterday's late night changes
Index: compiler/mil/milprint_summer.c
===================================================================
RCS file: /cvsroot/monetdb/pathfinder/compiler/mil/milprint_summer.c,v
retrieving revision 1.298.2.10
retrieving revision 1.298.2.11
diff -u -r1.298.2.10 -r1.298.2.11
--- compiler/mil/milprint_summer.c 7 Dec 2006 17:42:26 -0000 1.298.2.10
+++ compiler/mil/milprint_summer.c 7 Dec 2006 23:07:08 -0000 1.298.2.11
@@ -815,7 +815,7 @@
/* get the attributes of the subtree copy elements */
"{ create attribute subtree copies\n"
- "var temp_attr := mvaljoin( mposjoin(res_item, res_cont, ws.fetch(PRE_NID)), res_cont, ws.fetch(ATTR_OWN));\n"
+ "var temp_attr := get_attr_own(ws, mposjoin(res_item, res_cont, ws.fetch(PRE_NID)), res_cont);\n"
"var oid_attr := temp_attr.tmark(0@0);\n"
"var oid_cont;\n"
"if (is_constant(res_cont)) {\n"
@@ -1126,7 +1126,7 @@
milprintf(f, "var vid := v_vid%03u.ord_uselect(%i@0);\n",
cur_level, f->module_base + v->vid);
milprintf(f, "vid := vid.hmark(0@0);\n");
- milprintf(f, "iter := vid.leftfetchjoin(v_iter%03u);\n", cur_level);
+ milprintf(f, "iter := vid.leftfetchjoin(v_iter%03u).assert_order();\n", cur_level);
milprintf(f, "pos := vid.leftfetchjoin(v_pos%03u);\n", cur_level);
milprintf(f, "item := vid.leftfetchjoin(v_item%03u);\n", cur_level);
milprintf(f, "kind := vid.leftfetchjoin(v_kind%03u);\n", cur_level);
@@ -1896,7 +1896,7 @@
"var oid_iter := iter;\n"
"var oid_item := item.materialize(ipik);\n"
"var oid_cont := kind.get_container();\n"
- "var temp1 := mvaljoin ( mposjoin(oid_item, oid_cont, ws.fetch(PRE_NID)), oid_cont, ws.fetch(ATTR_OWN));\n"
+ "var temp1 := get_attr_own (ws, mposjoin(oid_item, oid_cont, ws.fetch(PRE_NID)), oid_cont);\n"
"oid_item := nil;\n"
"oid_cont := temp1.hmark(0@0).leftfetchjoin(oid_cont);\n"
"var oid_attr := temp1.tmark(0@0);\n"
@@ -2799,7 +2799,7 @@
"var oid_preNew := content_preNew_preOld.hmark(0@0);\n"
"content_preNew_preOld := nil;\n"
"var oid_cont := oid_preNew.leftfetchjoin(preNew_cont);\n"
- "var temp_attr := mvaljoin( mposjoin(oid_preOld, oid_cont, ws.fetch(PRE_NID)), oid_cont, ws.fetch(ATTR_OWN));\n"
+ "var temp_attr := get_attr_own(ws, mposjoin(oid_preOld, oid_cont, ws.fetch(PRE_NID)), oid_cont);\n"
"oid_preOld := nil;\n"
"var oid_attr := temp_attr.tmark(0@0);\n"
"oid_cont := temp_attr.reverse().leftfetchjoin(oid_cont);\n"
===================================================================
seems to fix this bug and http://sourceforge.net/tracker/index.php?func=detail&aid=1611373&group_id=56967&atid=482468 ,
without causing any other (obvious) harm.
I'm not sure, though, whether this is a valid fix, or whether it would simply undo (part of) Peter's performance improvements...
Peter,
Could you please comment?
Thanks!
## Comment 6553
Date: 2006-12-08 16:57:57 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
Hi Stefan,
No reverting is not an option.
There is probably something wrong still with get_attr_own(), under certain circumstances.
Peter
## Comment 6554
Date: 2006-12-08 18:50:38 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
fixed in CVS:
fixed get_attr_own:
after filtering false hits out of "shared", make sure that the final result
indeed holds the right oids in head and tail, i.e., a subset of the original "shared".
## Comment 6555
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1611400 at http://sourceforge.net/support/tracker.php?aid=1611400
| XQ: tests/W3C_use_cases/XQUF/Parts/ fail after late checkins | https://api.github.com/repos/MonetDB/MonetDB/issues/1169/comments | 0 | 2020-11-30T09:24:29Z | 2024-06-27T11:21:48Z | https://github.com/MonetDB/MonetDB/issues/1169 | 753,307,212 | 1,169 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-08 10:46:34 +0100
From: @drstmane
To: Jan Rittinger <<jan.rittinger>>
Version: -- development
Last updated: 2006-12-14 06:33:12 +0100
## Comment 6542
Date: 2006-12-08 10:46:34 +0100
From: @drstmane
Before this information gets lost:
On Fri Dec 1 2006, all XMark tests did work fine.
On Fri Dec 8 2006, XMark query q04 fails in all modularized cases, only, incorrectly yielding an empty result.
(Due to unattended compilation problems of Pathfinder and/or MonetDB, no pathfinder testing was performed between Fri Dec 1 2006 & Fri Dec 8 2006.)
For details, see:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XMark_XRpc/q04.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XMark_mod_MIL/q04.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XMark_mod_XQ/q04.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XMark_opt_MIL/q04.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/benchmarks_XMark_opt_XQ/q04.out.00.html
## Comment 6543
Date: 2006-12-12 15:07:32 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
[After a replaying a week of lost testing by hand:]
This problem is caused (or at least triggered) by
===================================================================
2006/12/01 - boncz: compiler/semantics/normalize.brg,1.11.2.1
The Moerkotte Queries (e.g. /ufs/boncz/moer) seduced me into implementing
the "syntactic variant" solution to the problem discussed by Jens in the wiki:
http://www.inf.uni-konstanz.de/dbis/pathfinder/wiki/index.php/Why_is_Predicate_Processing_Inefficient_in_Pathfinder
basically, this is a rewrite of X/Y[Z] into (X/Y)[Z] if:
- Z is *not* a positional predicate (a boolean or path expr)
- fn:dot, fn:position and fn:last are not used in Z
yes, I was initiated into the sect of burg, finally.
The Moerkotte Queries (e.g. /ufs/boncz/moer) seduced me into implementing
the "syntactic variant" solution to the problem discussed by Jens in the wiki:
http://www.inf.uni-konstanz.de/dbis/pathfinder/wiki/index.php/Why_is_Predicate_Processing_Inefficient_in_Pathfinder
basically, this is a rewrite of X/Y[Z] into (X/Y)[Z] if:
- Z is *not* a positional predicate (a boolean or path expr)
- fn:dot, fn:position and fn:last are not used in Z
yes, I was initiated into the sect of burg, finally.
The rewrite analysis is conservative, as this is done on the abstract syntax tree
(later in the core tree I could not find an easy way to recognize the pattern anymore)
thus we do not have any typing info available.
I do *not* propose to propagate this to the HEAD automatically. The decision whether
this rewrite is worth the while I leave to the Munich folks. It is a thing that helps
mps (a lot, in fact for programmers that use X[Z] XPath notation rather than xquery
notation for $x in X where Z return $x)
It makes the core tree easier, and generally removes a for-loop; but I am not
sure whether the algebra front-end benefits as much from this as mps.
Finding that out thus may give us some additional meta-insight into the power
of the algebra optimizations. See it as an experiment. It took me a day after all,
burg initiation effort inclusive.
Anyway, on our standard benchmarking machine db2 (non-optimized 64-bits 32-bits oids
gcc compile) we now get the following score on the Moerkotte Queries:
MXQ.14 MXQ.12 saxon
q0 0.67 1.71 1.5
q1 0.26 0.33 1.6
q2 0.68 1.23 60
q3 2.38 DNF 209
q4 1.83 2.02 61
q5 2.74 DNF 166
q6 4.82 DNF 130
q7 2312 DNF 211
q8 DNF DNF DNF
q9 DNF DNF DNF
q10 DNF DNF DNF
q11 DNF DNF 151
q12 0.14 DNF ?
this change obviously helped a bit (plus my previous mps checkin that limited the
expansion of user variables -- Guido likes to use complex let statements).
So we generally beat saxon now, but still lose q7 and q11 to it. There is plenty
of room for the algebra version to excel over mps still, though to handle queries
q8-q10 in linear time, we would need 'bandjoin' physical support in the backend
(i.e. inner.val >= outer.lo and inner.val <= outer.hi).
===================================================================
for details, see
cvs diff -r1.11 -r1.11.2.1 compiler/semantics/normalize.brg
Since this looks more like a "feature" than a "bug fix" to me, I would not mind at all, if these changes were rolled-back.
Otherwise, they should obviously be fixed.
## Comment 6544
Date: 2006-12-12 15:28:10 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
In fact, rolling-back the respective changes in compiler/semantics/normalize.brg does only "fix" the "non-optimized versions of XMark q04, i.e., benchmarks/XMark/mod/XQ/Tests/q04.xq.in & benchmarks/XMark/mod/MIL/Tests/q04.xq.in .
The "optimized" versions (benchmarks/XMark/opt/XQ/Tests/q04.xq.in, benchmarks/XMark/opt/MIL/Tests/q04.xq.in & benchmarks/XMark/XRpc/Tests/q04.xq.in) still fail, apparently due to some other change that has checked in since Fri Dec 1 2006benchmarks/XMark/mod/XQ/Tests/q04.xq.in .
(I'm not going to replay more testing by hand. Sorry!)
## Comment 6545
Date: 2006-12-12 15:46:48 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Well, given that the "optimized" versions basically replace
doc($doc)/site/open_auctions/open_auction
by
doc($doc)/descendant::open_auction
and (in particular)
$b/bidder/personref[@person=$person1]
by
($b/descendant::personref)[@person=$person1]
the problem *might* be related to / triggered by (the first point of)
===================================================================
2006/12/01 - boncz: compiler/mil/milprint_summer.c,1.298.2.7
mps again..
- fix or rather path a bug in simplifyCoretree
it checkes for a LR c_for and an LLR that is not c_var
that turned out not to be restrictive enough for let and orderby statements
appearing after the for
[ BUG example: paraphrashing pathfinder/tests/WebSite/Tests/voc2.xq
for $t in doc("voc.xml")//voyage[leftpage/boatname="ZEELANDIA"]
order by zero-or-one($t/leftpage/departure)
return $t/descendant::departure
worked, but the equivalent (note the parenthesis in the firrst pathe expr)
for $t in (doc("voc.xml")//voyage)[leftpage/boatname="ZEELANDIA"]
order by zero-or-one($t/leftpage/departure)
return $t/descendant::departure
did not. That cost me some time to find.. we all love mps ]
- bugfix in exactly-one() and zero-or-one()
two issues came up by the "Moerkotte Queries" (see /ufs.boncz/moer)
- restrict variable expansion with a single hack: do not expand user
introduced variables.
we do not want to expand "expensive" expressions, but I did not
want to define some heuristic for that. The above is trivial.
Compiler introduced variables are expanded (and thus eliminated) as before.
- bugfix in orderby. solved together with Jan R. in a short Skype call.
===================================================================
## Comment 6546
Date: 2006-12-12 23:53:26 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
[for those who care / are interested]
in fact, rolling back the 1.298.2.6 -> 1.298.2.7 changes of compiler/mil/milprint_summer.c (except the variable expansion changes, that got fixed/extended later on) does make the optimized version of XMark/q04 work as expected, again.
## Comment 6547
Date: 2006-12-14 18:07:56 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
fixed by Jan Rittinger (thx!)
was some problem with the variable usage analysis (afaik)
## Comment 6548
Date: 2006-12-14 18:08:12 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: NO
This bug was only triggered by Peters changes as they removed additional for loops that had hidden this bug. A missing entry in the variable usage BAT made some rows disappear.
It turned out that the bug was introduced by Santa (cvs log -r1.154 mil/milprint_summer.c) almost a year ago :)
Thanks to Stefan and Peter for hunting this one down!
## Comment 6549
Date: 2006-12-14 18:33:12 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
ok, closing it
## Comment 6550
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1611377 at http://sourceforge.net/support/tracker.php?aid=1611377
| XQ: XMark q04 fails only in (all) modularized cases | https://api.github.com/repos/MonetDB/MonetDB/issues/1168/comments | 0 | 2020-11-30T09:24:26Z | 2024-06-27T11:21:47Z | https://github.com/MonetDB/MonetDB/issues/1168 | 753,307,182 | 1,168 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-08 10:38:19 +0100
From: @drstmane
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2006-12-08 06:50:19 +0100
## Comment 6537
Date: 2006-12-08 10:38:19 +0100
From: @drstmane
Before this information get's lost:
On Fri Dec 1 2006, the tests in pathfinder/tests/Update/Tests did work fine (except from a timeout in test pathfinder/tests/Update/Tests/update.xq on the "slower" systems).
On Fri Dec 8 2006, some of the tests fail with
!ERROR: mposjoin: missing matches. The result requires the same number of tuples as the first two input arguments.
!ERROR: CMDmposjoin: operation failed.
!ERROR: interpret_params: mposjoin(param 1): evaluation error.
(Due to unattended compilation problems of Pathfinder and/or MonetDB no pathfinder testing was performed between Fri Dec 1 2006 & Fri Dec 8 2006.)
For details, see
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-3.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-3.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-4.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-5.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/symm-test-5.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/update.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_Update/update.err.00.html
## Comment 6538
Date: 2006-12-08 11:22:52 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
(This *might* be related to
http://sourceforge.net/tracker/index.php?func=detail&aid=1611400&group_id=56967&atid=482468
)
## Comment 6539
Date: 2006-12-08 16:48:30 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Reverting
===================================================================
@@ -1896,7 +1896,7 @@
"var oid_iter := iter;\n"
"var oid_item := item.materialize(ipik);\n"
"var oid_cont := kind.get_container();\n"
- "var temp1 := mvaljoin ( mposjoin(oid_item, oid_cont, ws.fetch(PRE_NID)), oid_cont, ws.fetch(ATTR_OWN));\n"
+ "var temp1 := get_attr_own (ws, mposjoin(oid_item, oid_cont, ws.fetch(PRE_NID)), oid_cont);\n"
"oid_item := nil;\n"
"oid_cont := temp1.hmark(0@0).leftfetchjoin(oid_cont);\n"
"var oid_attr := temp1.tmark(0@0);\n"
===================================================================
of Peter's yesterday's late night changes
Index: compiler/mil/milprint_summer.c
===================================================================
RCS file: /cvsroot/monetdb/pathfinder/compiler/mil/milprint_summer.c,v
retrieving revision 1.298.2.10
retrieving revision 1.298.2.11
diff -u -r1.298.2.10 -r1.298.2.11
--- compiler/mil/milprint_summer.c 7 Dec 2006 17:42:26 -0000 1.298.2.10
+++ compiler/mil/milprint_summer.c 7 Dec 2006 23:07:08 -0000 1.298.2.11
@@ -815,7 +815,7 @@
/* get the attributes of the subtree copy elements */
"{ create attribute subtree copies\n"
- "var temp_attr := mvaljoin( mposjoin(res_item, res_cont, ws.fetch(PRE_NID)), res_cont, ws.fetch(ATTR_OWN));\n"
+ "var temp_attr := get_attr_own(ws, mposjoin(res_item, res_cont, ws.fetch(PRE_NID)), res_cont);\n"
"var oid_attr := temp_attr.tmark(0@0);\n"
"var oid_cont;\n"
"if (is_constant(res_cont)) {\n"
@@ -1126,7 +1126,7 @@
milprintf(f, "var vid := v_vid%03u.ord_uselect(%i@0);\n",
cur_level, f->module_base + v->vid);
milprintf(f, "vid := vid.hmark(0@0);\n");
- milprintf(f, "iter := vid.leftfetchjoin(v_iter%03u);\n", cur_level);
+ milprintf(f, "iter := vid.leftfetchjoin(v_iter%03u).assert_order();\n", cur_level);
milprintf(f, "pos := vid.leftfetchjoin(v_pos%03u);\n", cur_level);
milprintf(f, "item := vid.leftfetchjoin(v_item%03u);\n", cur_level);
milprintf(f, "kind := vid.leftfetchjoin(v_kind%03u);\n", cur_level);
@@ -1896,7 +1896,7 @@
"var oid_iter := iter;\n"
"var oid_item := item.materialize(ipik);\n"
"var oid_cont := kind.get_container();\n"
- "var temp1 := mvaljoin ( mposjoin(oid_item, oid_cont, ws.fetch(PRE_NID)), oid_cont, ws.fetch(ATTR_OWN));\n"
+ "var temp1 := get_attr_own (ws, mposjoin(oid_item, oid_cont, ws.fetch(PRE_NID)), oid_cont);\n"
"oid_item := nil;\n"
"oid_cont := temp1.hmark(0@0).leftfetchjoin(oid_cont);\n"
"var oid_attr := temp1.tmark(0@0);\n"
@@ -2799,7 +2799,7 @@
"var oid_preNew := content_preNew_preOld.hmark(0@0);\n"
"content_preNew_preOld := nil;\n"
"var oid_cont := oid_preNew.leftfetchjoin(preNew_cont);\n"
- "var temp_attr := mvaljoin( mposjoin(oid_preOld, oid_cont, ws.fetch(PRE_NID)), oid_cont, ws.fetch(ATTR_OWN));\n"
+ "var temp_attr := get_attr_own(ws, mposjoin(oid_preOld, oid_cont, ws.fetch(PRE_NID)), oid_cont);\n"
"oid_preOld := nil;\n"
"var oid_attr := temp_attr.tmark(0@0);\n"
"oid_cont := temp_attr.reverse().leftfetchjoin(oid_cont);\n"
===================================================================
seems to fix this bug and http://sourceforge.net/tracker/index.php?func=detail&aid=1611400&group_id=56967&atid=482468 ,
without causing any other (obvious) harm.
I'm not sure, though, whether this is a valid fix, or whether it would simply undo (part of) Peter's performance improvements...
Peter,
Could you please comment?
Thanks!
## Comment 6540
Date: 2006-12-08 18:50:19 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
fixed in CVS:
fixed get_attr_own:
after filtering false hits out of "shared", make sure that the final result
indeed holds the right oids in head and tail, i.e., a subset of the original "shared".
## Comment 6541
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1611373 at http://sourceforge.net/support/tracker.php?aid=1611373
| XQ: update tests fail, again | https://api.github.com/repos/MonetDB/MonetDB/issues/1167/comments | 0 | 2020-11-30T09:24:24Z | 2024-06-27T11:21:46Z | https://github.com/MonetDB/MonetDB/issues/1167 | 753,307,149 | 1,167 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-08 09:52:59 +0100
From: Jan Rittinger <<jan.rittinger>>
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-12-14 05:25:21 +0100
## Comment 6533
Date: 2006-12-08 09:52:59 +0100
From: Jan Rittinger <<jan.rittinger>>
The mil proc add_qnames (in pf_support.mx) fails as the nested proc find_qn_bulk is not able to apply the first fetch in the following MIL snippet:
ws.fetch(QN_LOC_UPDATE).find(cont).insert(_loc);
The proc add_qname that does cope with a single only avoids to call the proc find_qn_bulk and thus works.
(try e.g. echo 'element { "foo" } {}' | pf -A | Mserver to trigger the error)
## Comment 6534
Date: 2006-12-14 17:25:20 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
fixed
## Comment 6535
Date: 2006-12-14 20:16:18 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
added test in
pathfinder/tests/BugTracker/Tests/Attic/add_qnames.SF-1611353.alg.*
seems to work fine.
## Comment 6536
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1611353 at http://sourceforge.net/support/tracker.php?aid=1611353
| XQ: add_qnames fails | https://api.github.com/repos/MonetDB/MonetDB/issues/1166/comments | 0 | 2020-11-30T09:24:21Z | 2024-06-27T11:21:45Z | https://github.com/MonetDB/MonetDB/issues/1166 | 753,307,126 | 1,166 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-08 09:09:17 +0100
From: @drstmane
To: @njnes
Version: -- development
Last updated: 2006-12-09 04:03:35 +0100
## Comment 6529
Date: 2006-12-08 09:09:17 +0100
From: @drstmane
MonetDB checkins on Wed Dec 6 2006 and/or Thu Dec 7 2006 (see below) make almost ALL SQL test segfault (SQL did work fine before Wed Dec 6 2006; there was no SQL testing on Thu Dec 7 2006, as MonetDB compilation failed after in complete changes; there were no related SQL checkins on Wed Dec 6 2006 and/or Thu Dec 7 2006); cf.
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/sql/.mTests103/index_short.html
===================================================================
2006/12/06 - boncz: src/gdk/gdk.mx,1.211.2.2 src/gdk/gdk_atoms.mx,1.125.2.1
src/gdk/gdk_bat.mx,1.166.2.1 src/gdk/gdk_heap.mx,1.84.2.2
src/monet/monet_multiplex.mx,1.59.2.1
PROBABLE M5 DEPENDENCY ALERT
- HEAPextend(heap, newmode, newsize)
signature reverts back to the original
HEAPextend(heap, newsize)
I had introduced the 'newmode' to indate e.g. STORE_MMAP as the
future desired mode of a heap. In case of re-allocation (extend)
that memory type is used for the newly allocated heap. This helps
large data imports (e.g. shreds) change on-the-fly into mmap
memory if things get big. Previously a huge malloced result was
constructed (with a lot of swapping pain), then saved to disk,
then mmapped. The on-the-fly strategy works better, as enlarging
mmap data is O(N) whereas extening mallocs is O(N^2); the reason
being that mmap-extends do not have to physically copy any data.
So far so good. Why the change, again?
A limitation was that this newmode was only passed in BATextend()
for the b->batBuns heap. A tail heap extend is managed implicitly in
strPut(), and did not receive the desired mode information.
(newmode is not part of the strPut parameters). Consequently, these
heaps stayed malloced until the end.
It turns out, that there was a byte left in the Heap struct, where
we can store the hp->newstorage mode. Now, BATmmap() puts the desired
mode there, and any strPut that extends the string heap will notice it
when needed.
Thus, the new solution is more general and caters for string heaps.
- continuing on this line, it was discovered (in XML shredding) that
large imports with mmapped string heaps could lead to performance
problems. It is Linux I/O scheduling stupidity again.
In cases the access pattern is sequential, there is no problem at
all. However, if the access pattern is random, think a double-eliminated
string column using a hash table (not the internal hash table, but an
external one), then after the extend we get random access to unloaded
mmap region. The consequence is <1% top CPU time, where 99% is possible.
The cure is to make a quick sequential pass over the mmapped region.
Thus, we make such string heaps warm using a new HEAPwarm() after an
extend if such a hash table is detected. This is done in
the "property-maintaining" BUN API BUNins/BUNinplace/BUNreplace/BUNappend.
Note there might be additional BAT algebra commands that access a result
bat in a not strictly sequential way, and who bypass this API. Such commands
should consider calling HEAPwarm explictly after a HEAPextend.
===================================================================
2006/12/07 - boncz: src/modules/plain/logger.mx,1.49.2.6
[ 1607207 ] XQ: ID.1204095-a. logger errors
usage of bats that have been deleted, led to errors because while
reading the catalog was not checked.
If in the same session an xml doc was shredded it was also deleted,
this condition would occur.
===================================================================
2006/12/07 - boncz: src/gdk/gdk.mx,1.211.2.3
- small performance fix in HASHloop_str
(check first letter before going into strcmp)
===================================================================
2006/12/07 - stmane: src/modules/plain/upgrade.mx,1.14.2.1
made MonetDB compile, again:
"padding" has been renamed to / replaced by "newstorage" in the BAT header,
hence, we also replace it in all places where the BAT header is used ...
===================================================================
2006/12/07 - stmane:
tests/Availability/Tests/02_Modules_plain.stable.out,1.127.2.3
tests/Availability/Tests/02_Modules_plain.stable.out.STATIC,1.114.2.3
tests/Availability/Tests/02_Modules_plain.stable.out.WindowsNT,1.107.2.3
updated stable output according to Jennie's recent checkin:
"msec" -> "milliseconds"
(... apparently, I'm still needed for these tasks ...)
===================================================================
2006/12/07 - stmane: src/gdk/gdk_relop.mx,1.119.2.3
Fixing BUG #1598310 property checking and leftjoin, (size?):
Avoid calling BATpropcheck without BAT in case the previous operator failed to return a proper BAT.
No test added, as it would require a join that runs out-of memory/address space.
===================================================================
2006/12/07 - stmane: RunMtest.bat.in,1.17.2.1 RunMtest.in,1.31.2.1
NT/Mserver.bat,1.6.6.1 src/common/monet_options.mx,1.7.2.1
src/testing/Mtest.py.in,1.247.2.5 src/testing/README,1.32.2.4
aligned setting & usage of xquery_logdir with that of sql_logdir
===================================================================
## Comment 6530
Date: 2006-12-08 13:10:03 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
Fix is simple (revert logger changes). Problem is that that will return a number of XQ problems. So I'm looking why these fail.
## Comment 6531
Date: 2006-12-09 16:03:35 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Fixed by Niels in CVS.
Thanks!
## Comment 6532
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1611339 at http://sourceforge.net/support/tracker.php?aid=1611339
| MonetDB checkins make (almost) ALL SQL tests segfault | https://api.github.com/repos/MonetDB/MonetDB/issues/1165/comments | 0 | 2020-11-30T09:24:19Z | 2024-06-27T11:21:44Z | https://github.com/MonetDB/MonetDB/issues/1165 | 753,307,090 | 1,165 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-06 05:25:51 +0100
From: Jan Rittinger <<jan.rittinger>>
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-12-14 05:26:02 +0100
## Comment 6525
Date: 2006-12-06 17:25:51 +0100
From: Jan Rittinger <<jan.rittinger>>
If I create elements in XQuery and apply a path step with name test afterwards the result is empty.
The following works:
echo '<a><b/></a>//b' | pf -M | Mserver
echo '<a><b/></a>//*' | pf -A | Mserver
echo '<a><b/></a>/*' | pf -A | Mserver
The following variants however fail:
echo '<a><b/></a>//b' | pf -A | Mserver
echo '<a><b/></a>/b' | pf -A | Mserver
Note that this happens only in the algebra generated code (element construction and path step implementation reside in pf_support.mx).
Perhaps the current version of the candidate list implementation breaks these queries.
## Comment 6526
Date: 2006-12-14 17:26:02 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
fixed
## Comment 6527
Date: 2006-12-14 20:15:41 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
added tests in
pathfinder/tests/BugTracker/Tests/Attic/path_steps_with_name_tests_on_transsient_nodes.SF-1610247.alg.*
seem to work fine.
## Comment 6528
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1610247 at http://sourceforge.net/support/tracker.php?aid=1610247
| XQ: path steps with name tests don't work on transient nodes | https://api.github.com/repos/MonetDB/MonetDB/issues/1164/comments | 0 | 2020-11-30T09:24:16Z | 2024-06-27T11:21:43Z | https://github.com/MonetDB/MonetDB/issues/1164 | 753,307,063 | 1,164 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-06 11:36:51 +0100
From: @romulogoncalves
To: Fabian Groffen <<fabian>>
Version: -- development
Last updated: 2006-12-06 02:11:51 +0100
## Comment 6518
Date: 2006-12-06 11:36:51 +0100
From: @romulogoncalves
Dear developers,
I am using M5 and SQL cvs head. If you execute the following queries:
create table t1(id int, name varchar(1024), age int);
create table t2(id int, age int);
create trigger test_0 after insert on t1
update t1 set id = 1, name = 'monetdb', age = 24 where id =1;
insert into t1 values(1, 'mo', 25);
select * from t1;
select * from t2;
create trigger test_1 after insert on t1
insert into t2 select id,age from t1;
insert into t1 values(2, 'mo', 26);
You will get the following error:
sql>insert into t1 values(2, 'mo', 26);
MAPI = monetdb@localhost:50000
QUERY = insert into t1 values(2, 'mo', 26);
ERROR = !ERROR: MALException:mal.assert:emptySet
sql>
I think this is a problem related with Martin optimizer.
Regards,
Romulo
## Comment 6519
Date: 2006-12-06 14:11:51 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
I think I just solved this. Could you reopen if the problem persists?
## Comment 6520
Date: 2006-12-06 14:14:52 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
OK, I will do it.
Regards,
Romulo
## Comment 6521
Date: 2006-12-06 14:18:58 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
What about adding a test, as usually, to monitor the problem?
## Comment 6522
Date: 2006-12-06 14:26:32 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
The test already exist. If you see the cascaded_triggers test you can see the sql queries that I used to present the bug and in the err output you can see the error message.
I do not report a bug which is not covered by a test. ;)
Regards,
Romulo
## Comment 6523
Date: 2006-12-06 16:04:03 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
I forgot to mention that this is covered by pretty much every SQL test. Testweb was pretty red this morning.
## Comment 6524
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1609984 at http://sourceforge.net/support/tracker.php?aid=1609984
| EmptySet problem.. | https://api.github.com/repos/MonetDB/MonetDB/issues/1163/comments | 0 | 2020-11-30T09:24:14Z | 2024-06-27T11:21:42Z | https://github.com/MonetDB/MonetDB/issues/1163 | 753,307,035 | 1,163 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-02 01:27:19 +0100
From: @peterboncz
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2009-11-27 08:36:12 +0100
## Comment 6504
Date: 2006-12-02 01:27:19 +0100
From: @peterboncz
Philippe Michiels reported at dagstuhl that the simple CHILD test of the MemBeR benchmark crashes when run in MapiClient -lx; crashing during the *compilation* process
It works with pf standalone, translating to MIL explicitly
http://ilps.science.uva.nl/Resources/MemBeR/CHILD.html
The query used in this micro-benchmark navigates downward from the root, performing n successive child navigation steps, where n is to be varied.
For every integer n, the corresponding query, named CHILD(n), is:
/t1/t2/.../tn
THE BUG ONLY OCCURS WITH n>15
AND WITH SERVER SIDE COMPILATION
may well be some buffer overflow..
## Comment 6505
Date: 2006-12-02 08:38:54 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
How does it crash?
Is there any error message or exit status?
Does it work for stand-alone pf also for n >> 15?
Might well be a stack-overflow in some burg generated code...
## Comment 6506
Date: 2006-12-11 16:10:09 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
I am pretty sure, this is a problem with recursive burg-generated code running out of stack space, which in unfortunately out of out stack space control, and hence IMHO very hard to prevent/fix...
With "pf" standalone, it happens as well, just with a longer path expression (29 child steps in my case):
$ n=28 ; i=0 ; q='doc("MyDoc.xml")' ; while [ $i -lt $n ] ; do i=$[i+1] ; q="$q`printf "/t%02d" $i`" ; done ; echo $q | tee /tmp/q.xq ; pf /tmp/q.xq >/dev/null
doc("MyDoc.xml")/t01/t02/t03/t04/t05/t06/t07/t08/t09/t10/t11/t12/t13/t14/t15/t16/t17/t18/t19/t20/t21/t22/t23/t24/t25/t26/t27/t28
$ n=29 ; i=0 ; q='doc("MyDoc.xml")' ; while [ $i -lt $n ] ; do i=$[i+1] ; q="$q`printf "/t%02d" $i`" ; done ; echo $q | tee /tmp/q.xq ; pf /tmp/q.xq >/dev/null
doc("MyDoc.xml")/t01/t02/t03/t04/t05/t06/t07/t08/t09/t10/t11/t12/t13/t14/t15/t16/t17/t18/t19/t20/t21/t22/t23/t24/t25/t26/t27/t28/t29
Segmentation fault
$ gdb --args pf /tmp/q.xq
GNU gdb Red Hat Linux (6.3.0.0-1.84rh)
Copyright 2004 Free Software Foundation, Inc.
GDB is free software, covered by the GNU General Public License, and you are
welcome to change it and/or distribute copies of it under certain conditions.
Type "show copying" to see the conditions.
There is absolutely no warranty for GDB. Type "show warranty" for details.
This GDB was configured as "x86_64-redhat-linux-gnu"...Using host libthread_db library "/lib64/libthread_db.so.1".
(gdb) r
Starting program: /net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Stable/.GNU.64.32.d.--disable-optimize_--enable-debug_--enable-assert.PREFIX./PATHFINDER/bin/pf /tmp/q.xq
[Thread debugging using libthread_db enabled]
[New Thread 46912496328800 (LWP 30516)]
Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 46912496328800 (LWP 30516)]
0x0000000000483666 in reduce (p=0x2aaaaaae0b90, goalnt=50) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/core/fs.brg:874
874 rule = PFfs_rule (STATE_LABEL (p), goalnt);
(gdb) li
869 PFpnode_t *kids[MAX_KIDS]; /* leaf nodes of this rule */
870 bool topdown; /* is this a top-down rule? */
871 PFcnode_t *c; /* temporary helper variable */
872
873 /* determine rule that matches for this non-terminal */
874 rule = PFfs_rule (STATE_LABEL (p), goalnt);
875
876 /* PFinfo (OOPS_NOTICE, "match for rule %i", rule); */
877 assert (rule);
878
(gdb) up
1 0x00000000004837d9 in reduce (p=0x2aaaaaae0b90, goalnt=49) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/core/fs.brg:929
929 reduce (kids[i], nts[i]);
(gdb) li
924 /*
925 * Recursively invoke compilation. This means bottom-up compilation.
926 */
927 if (!topdown)
928 for (unsigned short i = 0; nts[i]; i++)
929 reduce (kids[i], nts[i]);
930
931 /*
932 PFinfo (OOPS_NOTICE, "processing rule %i", rule);
933 */
(gdb) bt
0 0x0000000000483666 in reduce (p=0x2aaaaaae0b90, goalnt=50) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/core/fs.brg:874
1 0x00000000004837d9 in reduce (p=0x2aaaaaae0b90, goalnt=49) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/core/fs.brg:929
[...]
680 0x00000000004837d9 in reduce (p=0x2aaaaaae5f00, goalnt=1) at 680 0x00000000004837d9 in reduce (p=0x2aaaaaae5f00, goalnt=1) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/core/fs.brg:929
681 0x0000000000491eaf in PFfs (r=0x2aaaaaae5f00) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/core/fs.brg:4209
682 0x00000000004065da in PFcompile (url=0x7ffffffb672e "/tmp/q.xq", pfout=0x38ecf318c0, status=0x731f40) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/compile.c:381
683 0x0000000000405f11 in main (argc=2, argv=0x7ffffffb4e78) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/main.c:905
(gdb)
/ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/core/fs.brg:929
681 0x0000000000491eaf in PFfs (r=0x2aaaaaae5f00) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/core/fs.brg:4209
682 0x00000000004065da in PFcompile (url=0x7ffffffb672e "/tmp/q.xq", pfout=0x38ecf318c0, status=0x731f40) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/compile.c:381
683 0x0000000000405f11 in main (argc=2, argv=0x7ffffffb4e78) at /ufs/manegold/_/scratch0/Monet/Testing/Stable/pathfinder/compiler/main.c:905
(gdb)
## Comment 6507
Date: 2007-05-25 23:55:37 +0200
From: @peterboncz
Logged In: YES
user_id=591107
Originator: YES
fixed using a fence
## Comment 6508
Date: 2007-05-28 12:33:18 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Added test in pathfinder/tests/BugTracker/Tests/server-side_compilation_crash.SF-1607210*
As opposed to
[ 1207048 ] XQuery: crashtest
http://sourceforge.net/tracker/index.php?func=detail&aid=1207048&group_id=56967&atid=482468
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora6/tests_BugDay_2005-11-09_0.9.3/xquery_crashtest.SF-1207048.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora6/tests_BugDay_2005-11-09_0.9.3/xquery_crashtest.SF-1207048.err.00.html
it still crashes with a segfault, i.e.,
the new fence does not catch the stack overflow in this case, yet.
Not analysed, yet.
Re-opend to keep us reminded to have another look at this one.
## Comment 6509
Date: 2007-05-29 10:24:48 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
See also
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora6/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Fedora6/tests_BugTracker/server-side_compilation_crash.SF-1607210.out.00.html
## Comment 6510
Date: 2007-06-04 21:19:07 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
See also
[ 1730547 ] XQ: Mserver crashes on concatenated query
http://sourceforge.net/tracker/index.php?func=detail&aid=1730547&group_id=56967&atid=482468
## Comment 6511
Date: 2007-06-05 18:37:18 +0200
From: @peterboncz
Logged In: YES
user_id=591107
Originator: YES
fixed
## Comment 6512
Date: 2008-01-22 14:58:02 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
re-opened as it fails, again, since 2008.01.12, i.e., after checkins on 2008.01.11, now with
ERROR = !fatal error: aborted too deep recursion
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.32.32.d-Fedora6/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
## Comment 6513
Date: 2008-02-10 14:12:12 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Appears to work fine on most platforms, again;
exceptions:
Segfault on RedHat (on Itanium):
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/Int.64.64.d-RedHat4WS/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
Timeout on Windows:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/Int.64.32.d-Windows5.2/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
## Comment 6514
Date: 2009-11-16 22:18:37 +0100
From: @drstmane
re-opened as the respective test fails on all platform with timeout or segfault at least since 2009.03.25
cf.,
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d.1-Fedora10/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d.1-Fedora10/tests_BugTracker/server-side_compilation_crash.SF-1607210.out.00.html
## Comment 6515
Date: 2009-11-21 15:35:02 +0100
From: @drstmane
===================================================================
2009/11/21 - stmane: pathfinder/runtime/pathfinder.mx,1.462.4.12
use more conservative stack limit (PFmaxstack)
for too deep recusion check (PFrecursion_fence())
to prevent segfault (due to too deep recursion and
consequent stack overflow) in tests
tests/BugTracker/Tests/collection_management.SF-1726599.*
tests/BugTracker/Tests/crash_on_concatenated_query.SF-1730547.*
cf.,
https://sourceforge.net/tracker/?func=detail&aid=1607210&group_id=56967&atid=482468
https://sourceforge.net/tracker/?func=detail&aid=1730547&group_id=56967&atid=482468
===================================================================
Report can be closed once nightly testing confirms the fix for all platforms.
## Comment 6516
Date: 2009-11-27 08:36:11 +0100
From: @drstmane
After some more fixes (see below), the test now appears to work fine on all platforms except
- timeout on
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/Int.64.32.d.1-Fedora10/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/Int.64.64.d.1-Fedora10/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/Int.64.32.d.1-Windows6.0/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/Int.64.64.d.1-Windows6.0/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
This needs to be analyzed separately.
- abort too deep recursion on
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.32.32.d.1-SunOS5.11/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
This occurs with numerous tests on that platform and needs to be analyzed separately.
Closing.
===================================================================
2009/11/23 - stmane: pathfinder/runtime/pathfinder.mx,1.462.4.13
try even more conservative stack limit (PFmaxstack)
for too deep recusion check (PFrecursion_fence())
to check whether this helps to prevent the remaining segfaults with test
tests/BugTracker/Tests/server-side_compilation_crash.SF-1607210.*
on some platforms; cf.
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.32.32.d.1-Debian4.0/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d.1-Gentoo2.0.1/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.32.32.d.1-SunOS5.11/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
===================================================================
2009/11/24 - stmane:
pathfinder/tests/BugTracker/Tests/server-side_compilation_crash.SF-1607210_s00.xq,1.4.22.1
pathfinder/tests/BugTracker/Tests/server-side_compilation_crash.SF-1607210_s02.xq,1.4.22.1
after
"
try even more conservative stack limit (PFmaxstack)
for too deep recusion check (PFrecursion_fence())
to check whether this helps to prevent the remaining segfaults with test
tests/BugTracker/Tests/server-side_compilation_crash.SF-1607210.*
on some platforms; cf.
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.32.32.d.1-Debian4.0/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d.1-Gentoo2.0.1/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.32.32.d.1-SunOS5.11/tests_BugTracker/server-side_compilation_crash.SF-1607210.err.00.html
"
this test does not segfault any more,
but runs into "too deep recursion" on the mentioned platforms
earlier than on others; hence, we shrink parts of the test
a bit.
===================================================================
2009/11/26 - stmane:
pathfinder/tests/BugTracker/Tests/server-side_compilation_crash.SF-1607210_s00.xq,1.4.22.2
pathfinder/tests/BugTracker/Tests/server-side_compilation_crash.SF-1607210_s02.xq,1.4.22.2
shrink queries even more
to avoid "aborted too deep recursion"
also on 32-bit Debian4.0 & 32-bit SunOS5.11
===================================================================
## Comment 6517
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1607210 at http://sourceforge.net/support/tracker.php?aid=1607210
| XQ: server-side compilation crash (member benchmark) | https://api.github.com/repos/MonetDB/MonetDB/issues/1162/comments | 0 | 2020-11-30T09:24:10Z | 2024-06-27T11:21:41Z | https://github.com/MonetDB/MonetDB/issues/1162 | 753,306,994 | 1,162 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-12-02 01:20:33 +0100
From: @peterboncz
To: @njnes
Version: -- development
Last updated: 2006-12-09 04:23:30 +0100
## Comment 6500
Date: 2006-12-02 01:20:33 +0100
From: @peterboncz
It happens at startup, so the problem may well be caused by the previous script:
ID.1204965-d_run
errors are:
!ERROR: logger could not use bat (142) for 1000000002_map_pid
!WARNING: BATdescriptor: range error 71
!ERROR: logger could not use bat (71) for 1000000002_rid_size
!WARNING: BATdescriptor: range error 145
!ERROR: logger could not use bat (145) for 1000000002_rid_level
!WARNING: BATdescriptor: range error 70
!ERROR: logger could not use bat (70) for 1000000002_rid_prop
!WARNING: BATdescriptor: range error 146
!ERROR: logger could not use bat (146) for 1000000002_rid_kind
!WARNING: BATdescriptor: range error 147
!ERROR: logger could not use bat (147) for 1000000002_rid_nid
!WARNING: BATdescriptor: range error 148
!ERROR: logger could not use bat (148) for 1000000002_nid_rid
!WARNING: BATdescriptor: range error 162
!ERROR: logger could not use bat (162) for 1000000002_attr_own
!WARNING: BATdescriptor: range error 163
!ERROR: logger could not use bat (163) for 1000000002_attr_qn
!WARNING: BATdescriptor: range error 164
!ERROR: logger could not use bat (164) for 1000000002_attr_prop
!WARNING: BATdescriptor: range error 160
!ERROR: logger could not use bat (160) for 1000000002_prop_val
!WARNING: BATdescriptor: range error 156
!ERROR: logger could not use bat (156) for 1000000002_prop_text
!WARNING: BATdescriptor: range error 157
!ERROR: logger could not use bat (157) for 1000000002_prop_com
!WARNING: BATdescriptor: range error 158
!ERROR: logger could not use bat (158) for 1000000002_prop_ins
!WARNING: BATdescriptor: range error 159
!ERROR: logger could not use bat (159) for 1000000002_prop_tgt
!WARNING: BATdescriptor: range error 155
!ERROR: logger could not use bat (155) for 1000000002_qn_loc
!WARNING: BATdescriptor: range error 154
!ERROR: logger could not use bat (154) for 1000000002_qn_uri
!WARNING: BATdescriptor: range error 153
!ERROR: logger could not use bat (153) for 1000000002_qn_prefix
!WARNING: BATdescriptor: range error 152
!ERROR: logger could not use bat (152) for 1000000002_qn_uri_loc
!WARNING: BATdescriptor: range error 151
!ERROR: logger could not use bat (151) for 1000000002_qn_prefix_uri_loc
!WARNING: BATdescriptor: range error 149
!ERROR: logger could not use bat (149) for 1000000002_qn_histogram
## Comment 6501
Date: 2006-12-09 03:26:53 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Test seems to work fine both in the XQuery_0-14 release branch and the development trunk:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugsViaSourgeforce/ID.1204965-d_shred.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_BugsViaSourgeforce/ID.1204965-d_shred.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_BugsViaSourgeforce/ID.1204965-d_run.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_BugsViaSourgeforce/ID.1204965-d_run.err.00.html
Hence, could/should this one be closed?
## Comment 6502
Date: 2006-12-09 16:23:30 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Fixed by Niels in CVS.
Thanks!
## Comment 6503
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1607207 at http://sourceforge.net/support/tracker.php?aid=1607207
| XQ: ID.1204095-a. logger errors | https://api.github.com/repos/MonetDB/MonetDB/issues/1161/comments | 0 | 2020-11-30T09:24:06Z | 2024-06-27T11:21:41Z | https://github.com/MonetDB/MonetDB/issues/1161 | 753,306,944 | 1,161 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-29 01:17:49 +0100
From: @romulogoncalves
To: @njnes
Version: -- development
Last updated: 2007-01-17 08:20:56 +0100
## Comment 6497
Date: 2006-11-29 13:17:49 +0100
From: @romulogoncalves
Dear developers,
I am using SQL cvs head and M5.
If I run the following queires:
create table t1 (id int, name varchar(1024));
create table t2 (id int);
insert into t1 values(10, 'monetdb');
insert into t1 values(20, 'monet');
create trigger test1
after update on t1
delete from t1 where name = 'mo' and id = 10;
create trigger test2
before update on t1
select id into t2 where id = 10;
create trigger test3
after update on t1
insert into t1 values(10, 'update_from_test3');
create trigger test4
after update on t1
insert into t1 values(20, 'update_from_test4');
ALTER TABLE t1 add CONSTRAINT t1_constraint PRIMARY KEY(id);
I get this error:
sql>ALTER TABLE t1 add CONSTRAINT t1_constraint PRIMARY KEY(id);
MAPI = monetdb@localhost:50000
QUERY = ALTER TABLE t1 add CONSTRAINT t1_constraint PRIMARY KEY(id);
ERROR = !ERROR: 'aggr.count' undefined in user.s1_1[15] _35:bat[:oid,:int] := aggr.count(_31:bat[:void,:oid], _31:bat[:void,:oid], _33:bat[:void,:void])
!ERROR: SQLerror:assert:INSERT INTO: PRIMARY KEY constraint 't1.t1_constraint' violated
sql>
Any idea?
Regards,
Romulo
## Comment 6498
Date: 2007-01-17 20:20:56 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
This BUG was fixed and it covered by a Test.
Note: One of the triggers in the example is not working. I run the example without this trigger.
Regards,
Romulo
## Comment 6499
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1605209 at http://sourceforge.net/support/tracker.php?aid=1605209
| Triggers combine with constraints | https://api.github.com/repos/MonetDB/MonetDB/issues/1160/comments | 0 | 2020-11-30T09:24:03Z | 2024-06-28T07:36:51Z | https://github.com/MonetDB/MonetDB/issues/1160 | 753,306,891 | 1,160 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-29 01:11:58 +0100
From: @romulogoncalves
To: @njnes
Version: -- development
Last updated: 2007-01-17 08:10:21 +0100
## Comment 6494
Date: 2006-11-29 13:11:58 +0100
From: @romulogoncalves
Dear developers,
I am using SQL cvs head and M5.
If I run the following quereies the Mserver crash:
create table t1 (id int, name varchar(1024));
insert into t1 values(10, 'monetdb');
insert into t1 values(20, 'monet');
create trigger test1
after update on t1 referencing old row as old_row
for each row insert into t1 values(0, 'update_old_row');
Regards,
Romulo
## Comment 6495
Date: 2007-01-17 20:10:21 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
The bug was resolved and it is covered by SQL test.
Regards,
Romulo
## Comment 6496
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1605205 at http://sourceforge.net/support/tracker.php?aid=1605205
| Triggers - old row | https://api.github.com/repos/MonetDB/MonetDB/issues/1159/comments | 0 | 2020-11-30T09:23:59Z | 2024-06-28T07:36:50Z | https://github.com/MonetDB/MonetDB/issues/1159 | 753,306,848 | 1,159 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-29 01:05:58 +0100
From: @romulogoncalves
To: @njnes
Version: -- development
Last updated: 2007-01-17 10:20:50 +0100
## Comment 6491
Date: 2006-11-29 13:05:58 +0100
From: @romulogoncalves
Dear developers,
I am using SQL cvs head and M5.
To see me problem run the following queries:
create table t1(id int, name varchar(1024), age int);
create table t2(id int, age int);
create trigger test_0 after insert on t1
update t1 set id = 1, name = 'monetdb', age = 24 where id =1;
insert into t1 values(1, 'mo', 25);
select * from t1;
select * from t2;
create trigger test_1 after insert on t1
insert into t2 select id,age from t1;
insert into t1 values(2, 'mo', 26);
select * from t1;
select * from t2;
create trigger test_2 after insert on t1
insert into t2 values(1,23);
insert into t1 values(3, 'mo', 27);
select * from t1;
select * from t2;
If you check the result of the selection over t2 it does not contains the row:
[1, 23 ]
it only contains:
[ 1, 25 ]
[ 1, 24 ]
[ 2, 26 ]
Which should had been inserted by action body of trigger test_2.
Regards,
Romulo
## Comment 6492
Date: 2007-01-17 22:20:50 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
This BUG was correct and it is coverd by a SQL test.
Regards,
Romulo
## Comment 6493
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1605200 at http://sourceforge.net/support/tracker.php?aid=1605200
| Triggers- Insert action | https://api.github.com/repos/MonetDB/MonetDB/issues/1158/comments | 0 | 2020-11-30T09:23:56Z | 2024-06-28T07:36:50Z | https://github.com/MonetDB/MonetDB/issues/1158 | 753,306,815 | 1,158 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-29 12:54:32 +0100
From: @romulogoncalves
To: @njnes
Version: -- development
Last updated: 2007-01-17 08:16:15 +0100
## Comment 6488
Date: 2006-11-29 12:54:32 +0100
From: @romulogoncalves
Dear developers,
Just run the following queries:
create table t1(id int, name varchar(1024), age int);
create trigger test_0 after insert on t1
update t1 set id = 1, name = 'monetdb', age = 24 where id =1;
insert into t1 values(1, 'mo', 25);
select * from t1;
How you can see the result in t1 is:
[ 1, "mo", 25 ]
and it should be:
[ 1, "monetdb", 24 ]
Regards,
Romulo
## Comment 6489
Date: 2007-01-17 20:16:15 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
This BUG was fixed and it is covered by a SQL test.
Regards,
Romulo
## Comment 6490
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1605193 at http://sourceforge.net/support/tracker.php?aid=1605193
| Trigger - Update action | https://api.github.com/repos/MonetDB/MonetDB/issues/1157/comments | 0 | 2020-11-30T09:23:53Z | 2024-06-28T13:31:20Z | https://github.com/MonetDB/MonetDB/issues/1157 | 753,306,766 | 1,157 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-27 12:05:35 +0100
From: @BitGrit
To: GDK devs <<bugs-common>>
Version: -- development
Last updated: 2007-02-19 03:06:23 +0100
## Comment 6483
Date: 2006-11-27 12:05:35 +0100
From: @BitGrit
Under certain conditions, procedures defined in a running Mserver, are not visible from a MapiClient connecting to that server:
Start Mserver and run:
MonetDB>module(mapi);
MonetDB>PROC test() : void { print(1); }
MonetDB>mil_start();
When connecting with MapiClient and running above procedure:
mil>test();
MAPI = monetdb@localhost:50000
QUERY = test();
ERROR = !ERROR: interpret: no matching MIL operator to 'test()'.
However, starting the Mserver with either of the following dbinit strings does seem to work:
Mserver --dbinit="module(mapi);mil_start(); PROC test() : void { print(1); }"
AND:
Mserver --dbinit="module(mapi); PROC test() : void { print(1); }; mil_start();"
result in MapiClient:
mil>test();
[ 1 ]
mil>
## Comment 6484
Date: 2006-11-27 14:26:22 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
what's the behaviour if you define the PROC in a script that's given as a command line option to Mserver, i.e.:
cat <<EOF >x.mil
module(mapi);
PROC test() : void { print(1); }
mil_start();
EOF
Mserver x.mil
?
## Comment 6485
Date: 2006-11-27 18:24:54 +0100
From: @BitGrit
Logged In: YES
user_id=1292024
Originator: YES
The suggusted approach does work, e.g.:
what's the behaviour if you define the PROC in a script that's given as a
command line option to Mserver, i.e.:
cat <<EOF >x.mil
module(mapi);
PROC test() : void { print(1); }
mil_start();
EOF
Mserver x.mil
Does produce the desired output when calling test() over MapiClient.
However, the below approach ('sourcing' the PROC from x.mil from an external file
t.mil) used to work in the past as well, but does not work anymore:
cat <<EOF >x.mil
module(mapi);
var p := environment().find("monet_cwd");
var t := p + "/t.mil";
source(t);
mil_start();
EOF
cat <<EOF >t.mil
PROC test() : void { print(1); }
EOF
Mserver x.mil
calling test from MapiClient now again results in:
mil>test();
MAPI = monetdb@localhost:50000
QUERY = test();
ERROR = !ERROR: interpret: no matching MIL operator to 'test()'.
Hope this makes things more clear.
Greetings
Sandor
## Comment 6486
Date: 2007-02-19 15:06:23 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Use the workaround. It's unlikely that this will get fixed.
## Comment 6487
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1603655 at http://sourceforge.net/support/tracker.php?aid=1603655
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-common@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| Inconsistency in Mapi Client context locality | https://api.github.com/repos/MonetDB/MonetDB/issues/1156/comments | 0 | 2020-11-30T09:23:50Z | 2024-06-28T13:31:19Z | https://github.com/MonetDB/MonetDB/issues/1156 | 753,306,736 | 1,156 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-26 12:54:09 +0100
From: @romulogoncalves
To: @njnes
Version: -- development
Last updated: 2007-01-17 08:05:41 +0100
## Comment 6480
Date: 2006-11-26 12:54:09 +0100
From: @romulogoncalves
Dear developers,
If you run the following queries the Mserver crash.
create table t1 (id int, name varchar(1024));
insert into t1 values(1, 'monetdb');
create trigger test1
after update on t1
create table t2 (id int);
update t1 set name = 'mo' where id = 1;
I am using SQL cvs head and M5.
Regards,
Romulo Goncalves
## Comment 6481
Date: 2007-01-17 20:05:41 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
This bug was resolved and it is covered by a SQL test.
Regards,
Romulo
## Comment 6482
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1603089 at http://sourceforge.net/support/tracker.php?aid=1603089
| Triggers action - create table | https://api.github.com/repos/MonetDB/MonetDB/issues/1155/comments | 0 | 2020-11-30T09:23:47Z | 2024-06-28T13:31:18Z | https://github.com/MonetDB/MonetDB/issues/1155 | 753,306,704 | 1,155 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-26 12:48:42 +0100
From: @romulogoncalves
To: @njnes
Version: -- development
Last updated: 2006-12-04 10:55:41 +0100
## Comment 6476
Date: 2006-11-26 12:48:42 +0100
From: @romulogoncalves
Dear developers,
If you run the the following queries the Mserver will crash.
create table t1 (id int, name varchar(1024));
insert into t1 values(10, 'monetdb');
insert into t1 values(20, 'monet');
create trigger test1
after update on t1 referencing new row as new_row
for each row insert into t1 values(0, 'update_new_row');
I am using the SQL cvs head and M5.
Regards,
Romulo Goncalves
## Comment 6477
Date: 2006-12-04 10:55:41 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
The parser was fixed to properly handle the new row as part. A test exists for this in the trigger test directory.
## Comment 6478
Date: 2006-12-04 10:58:15 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
Well, the parser was fixed. However the semantics is still not working.
If you run Mtest in the Triggers directory you will see the problem.
Regards,
Romulo
## Comment 6479
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1603087 at http://sourceforge.net/support/tracker.php?aid=1603087
| Triggers new row | https://api.github.com/repos/MonetDB/MonetDB/issues/1154/comments | 0 | 2020-11-30T09:23:44Z | 2024-06-27T11:21:33Z | https://github.com/MonetDB/MonetDB/issues/1154 | 753,306,668 | 1,154 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-23 02:39:14 +0100
From: @drstmane
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-12-02 12:54:56 +0100
## Comment 6472
Date: 2006-11-23 14:39:14 +0100
From: @drstmane
My recent checkin of Peter's Dagstuhl-shredded-bug-fix
(
pathfinder/runtime/shredder.mx, revision 1.106.2.1
"
Peter's Dagstuhl bug-fixes
for the Pathfinder shredder
[I hope my interpretation/comment is correct...]
shredBAT_extend needs to set the current size of shredBAT(s)
before extending them
"
)
seems to break the shredding of several benchmark documents.
Errors are as follows
benchmarks/XMach-1/:
benchmarks/XBench/TC/MD/:
!ERROR: [shred_url]: 100 times inserted nil due to errors at tuples 1@0, 2@0, 3@0, 4@0, 5@0, 6@0, 7@0, 8@0, 9@0, 10@0, 11@0, 12@0, 13@0, 14@0, 15@0.
!ERROR: [shred_url]: first error was:
!ERROR: shredder_create: _qn_prefix_uri_loc (batid 102) is not aligned with _qn_histogram (batid 101).
benchmarks/X007/:
benchmarks/XBench/DC/MD/:
!ERROR: [shred_url]: 1 times inserted nil due to errors at tuples 1@0.
!ERROR: [shred_url]: first error was:
!ERROR: shredder_create: _rid_level (batid 193) is not aligned with _rid_size (batid 190).
[... sorry, I should have tested BEFORE checking in ...]
## Comment 6473
Date: 2006-11-23 14:45:23 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
similar error now also occur with the W3C XQUF use case tests in
tests/W3C_use_cases/XQUF/Parts/
## Comment 6474
Date: 2006-12-02 00:54:56 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
this was fixed in the shredder
## Comment 6475
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1601761 at http://sourceforge.net/support/tracker.php?aid=1601761
| XQ: Dagstuhl shredder bug-fix not correct | https://api.github.com/repos/MonetDB/MonetDB/issues/1153/comments | 0 | 2020-11-30T09:23:41Z | 2024-06-27T11:21:32Z | https://github.com/MonetDB/MonetDB/issues/1153 | 753,306,628 | 1,153 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-23 12:12:28 +0100
From: Wouter Alink <<alink>>
To: @njnes
Version: -- development
Last updated: 2006-12-04 10:53:47 +0100
## Comment 6465
Date: 2006-11-23 00:12:28 +0100
From: Wouter Alink <<alink>>
I was able to create a new user, but could not log in as that user (ERROR: schema authorization error). See output below.
I don't know what I am doing wrong. Fabian reported that he did not have any problems doing exactly the same thing.
the Mserver is running as 'root'. logs and databases are stored under /var/MonetDB_SQL/. all other monetdb.conf options are equal to the default.
My system: Fedora Core 6, compiled from yesterday's CVS SQL 2.14.0, MonetDB 4.14.0 (32-bits).
[xiraf@client lib]$ java -jar jdbcclient-1.5.jar -umonetdb
password:
Welcome to the MonetDB interactive JDBC terminal!
Database: MonetDB 4.14.0
Driver: MonetDB Native Driver 1.5 (Steadfast_pre3 20061121)
Type \q to quit, \h for a list of available commands
auto commit mode: on
monetdb-> CREATE USER "nsrl" WITH PASSWORD 'nsrl' NAME 'NSRL Tool' SCHEMA "sys";
CREATE SCHEMA "nsrl" AUTHORIZATION "nsrl";
ALTER USER "nsrl" SET SCHEMA "nsrl";
Operation successful
monetdb-> Operation successful
monetdb-> Operation successful
monetdb-> \q
[xiraf@client lib]$ java -jar jdbcclient-1.5.jar -unsrl
password:
Database connect failed: ERROR: schema authorization error
[xiraf@client lib]$ java -jar jdbcclient-1.5.jar -umonetdb
monetdb-> SELECT * FROM db_users;
+---------+---------+---------------+----------------+
| name | passwd | fullname | default_schema |
+=========+=========+===============+================+
| monetdb | monetdb | MonetDB Admin | 4557 |
| nsrl | nsrl | NSRL Tool | 5984 |
+---------+---------+---------------+----------------+
2 rows
monetdb->
## Comment 6466
Date: 2006-11-23 00:17:28 +0100
From: Wouter Alink <<alink>>
Logged In: YES
user_id=621590
Originator: YES
perhaps useful information (or another bug??):
[xiraf@client ~]$ java -jar /usr/local/share/MonetDB/lib/jdbcclient-1.5.jar -dnsrl -unsrl -Xdebug
password:
Exception in thread "main" java.lang.StringIndexOutOfBoundsException
at java.lang.String.substring(libgcj.so.7rh)
at nl.cwi.monetdb.jdbc.MonetSocketBlockMode.read(MonetSocketBlockMode.java:283)
at nl.cwi.monetdb.jdbc.MonetConnection.<init>(MonetConnection.java:248)
at nl.cwi.monetdb.jdbc.MonetDriver.connect(MonetDriver.java:160)
at java.sql.DriverManager.getConnection(libgcj.so.7rh)
at java.sql.DriverManager.getConnection(libgcj.so.7rh)
at nl.cwi.monetdb.client.JdbcClient.main(JdbcClient.java:257)
[xiraf@client ~]$ cat monet_1164237279983.log
RX 1164237279984: 12
RX 1164237279984: ::mserver:6
TX 1164237279984: nsrl:nsrl:sql:blocked:nsrl
TX 1164237279984:
## Comment 6467
Date: 2006-11-23 00:21:39 +0100
From: Wouter Alink <<alink>>
Logged In: YES
user_id=621590
Originator: YES
The crash in the previous comment is caused by the gcj... running it with sun's java gives:
[xiraf@client ~]$ cat monet_1164237456663.log
RX 1164237456675: 12
RX 1164237456675: ::mserver:6
TX 1164237456675: nsrl:nsrl:sql:blocked:nsrl
TX 1164237456676: �
RX 1164237456676: �
RD 1164237456677: final block: 35 bytes
RD 1164237456677: read chunk: 35 bytes, left: 0 bytes
RX 1164237456677: !ERROR: schema authorization error
RD 1164237456677: generating prompt
## Comment 6468
Date: 2006-11-27 15:41:04 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
I tried with gcj too, but can't reproduce your problem there either:
(pegasus:cwi/monetdb/jdbc) fabian% gcj --version
gcj (GCC) 4.1.1 (Gentoo 4.1.1-r2)
Copyright (C) 2006 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
(pegasus:cwi/monetdb/jdbc) fabian% ./mjclient -unsrl
password:
Welcome to the MonetDB interactive JDBC terminal!
Database: MonetDB 4.14.0
Driver: MonetDB Native Driver 1.5 (Steadfast_pre4 20061127)
Type \q to quit, \h for a list of available commands
auto commit mode: on
nsrl-> \d
nsrl-> select 'hello';
+--------------+
| single_value |
+==============+
| hello |
+--------------+
1 row
nsrl->
## Comment 6469
Date: 2006-11-27 15:42:51 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
mjclient is a binary made using:
% gcj -o mjclient jdbcclient-1.5.jar --main=nl.cwi.monetdb.client.JdbcClient
## Comment 6470
Date: 2006-12-04 10:53:47 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
As the problem isn't reproducable with current gcj and doesn't exist with the java
sdk >= 1.4 we close this bug.
Also login with different users is tested using the SkyServer and VOC tests.
## Comment 6471
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1601455 at http://sourceforge.net/support/tracker.php?aid=1601455
| SQL: cannot login using different user | https://api.github.com/repos/MonetDB/MonetDB/issues/1152/comments | 0 | 2020-11-30T09:23:37Z | 2024-06-28T13:31:18Z | https://github.com/MonetDB/MonetDB/issues/1152 | 753,306,580 | 1,152 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-21 09:31:32 +0100
From: Wouter Alink <<alink>>
To: XQuery devs <<bugs-xquery>>
Version: -- development
Last updated: 2006-12-02 01:10:40 +0100
## Comment 6463
Date: 2006-11-21 21:31:32 +0100
From: Wouter Alink <<alink>>
This bug is in 0.12 as well as in 0.14
when writing a more complex query (don't know exactly how complex it needs to be), the subsequence-function returns either no results at all, or all nodes without attributes. In 0.12 a internal BAT-alignment error is generated (don't have the error-message at hand, but something with BAT 2 and BAT 4 ;).
I shall add the queries to the testweb, but for clarity I provide an example here:
let $d := doc("test.xml")
let $nodes := for $i in $d//noot
where $i/@id
return $i
return
element { "result" } {
element { "count" } { count($nodes) },
for $i in subsequence($nodes, 11, 20)
return element { name($i) } { $i/@* }
}
which outputs:
<?xml version="1.0" encoding="utf-8"?>
<XQueryResult><result><count>55</count></result></XQueryResult>
which is obviously a wrong result (independant of the document.
Note that the same query, but with the subsequence offset parameter "11" replaced by "1" yields correct results.
## Comment 6464
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1600704 at http://sourceforge.net/support/tracker.php?aid=1600704
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-xquery@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| XQuery: subsequence > 1 fails | https://api.github.com/repos/MonetDB/MonetDB/issues/1151/comments | 0 | 2020-11-30T09:23:34Z | 2024-06-27T11:21:30Z | https://github.com/MonetDB/MonetDB/issues/1151 | 753,306,550 | 1,151 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-21 01:59:50 +0100
From: Wouter Alink <<alink>>
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2006-12-12 10:20:25 +0100
## Comment 6458
Date: 2006-11-21 13:59:50 +0100
From: Wouter Alink <<alink>>
(i'm using MonetDB 4.14, yesterday's CVS)
should 'make check' use the "{installdir}/var" for testing?
Traceback (most recent call last):
File "/home/alink/Desktop/monetdb-0.14-work/MonetDB/Linux/src/testing/Mtest.py", line 3361, in ?
main(sys.argv)
File "/home/alink/Desktop/monetdb-0.14-work/MonetDB/Linux/src/testing/Mtest.py", line 3115, in main
os.makedirs(env['XQUERY_LOGDIR'])
File "/usr/lib64/python2.4/os.py", line 156, in makedirs
makedirs(head, mode)
File "/usr/lib64/python2.4/os.py", line 156, in makedirs
makedirs(head, mode)
File "/usr/lib64/python2.4/os.py", line 159, in makedirs
mkdir(name, mode)
OSError: [Errno 13] Permission denied: '/usr/local/var'
## Comment 6459
Date: 2006-11-21 14:04:39 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
I guess your prefix is /usr/local (autoconf default) so it looks ok to me...
## Comment 6460
Date: 2006-11-21 20:07:35 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
The initial question of this bug report is:
should 'make check' use the "{installdir}/var" for testing?
Which I would answer with NO. Make check should run before installation and preferably without any
special privileges. So using an installdir which could require those special privileges isn't an option.
Probably we should use a directory under the builddir.
(lowered the priority as its less important to fix this than real MonetDB product bugs)
## Comment 6461
Date: 2006-12-12 10:20:25 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Stefan added the --xquery_logdir option to RunMtest which should solve this problem.
## Comment 6462
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1600392 at http://sourceforge.net/support/tracker.php?aid=1600392
| MonetDB: should 'make check' use installdir? | https://api.github.com/repos/MonetDB/MonetDB/issues/1150/comments | 0 | 2020-11-30T09:23:32Z | 2024-06-27T11:21:29Z | https://github.com/MonetDB/MonetDB/issues/1150 | 753,306,505 | 1,150 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-20 03:46:57 +0100
From: @romulogoncalves
To: @njnes
Version: -- development
Last updated: 2007-01-17 08:12:50 +0100
## Comment 6455
Date: 2006-11-20 15:46:57 +0100
From: @romulogoncalves
Dear developers,
I created the following tables and trigger:
create table t1(id int, name varchar(1024), age int);
create table t2(id int, age int);
create trigger test_1 after insert on t1
insert into t2 select id,age from t1;
I done the following insertion on t1:
sql>insert into t1 values(2, 'mo', 26);
[ 0 ]
[ 1 ]
And then I checked the status of each table:
sql>select * from t1;
% sys.t1, sys.t1, sys.t1 table_name
% id, name, age name
% int, varchar, int type
% 1, 2, 2 length
[ 2, "mo", 26 ]
sql>select * from t2;
% sys.t2, sys.t2 table_name
% id, age name
% int, int type
% 1, 1 length
I done another insertion:
sql>insert into t1 values(2, 'mo', 26);
[ 1 ]
[ 1 ]
and I checked again the status:
sql>select * from t2;
% sys.t2, sys.t2 table_name
% id, age name
% int, int type
% 1, 2 length
[ 2, 26 ]
another insertion and status checked:
sql>insert into t1 values(2, 'mo', 28);
[ 2 ]
[ 1 ]
sql>select * from t2;
% sys.t2, sys.t2 table_name
% id, age name
% int, int type
% 1, 2 length
[ 2, 26 ]
[ 2, 26 ]
[ 2, 26 ]
Any idea about the problem?
In my opinion the problem is related with the commit.
Best Regards,
Romulo Goncalves
## Comment 6456
Date: 2007-01-17 20:12:50 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
The bug was fixed and it is covered by a test.
Regards,
Romulo
## Comment 6457
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1599763 at http://sourceforge.net/support/tracker.php?aid=1599763
| Triggers | https://api.github.com/repos/MonetDB/MonetDB/issues/1149/comments | 0 | 2020-11-30T09:23:28Z | 2024-06-28T13:31:17Z | https://github.com/MonetDB/MonetDB/issues/1149 | 753,306,466 | 1,149 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-20 11:36:07 +0100
From: @romulogoncalves
To: @njnes
Version: -- development
Last updated: 2006-11-23 12:43:00 +0100
## Comment 6443
Date: 2006-11-20 11:36:07 +0100
From: @romulogoncalves
Dear developers,
I just updated my SQL cvs head and the slam (ADT2006) test is red.
In the selection with the clause " like '%missing%' " one result is missing.
An similiar bug in the same query was founded one week ago in the suse and sun systems. Maybe the fix of that bug created this bug.
Best regards,
Romulo Goncalves
## Comment 6444
Date: 2006-11-20 11:43:24 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
fix might not have been propagated yet?
## Comment 6445
Date: 2006-11-21 17:55:54 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
slam fails on all platforms. Test output simply wasn't correct (= the same as stable).
Romulo, you could have figured and fixed this out yourself.
NO TEST / ALREADY IN TESTWEB
## Comment 6446
Date: 2006-11-21 18:24:30 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
One of the problems is the order created by the order by. In this case the result is correct only the order is different.
However I can not approve the test because three results are missing. In the current out only the second selecttion result of the following list appears:
[ "Mary Ann Walker", "brown", "brown hair turning grey", "five front teeth missing (Rumbelow); two bottom-one top\n\t\t\t\tfront (Fido), her teeth are slightly discoloured. She is\n\t\t\t\tdescribed as having small, delicate features with high\n\t\t\t\tcheekbones and grey eyes. She has a small scar on her\n\t\t\t\tforehead from a childhood injury. She is described by\n\t\t\t\tEmily Holland as \"a very clean woman who always seemed\n\t\t\t\tto keep to herself.\" The doctor at the post mortem\n\t\t\t\tremarked on the cleanliness of her thighs. She is also\n\t\t\t\tan alcoholic." ]
[ "Annie Chapman", "blue", "dark brown, wavy", "Pallid complexion, excellent teeth (possibly two missing in lower jaw), strongly built (stout), thick nose" ]
[ "Elisabeth Stride", "light gray", "curly dark brown", "pale complexion, all the teeth in her lower left jaw were missing" ]
[ "Pinchin Street Murder, possibly Lydia Hart", "", "", "body, missing both head and legs" ]
The query responsible for this selection is:
SELECT name, eyes, hair, features from Jtr_victim WHERE features LIKE '%missing%';
The problem that I reported in this bug is not yet complete resolved. Maybe the propagations from the stable branch to the cvs head were not complete/correctly done. This is a bug not realted with me. ;)
Best Regards,
Romulo Goncalves
## Comment 6447
Date: 2006-11-21 18:45:04 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
Sorry. I said that only the second one appears, but I was wrong.
I forgot that my machine crashed this morning during the updates.
After an update of all my system only the first one does not appear.
Best regards,
Romulo Goncalves
## Comment 6448
Date: 2006-11-21 20:59:59 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
so, you suggest to reopen the bug?
## Comment 6449
Date: 2006-11-21 21:03:10 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: YES
No.
I suggest to keep open the bug. ;)
Regards,
Romulo
## Comment 6450
Date: 2006-11-21 21:11:57 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
in that case we need to reopen the bug...
## Comment 6451
Date: 2006-11-22 10:31:06 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
This is an M5 issue. Now on Romulo's desk.
## Comment 6452
Date: 2006-11-22 10:35:54 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
By the way, the test fails in the current branch of M4 because of an ORDER issue. Please fix that too.
## Comment 6453
Date: 2006-11-23 12:43:00 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: NO
I fixed the orders, niels the PCRE on 5 in the end. All happy. :)
## Comment 6454
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1599597 at http://sourceforge.net/support/tracker.php?aid=1599597
| Slam out is red | https://api.github.com/repos/MonetDB/MonetDB/issues/1148/comments | 0 | 2020-11-30T09:23:24Z | 2024-06-27T11:21:27Z | https://github.com/MonetDB/MonetDB/issues/1148 | 753,306,420 | 1,148 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-17 12:39:37 +0100
From: Jan Rittinger <<jan.rittinger>>
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2006-12-07 07:38:23 +0100
## Comment 6439
Date: 2006-11-17 12:39:37 +0100
From: Jan Rittinger <<jan.rittinger>>
Hi,
The problem occurs in a leftjoin (see script where problematic operator is wrapped with print() statements).
Perhaps the reason is that the result gets to big.
> time Mserver foo_delta.mil
Monet Database Server V4.15.0
Copyright (c) 1993-2006, CWI. All rights reserved.
Compiled for i686-suse-linux/32bit with 32bit OIDs; dynamically linked.
Visit http://monetdb.cwi.nl/ for further information.
!ERROR: GDKload failed: name=07/771, ext=buns.priv
!ERROR: BATpropcheck: BAT parameter: BAT required.
!ERROR: CMDleftjoin: operation failed.
MonetDB>
real 2m11.783s
user 0m17.733s
sys 0m15.661s
## Comment 6440
Date: 2006-12-06 17:35:26 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
Originator: YES
Stefan already had a look at this problem and decided that the script runs out of address space. The error message however is misleading.
## Comment 6441
Date: 2006-12-07 19:38:23 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Fixed in CVS (MonetDB_4-14 release branch; propagation to development trunk pending):
Avoid calling BATpropcheck without BAT in case the previous operator failed to return a proper BAT.
No test added, as it would require a join that runs out-of memory/address space.
## Comment 6442
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1598310 at http://sourceforge.net/support/tracker.php?aid=1598310
| property checking and leftjoin, (size?) | https://api.github.com/repos/MonetDB/MonetDB/issues/1147/comments | 0 | 2020-11-30T09:23:21Z | 2024-06-27T11:21:26Z | https://github.com/MonetDB/MonetDB/issues/1147 | 753,306,386 | 1,147 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-17 12:39:07 +0100
From: @yzchang
To: @njnes
Version: -- development
Last updated: 2006-12-04 11:00:21 +0100
## Comment 6434
Date: 2006-11-17 12:39:07 +0100
From: @yzchang
I ran MonetDB with valgrind and the following error has been reported. Is this a bug or feature?
==12509== 1 errors in context 1 of 1:
==12509== Syscall param write(buf) points to uninitialised byte(s)
==12509== at 0x3A06A0B2EB: (within /lib64/libpthread-2.3.6.so)
==12509== by 0x5B4DF6F: GDKsave (gdk_storage.mx:278)
==12509== by 0x5A1FA73: HEAPsave (gdk_heap.mx:397)
==12509== by 0x5B4F7B8: BATsave (gdk_storage.mx:525)
==12509== by 0x5A1C297: BBPsync (gdk_bbp.mx:2767)
==12509== by 0x5A05831: TMsubcommit (gdk_tm.mx:160)
==12509== by 0xFE0721E: bm_subcommit (logger.mx:946)
==12509== by 0xFE098EA: bm_commit (logger.mx:1538)
==12509== by 0xFE06C4E: logger_commit (logger.mx:902)
==12509== by 0xFE07DCE: logger_exit (logger.mx:1146)
==12509== by 0xFE07ECE: logger_restart (logger.mx:1170)
==12509== by 0xFE07B49: logger_create (logger.mx:1086)
==12509== by 0xFE09AE2: logger_create_wrap (logger.mx:1584)
==12509== by 0xFE03C73: logger_create_wrap_unpack1696550061 (logger.glue.c:45)
==12509== by 0x5361399: interpret (monet_interpreter.mx:1173)
==12509== by 0x536A89C: interpret_assignment (monet_interpreter.mx:1834)
==12509== Address 0xDC8545C is 5,172 bytes inside a block of size 16,392 alloc'd
==12509== at 0x4904A66: malloc (vg_replace_malloc.c:149)
==12509== by 0x5AB3201: GDKmallocmax (gdk_utils.mx:1055)
==12509== by 0x5AB338C: GDKmalloc (gdk_utils.mx:1076)
==12509== by 0x5B4E213: GDKload (gdk_storage.mx:323)
==12509== by 0x5A1F8A8: HEAPload (gdk_heap.mx:359)
==12509== by 0x5B4FBC8: BATload_intern (gdk_storage.mx:578)
==12509== by 0x5A15998: BBPdescriptor (gdk_bbp.mx:1539)
==12509== by 0xFE05035: BATdescriptor (gdk.h:1171)
==12509== by 0xFE076AE: logger_new (logger.mx:1024)
==12509== by 0xFE07AFF: logger_create (logger.mx:1078)
==12509== by 0xFE09AE2: logger_create_wrap (logger.mx:1584)
==12509== by 0xFE03C73: logger_create_wrap_unpack1696550061 (logger.glue.c:45)
==12509== by 0x5361399: interpret (monet_interpreter.mx:1173)
==12509== by 0x536A89C: interpret_assignment (monet_interpreter.mx:1834)
==12509== by 0x536366C: interpret_var (monet_interpreter.mx:1328)
==12509== by 0x535F9DC: interpret monet_interpreter.mx:831)
Regards,
Jennie
## Comment 6435
Date: 2006-11-17 15:11:46 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
I haven't investigated this, but this could well be caused by having a BAT with unequal sized columns. If you have, e.g., a BAT with chr and lng columns, the chr column takes up 8 bytes for padding purposes, but only a single byte gets written, the other 7 remaining uninitialized. This is normally not a problem since nobody will read those 7 bytes. But when you write the BUN heap to disk, the hole thing is written, including the 7 uninitialized bytes. This is totally harmless.
But not knowing more about the BAT involved I can't say for sure that this is the cause for the report.
## Comment 6436
Date: 2006-11-17 18:15:56 +0100
From: @yzchang
Logged In: YES
user_id=341633
Originator: YES
I don't have time to look further into this now. The query I used is:
$ cat b.xq
import module namespace tf = "testfunctions"
at "/ufs/zhang/rpctest/testfunc.xq";
tf:add(3, 4)
$ cat testfunc.xq
module namespace tf = "testfunctions";
declare function tf:echoVoid()
{ () };
declare function tf:add($v1 as xs:integer, $v2 as xs:integer) as xs:integer
{ $v1 + $v2 };
declare function tf:echoAll($n as xs:anyNode*) as xs:anyNode*
{ $n };
I can try to find if this has been caused by unequal sized columns, when I have time.
Jennie
## Comment 6437
Date: 2006-12-04 11:00:21 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
The buns of a bat are stored as pairs, which sometimes requires a hole between them as data types have aligning constraints.
Also the var sized heaps (to store strings for example) have similar holes which are never initialized. For example are string lengths rounded up to multiples of 4 bytes.
Conclusion valgrinds complian is correct but we consider this not problem.
## Comment 6438
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1598309 at http://sourceforge.net/support/tracker.php?aid=1598309
| MonetDB: error in logger found by valgrind | https://api.github.com/repos/MonetDB/MonetDB/issues/1146/comments | 0 | 2020-11-30T09:23:18Z | 2024-06-28T13:32:03Z | https://github.com/MonetDB/MonetDB/issues/1146 | 753,306,343 | 1,146 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-14 12:56:45 +0100
From: @drstmane
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2006-11-17 03:11:22 +0100
## Comment 6429
Date: 2006-11-14 12:56:45 +0100
From: @drstmane
Shredding the document (incl. DTD) as provided with
test "SEQ" of the W3C XQuery use cases fails on parsing
the DTD with:
!ERROR:
/export/scratch0/monet/monet.GNU.64.64.d.413/pathfinder/tests/W3C_use_cases/XQ/SEQ/Tests/shred_doc.report1.dtd:7:
parser error : ContentDecl : Name or '(' expected
!ERROR: <!ELEMENT prep ( (PCDATA | action)* )>
!ERROR: ^
!ERROR:
/export/scratch0/monet/monet.GNU.64.64.d.413/pathfinder/tests/W3C_use_cases/XQ/SEQ/Tests/shred_doc.report1.dtd:7:
parser error : ContentDecl : ',' '|' or ')' expected
!ERROR: <!ELEMENT prep ( (PCDATA | action)* )>
!ERROR: ^
!ERROR:
/export/scratch0/monet/monet.GNU.64.64.d.413/pathfinder/tests/W3C_use_cases/XQ/SEQ/Tests/shred_doc.report1.dtd:7:
parser error : expected '>'
!ERROR: <!ELEMENT prep ( (PCDATA | action)* )>
!ERROR: ^
!ERROR:
/export/scratch0/monet/monet.GNU.64.64.d.413/pathfinder/tests/W3C_use_cases/XQ/SEQ/Tests/shred_doc.report1.dtd:7:
parser error : Content error in the external subset
!ERROR: <!ELEMENT prep ( (PCDATA | action)* )>
!ERROR: ^
!ERROR: [shred_url]: 1 times inserted nil due to errors
at tuples 0@0.
!ERROR: [shred_url]: first error was:
!ERROR: shred_external_subset: WARNING:
xmlParseDTD("/export/scratch0/monet/monet.GNU.64.64.d.413/pathfinder/tests/W3C_use_cases/XQ/SEQ/Tests/shred_doc.report1.dtd")
FAILED, NO ID/IDREF QUERIES
!ERROR: shred_external_subset: NOTE : maybe using
absolute filenames works, sorry!
cf.,
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_SEQ/shred_doc.out.00.html
Shreding the same document without the respective DTD
works fine:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_SEQ/shred_doc_noDTD.out.00.html
I haven't had the time to analyse this, and check,
whether the DTD is incorrect, or where (else) the
problems originate from.
Until we find the time to analyse the problem, we have
this bug report to document the situation and keep us
reminded.
## Comment 6430
Date: 2006-11-14 13:15:18 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
The test is failing correctly since the DTD is invalid.
An element declaration should follow the following rules:
elementdecl ::= '<!ELEMENT' S Name S contentspec S? '>'
contentspec ::= 'EMPTY' | 'ANY' | Mixed | children
And in this case the content specification is Mixed:
Mixed ::= '(' S? 'PCDATA' (S? '|' S? Name)* S? ')*' | '('
S? 'PCDATA' S? ')'
In other words, the DTD contains too many parentheses.
There is only one set allowed (and required) around the
"PCDATA | action".
## Comment 6431
Date: 2006-11-14 13:21:08 +0100
From: @Teggy
Logged In: YES
user_id=642632
Hi Stefan,
the DTD is syntactically wrong. An element with mixed
content has to
be declared as follows:
<!ELEMENT prep (PCDATA | action)*>
(note placement of the `*').
Cheers,
--Teggy
## Comment 6432
Date: 2006-11-17 15:11:22 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
I fixed the DTD in CVS.
Thanks to Teggy & Sjoerd for enlightening me.
## Comment 6433
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1596256 at http://sourceforge.net/support/tracker.php?aid=1596256
| XQ: shredding fails due to DTD with W3C XQ use case | https://api.github.com/repos/MonetDB/MonetDB/issues/1145/comments | 0 | 2020-11-30T09:23:14Z | 2024-06-27T11:21:24Z | https://github.com/MonetDB/MonetDB/issues/1145 | 753,306,305 | 1,145 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-14 12:56:40 +0100
From: @drstmane
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2007-06-05 08:15:05 +0200
## Comment 6425
Date: 2006-11-14 12:56:40 +0100
From: @drstmane
"Just" for the sake of finishing the (my) assessment
and documentation of the current status of the
Pathfinder (and MonetDB) code base wrt. the available
tests:
(Most probably) since Thursday Nov 2 2006, hence, after
(due to?) checkins on Tuesday Oct 31 2006 and/or
Wednesday Nov 1 2006 (there were no testing results for
this test on Wednesday Nov 1 2006),
pathfinder test runtime/Tests/merged_union.milS
produces different BAT IDs on 32-bit systems than on
64-bit systems.
See e.g.
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.32.d-Fedora4/runtime/merged_union.out.00.html
Note, the difference does not compromise the correct
behaviour of the test --- we could easily "fix" the
problem by either approving the 32-bit OID specific
output, or removing the statement that prints the
BAT-of-BATs result of mreged_union (and hence the BAT
IDs) --- but the actual fact that the BAT IDs
"suddently" differ *might* indecate, that there *could*
a potenial BUG somewhere else in the code ...
I didn't have the time to analyse this, yet.
Until we find the time to analyse the problem (or
decide that it is none), we have this bug report to
document the situation and keep us reminded.
## Comment 6426
Date: 2007-02-19 15:12:25 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
It looks like the change is due to bat reuse. Note that there are two ways in which the results differ from the old stable output. The results differ based on the size of OIDs: 32 or 64 bit.
We need to approve the new output and then close this report.
## Comment 6427
Date: 2007-06-05 08:15:05 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
(finally) approved specific stable output.
## Comment 6428
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1596255 at http://sourceforge.net/support/tracker.php?aid=1596255
| XQ: different BAT IDs with 32-bit OIDs in merged_union test | https://api.github.com/repos/MonetDB/MonetDB/issues/1144/comments | 0 | 2020-11-30T09:23:11Z | 2024-06-27T11:21:23Z | https://github.com/MonetDB/MonetDB/issues/1144 | 753,306,258 | 1,144 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-14 12:55:56 +0100
From: @romulogoncalves
To: Fabian Groffen <<fabian>>
Version: -- development
Last updated: 2006-11-14 02:16:14 +0100
## Comment 6419
Date: 2006-11-14 12:55:56 +0100
From: @romulogoncalves
Dear developers the Slam test fails for Suse and sun
systems.
Regards,
Romulo
## Comment 6420
Date: 2006-11-14 12:58:05 +0100
From: @grobian
Logged In: YES
user_id=963970
MonetDB Server v4.99.19
Copyright (c) 1993-2006 CWI, all rights reserved
Compiled for x86_64-pc-linux-gnu/64bit with 64bit OIDs
dynamically linked
config:/ufs/fabian/scratch/monetdb/five/program-x86_64/etc/MonetDB5.conf
dbfarm:/ufs/fabian/scratch/monetdb/five/program-x86_64/var/MonetDB5/dbfarm
dbname:demo
Visit http://monetdb.cwi.nl/ for further information
warning: please don't forget to set your vault key! (see
MonetDB5.conf file)
>sql.start();
>select 'bla missing (bla)' like '%missing%';
&1 0 1 1 1
% . table_name
% like_single_value name
% boolean type
% 5 length
[ true ]
>select 'ive front teeth missing (Rumbelow); two bottom-one
top\n\t\t\t\tfro' like '%missing%';
&1 0 1 1 1
% . table_name
% like_single_value name
% boolean type
% 5 length
[ false ]
>
this is tested by the recently added slam test.
## Comment 6421
Date: 2006-11-14 13:12:58 +0100
From: @drstmane
Logged In: YES
user_id=572415
Just a comment on Romulo's original report:
Indeed the output on SuSE Linux and Sun Solaris does differ
from the output on other systems:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/sql/.mTests103/GNU.32.32.d-SuSE9.3/src_test_ADT2006/slam.out.00.html
I suppose, this difference is caused by differences in the
available PCRE version(s) (resp. their configuration), just
as with this test:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/sql/.mTests103/GNU.32.32.d-SuSE9.3/src_test/like_tests.out.00.html
## Comment 6422
Date: 2006-11-14 13:27:51 +0100
From: @grobian
Logged In: YES
user_id=963970
in this case they don't have PCRE (afaict) and PCRE is wrong.
## Comment 6423
Date: 2006-11-14 14:16:14 +0100
From: @grobian
Logged In: YES
user_id=963970
I checked in a fix.
NO TEST / ALREADY A TEST
## Comment 6424
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1596253 at http://sourceforge.net/support/tracker.php?aid=1596253
| PCRE matching with multi-line strings fails | https://api.github.com/repos/MonetDB/MonetDB/issues/1143/comments | 0 | 2020-11-30T09:23:08Z | 2024-06-27T11:21:22Z | https://github.com/MonetDB/MonetDB/issues/1143 | 753,306,215 | 1,143 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-13 05:08:34 +0100
From: @drstmane
To: @sjoerdmullender
Version: -- development
Last updated: 2006-12-13 09:40:44 +0100
## Comment 6412
Date: 2006-11-13 17:08:34 +0100
From: @drstmane
One of the "relace value" statements in test
pathfinder/tests/W3C_use_cases/XQUF/AddressBook/Tests/Q1x.xq
seems to concat the new value to the old value of
doc("archive.xml")/*/last-synch-time instead of
replacing the latter with the former;
see
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_AddressBook/Q1x.out.00.html
for the update query, and
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQUF_AddressBook/check_docs.out.00.html
for the differences between the expected and the actual
result.
## Comment 6413
Date: 2006-11-13 17:11:17 +0100
From: @drstmane
Logged In: YES
user_id=572415
*might* be related to
[ 1591013 ] XQ: BATsetaccess warning/error with 32-bit OIDs
http://sourceforge.net/tracker/index.php?func=detail&aid=1591013&group_id=56967&atid=482468
## Comment 6414
Date: 2006-12-09 16:12:49 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
The problem seems to occur only with the second replacement of the same textnode's value in one query.
The first replace (with "2006-04-23T12:01") works fine (i.e., replaces the original "2005-10-05T10:00"), the second replace (with "2006-04-23T12:02") is never triggered the third replace (with "2006-04-23T12:03") then incorrectly appends "2006-04-23T12:03" to "2006-04-23T12:01" instead of replacing it.
When disabling the first replace, the third one correctly replaces the original "2005-10-05T10:00" by "2006-04-23T12:03".
## Comment 6415
Date: 2006-12-12 15:31:37 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
The problem is that the compiler translates do replace value of CDATA node with "string" into a combination of insert and delete. When the value of a text node is replaced twice, this results in two inserts and two deletes *of the same node*. The two inserts succeed, but only the first delete succeeds since by the time the second delete is attempted, the node is already gone. The net result is that the origninal text is replaced with the concatenation of the two new texts.
Since this is in the first instance a compiler problem, I assign it to Jens.
## Comment 6416
Date: 2006-12-12 15:46:46 +0100
From: Jens Teubner <<teubner>>
Logged In: YES
user_id=731390
Originator: NO
Apparently, the specs have changed in the respect. In the draft that I had implemented, the delete-insert combination was the correct behavior.
I'll do the necessary front-end stuff.
## Comment 6417
Date: 2006-12-13 21:40:44 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Jens changed the compiler to produce a single call for the do replace value query, and I did the implementation of the replacement.
No test needed since an existing test uncovered the bug.
## Comment 6418
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1595701 at http://sourceforge.net/support/tracker.php?aid=1595701
| XQ: "replace value" concatenates value iso. replacing it | https://api.github.com/repos/MonetDB/MonetDB/issues/1142/comments | 0 | 2020-11-30T09:23:04Z | 2024-06-27T11:21:21Z | https://github.com/MonetDB/MonetDB/issues/1142 | 753,306,171 | 1,142 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-13 12:35:43 +0100
From: @grobian
To: Romulo Goncalves <<goncalve>>
Version: -- development
Last updated: 2007-01-18 10:50:47 +0100
## Comment 6405
Date: 2006-11-13 12:35:43 +0100
From: @grobian
src/jdbc/tests/Test_JdbcClient fails to properly dump
the constraints entered into the database. This
indicates something has changed in the layout and
meaning of the various system tables. I believe this
might have to do with the recent addition of
dependencies in SQL, so assigning to Romulo. Feel free
to reassign if this bug is not caused by the dependencies.
## Comment 6406
Date: 2006-11-13 12:43:06 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
I just run the Mtest in sql/src/jdbc/tests and I did not get
any error.
Note: In the development trunk.
Regards,
Romulo
## Comment 6407
Date: 2007-01-15 22:00:59 +0100
From: @njnes
Logged In: YES
user_id=43556
Originator: NO
Whats the status of this bug report?
## Comment 6408
Date: 2007-01-15 22:09:18 +0100
From: @grobian
Logged In: YES
user_id=963970
Originator: YES
it still fails
## Comment 6409
Date: 2007-01-17 22:14:20 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: NO
Niels,
This test is already green, why the bug was not closed?
Regards,
Romulo
## Comment 6410
Date: 2007-01-18 10:50:47 +0100
From: @romulogoncalves
Logged In: YES
user_id=1498628
Originator: NO
The problem was fixed. There is already a test to cover it.
Regards,
Romulo
## Comment 6411
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1595556 at http://sourceforge.net/support/tracker.php?aid=1595556
| JDBC: constraint dumping fails | https://api.github.com/repos/MonetDB/MonetDB/issues/1141/comments | 0 | 2020-11-30T09:23:01Z | 2024-06-28T13:32:02Z | https://github.com/MonetDB/MonetDB/issues/1141 | 753,306,137 | 1,141 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-12 06:50:05 +0100
From: @drstmane
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2009-11-16 05:58:44 +0100
## Comment 6397
Date: 2006-11-12 18:50:05 +0100
From: @drstmane
The following three tests fail only in 64-bit Darwin
(MacOS X) with
ERROR = !WARNING: BATpropcheck: BAT tmp_556(366) was
incorrectly marked keyed!
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Darwin8.8.0/tests_Update/symm-test-5.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Darwin8.8.0/tests_W3C_use_cases_XQUF_R/Q7.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Darwin8.8.0/tests_W3C_use_cases_XQUF_R/Q7x.out.00.html
This *could* be a 64-bit big-endian related problem
either in some code setting BAT properties
(incorrectly) or in BATpropcheck itself.
## Comment 6398
Date: 2006-11-18 11:03:10 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Since Friday Nov 17 2006, i.e., after checkins on Thursday Nov 16 2006, test
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Darwin8.8.0/tests_W3C_use_cases_XQUF_R/Q7.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Darwin8.8.0/tests_W3C_use_cases_XQUF_R/Q7x.out.00.html
work fine also on 64-bit Darwin;
however since then, tests
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Darwin8.8.0/tests_BugTracker/insert_multiple.SF-1590580.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Darwin8.8.0/tests_BugTracker/inserting_multiple_elements.SF-1590583.out.00.html
fail with the above BATpropcheck warning/error only on 64-bit Darwin
(these tests used to fail (differently) on all platform before, but the respective bugs were fixed by Sjoerd on Thursday Nov 16 2006).
## Comment 6399
Date: 2006-12-20 12:04:52 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Most probably since its addition on Dec 14 2006,
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.64.d-Darwin8.8.0/tests_Update/replacevaluetest.err.00.html
fails only on 64-bit Darwin with
!ERROR: [swizzle]: tmp_424 contains illegal offsets.
## Comment 6400
Date: 2007-06-04 20:41:44 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Lowered priority, as we don't have access to any 64-bit Darwin machine, anymore.
## Comment 6401
Date: 2009-02-21 19:49:15 +0100
From: @grobian
to be more precise: "we don't have access to any 64-bit big-endian Darwin machine"
## Comment 6402
Date: 2009-02-21 20:07:18 +0100
From: @drstmane
right --- well, in fact, I personally and MonetDB testing currently don't have access to any MacOS X machine at all.
(details via personal communication, if required/desired)
## Comment 6403
Date: 2009-11-16 17:58:44 +0100
From: @drstmane
big-endian Darwin is considered history --- at least we don't have access to such a platform (anymore) --- and our only big-endian machine (a Sun T1000) is unfortunately too slow to also do Pathfinder/XQuery testing.
This means, there's practically no chance to investigate this one.
Closing.
If at all, the problem will re-appear by itself.
## Comment 6404
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1595129 at http://sourceforge.net/support/tracker.php?aid=1595129
| XQ: property error on 64-bit Darwin - big-endian related? | https://api.github.com/repos/MonetDB/MonetDB/issues/1140/comments | 0 | 2020-11-30T09:22:58Z | 2024-06-27T11:21:19Z | https://github.com/MonetDB/MonetDB/issues/1140 | 753,306,106 | 1,140 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-12 01:26:57 +0100
From: @peterboncz
To: @sjoerdmullender
Version: -- development
Last updated: 2007-01-10 02:51:28 +0100
## Comment 6393
Date: 2006-11-12 01:26:57 +0100
From: @peterboncz
the litmus test for 'acceptable' update performance, is
that it should be possible to execute 'trvivial'
updates at little cost (<100msec); regardless the size
of the data.
well, currently, we are a factor 100 from that goal.
my test is the 1GB Xmark document, shredded with
percFree = 10LL, e.g. (on the cluster):
shred_doc("/export/scratch1/manegold/ExpDB/Benchmarks/XMark/docs/XMark-f10.xml",
"a.xml, "a.xml", 10LL);
then, the query:
count(id("32297729",exactly-one(doc("a.xml")))/*)
should count 9 children of one particular description
element node, and we can make that 10 using the
following update:
do insert <test/> into
exactly-one(id("32297729",exactly-one(doc("a.xml"))))
I have attached a performance trace, which shows that
the performance problem is in do_update_insert, and in
particular in two loops that are taken around 600 times
and which do a lot of MIL processing .
use sort -k 2 -n -r aw.out
to sort the trace most expensive operations first.
Please ignore the index creation statements at the top;
this is only part of the first query.
After those, you will see a lot of repeated MIL
statements that individually do not take that much time
(though still 80 msec and downwards), but where the
problem is that they are executed in a loop that runs
hundreds of iterations.
Those loops can be found in pf_support.mx:
20270 while (holeneeded > 0) {
and:
20370 while (i <= pageno) {
I marked those and some other places up in the code
with a 'peter perf issue' comment.
My questions are:
- can we make those loops take less iterations (now, it
walks through the nodes one by one)
- can we move computations inside the loop outside?
- how much effort would it be to express those loops in C?
Currently, I see this as a show stopper for using our
updating functionality to prove anything about the
updatability of our XML storage.
## Comment 6394
Date: 2006-11-15 15:27:25 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
After my checkin today, the result on db2 is:
+ MapiClient -T -lx -s 'do insert <test/> into exactly-one(id("32297729",exactly-one(doc("a.xml"))))'
Trans 20.826 msec
Shred 851.248 msec
Query 5.296 msec
Update 1335.382 msec
## Comment 6395
Date: 2007-01-10 14:51:28 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
I'm closing this since the insert speed has been improved considerably since the bug was first reported.
## Comment 6396
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1594872 at http://sourceforge.net/support/tracker.php?aid=1594872
| PF: inserts really slow | https://api.github.com/repos/MonetDB/MonetDB/issues/1139/comments | 0 | 2020-11-30T09:22:54Z | 2024-06-27T11:21:18Z | https://github.com/MonetDB/MonetDB/issues/1139 | 753,306,064 | 1,139 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-10 11:27:21 +0100
From: @mlkersten
To: GDK devs <<bugs-common>>
Version: -- development
Last updated: 2006-11-11 12:26:02 +0100
## Comment 6387
Date: 2006-11-10 23:27:21 +0100
From: @mlkersten
Peter,
If you call BATkunion(l,r) with BATcount(l)>0 and
BATcount(r)=0 a BATcopy(l) will be triggered.
In turn this calls VIEWcreate_,
which increases the BBPref of l with one.
The result is that BATkunion has a side-effect.
It is left with 2 hard reference counts to the BAT l,
while the returned VIEW is already isolated.
The gross effect is that in M5 garbage collection
does not work, and temporary BATs fill up your memory.
line 736: in gdk_setop
if (BATcount(r) == 0) {
return BATcopy(l, l->htype, l->ttype, FALSE);
}
Shouldn't we decrement the ref before returning?
Is it safe to do so?
if( BATcount(r) == 0) {
BAT v= BATcopy(l,l->htype,l->ttype,FALSE);
BBPunfix(l);
return v;
}
The easiest way to test this behavior is to run 09.sql
against M5. Statement _19:= algebra.kunion(...) is the
point of error.
## Comment 6388
Date: 2006-11-11 00:22:34 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Hi Martin,
Sorry, I cannot see any error in the code. Each GDK call
that delivers a BAT result should set a refcount on that
BAT. This protects the result bat from being deallocated.
The view mechanism adds another fix to l, but when the view
is deallocated, this fix is removed. This is also correct;
the source BAT cannot disappear while a view on it is still
in use (sharing its resources).
The MIL/MAL interpreter should take care of pinning bats
while they are in use (i.e. when they are being handled by a
command). This is the reason of the first fix on l. After
executing the command, the arguments should be unfixed again
by the MIL/MAL interpreter. The same goes for the results of
expressions. Maybe the cause of the BAT leak is there.
It is not in the BATkunion, as far as I can see.
Peter
does not really change that.
## Comment 6389
Date: 2006-11-11 00:26:02 +0100
From: @njnes
Logged In: YES
user_id=43556
The refcnt is a result from the resulting view using the
left operant.
## Comment 6390
Date: 2006-11-11 10:29:08 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Niels,
I know that. What is your point?
Peter
## Comment 6391
Date: 2006-11-11 11:01:42 +0100
From: @mlkersten
Logged In: YES
user_id=490798
Indeed the initial bugreport was invalid, but it was triggered
by the assumption that any view is recognized in the target BAT
with a sharecnt >0. This already protects the hard reference
of a BAT* from the view. As long as you don't free a BAT
when this
assumption hold the VIEW is safe.
In my case, I first released the target (decrement its
lrefs) and then the view lref, but which did not trigger proper
decrement of its hard ref on the target when lref becomes zero.
## Comment 6392
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1594453 at http://sourceforge.net/support/tracker.php?aid=1594453
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-common@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| GDK: BATkunion side effect considered harmfull | https://api.github.com/repos/MonetDB/MonetDB/issues/1138/comments | 0 | 2020-11-30T09:22:51Z | 2024-06-28T13:32:02Z | https://github.com/MonetDB/MonetDB/issues/1138 | 753,306,024 | 1,138 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-09 05:46:46 +0100
From: @yzchang
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2006-11-09 06:43:02 +0100
## Comment 6384
Date: 2006-11-09 17:46:46 +0100
From: @yzchang
I just did a fresh check out of all modules belonging
to the release. Compiling MonetDB (buildtools is
successfully recompiled) failed with the following error:
libtool: install: warning: relinking `libmonet.la'
/ufs/zhang/scratch/monet/stable/program/bin/mel:
"../../../../src/modules/plain/str.mx", line 69, Atom
not found at "3" of wrd.
/ufs/zhang/scratch/monet/stable/program/bin/mel:
"../../../../src/modules/plain/str.mx", line 69, Atom
not found at "3" of wrd.
/ufs/zhang/scratch/monet/stable/program/bin/mel:
"../../../../src/modules/plain/str.mx", line 69, Atom
not found at "3" of wrd.
Found 3 syntax errors.
make[6]: *** [str.proto.h] Error 1
make[5]: *** [install-recursive] Error 1
make[4]: *** [install] Error 2
make[3]: *** [install-recursive] Error 1
make[2]: *** [install] Error 2
make[1]: *** [install-recursive] Error 1
make: *** [install] Error 2
Does anybody have an idea what is wrong? Thanks!
Jennie
## Comment 6385
Date: 2006-11-09 18:43:02 +0100
From: @drstmane
Logged In: YES
user_id=572415
Jennie,
your buildtools are outdated, you must use branch
MonetDB_4-14, not MonetDB_4-12 (also for buildtools):
$ grep . /ufs/zhang/scratch/monet/stable/*/CVS/Tag
/ufs/zhang/scratch/monet/stable/buildtools/CVS/Tag:TMonetDB_4-12
/ufs/zhang/scratch/monet/stable/MonetDB/CVS/Tag:TMonetDB_4-14
/ufs/zhang/scratch/monet/stable/pathfinder/CVS/Tag:TXQuery_0-14
/ufs/zhang/scratch/monet/stable/sql/CVS/Tag:TSQL_2-12
## Comment 6386
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1593583 at http://sourceforge.net/support/tracker.php?aid=1593583
| Compilation error MonetDB 4.14 | https://api.github.com/repos/MonetDB/MonetDB/issues/1137/comments | 0 | 2020-11-30T09:22:48Z | 2024-06-28T13:32:01Z | https://github.com/MonetDB/MonetDB/issues/1137 | 753,305,975 | 1,137 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-09 12:05:05 +0100
From: @peterboncz
To: @sjoerdmullender
Version: -- development
Last updated: 2009-04-07 10:56:27 +0200
## Comment 6380
Date: 2006-11-09 12:05:05 +0100
From: @peterboncz
map_pid is currently logged in full for each transaction.
in the 10GB xmark, we map-pid has size 8000, which will
times 1+8+8 bytes, thus 136KB per single update.
given that map_pid is [void,oid], it would be better
not log the head, and use the array-write functionality
in the logger (that is faster and shorter than a log
call for each bun).
[ SIDE ISSUE: it would also be faster and shorter to
use array-write for all other [oid,any] delta bats;
splitting them by column ]
additionally, given that the mapi_pid contains page
numbers, we could convert it in te log to an [void,int]
bat, or maybe even a [void,sht] bat, and dynamically
cast upwards such datatypes during log recovery. In the
10GB case, this would reduce map_pid logging overhead
to to 16KB per query.
[ SIDE ISSUE AGAIN: array write save the log mode byte
plus this upcasting will probably save additional
space (on oids) on all other delta bats as well.
An interesting idea would be to use the X100
compression interface for writing to the log,
especially when all deltas are written with the array
interface. ]
But, for map_pid, it would be better to encode inserts
and deletes, rather than log the entire thing:
e.g. delta [oid,oid] head = insertpos, oid newtail
where
- insertpos is the original position after
which the insert is done.
- multiple identical insertpos entries indicate multipl
tuples inserted at that pos
- negative (higest bit set) insertpos are deletes
## Comment 6381
Date: 2007-06-04 20:38:00 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
Peter, Sjoerd,
is this one still an open issues after all the recent XQUF code changes/improvements?
Stefan
## Comment 6382
Date: 2009-04-07 10:56:26 +0200
From: @peterboncz
will not be fixed anymore
## Comment 6383
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1593305 at http://sourceforge.net/support/tracker.php?aid=1593305
| PF: perf issue: map_pid logging | https://api.github.com/repos/MonetDB/MonetDB/issues/1136/comments | 0 | 2020-11-30T09:22:45Z | 2024-06-28T13:32:00Z | https://github.com/MonetDB/MonetDB/issues/1136 | 753,305,944 | 1,136 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-09 11:38:49 +0100
From: @peterboncz
To: @njnes
Version: -- development
CC: @sjoerdmullender
Last updated: 2010-07-09 15:17:29 +0200
## Comment 6378
Date: 2006-11-09 11:38:49 +0100
From: @peterboncz
If reading the logs fails, logger_create() should
create an error. Now the return value of
logger_readlogs() is ignored, see line 800 in logger.mx:
logger_readlogs(logger, fp, filename);
the database is then just started as if nothing
happened. With a corrupt log, subsequent updates will
also be lost, because after hitting the corrupt point
in the logfile, logger_readlogs() gives up.
In my case, the logger had been corrupted because my
test repository was old. So my fault, but
nothwitstanding that, the logger should be designed to
handle failures.
I'm actually wondering what will happen if the log
contains an incomplete transaction. That is a legal
situation. According to the rules, such a transaction
is incomplete and thus aborted. In that case, I expect
all previous transactions to be recovered, and in fact
no error should be reported. Maybe this is the reason
that the return value of logger_readlogs() is ignored?
However, the incomplete trasaction (that will cause
error when reading the log) may obstruct future
succesful transactions appended to the log. Is that
indeed a problem?? The way to handle this would be to
start writing in the logfile at the byte position where
the imcomplete transaction started. I do not know
whether that is the case now.
[in any case, I would prefer more flexible logging,
where transactions could log deltas intertwined,
because this releaves lock pressure. This also leads to
a situation where the logger may contain any amount of
incomplete information. Still, each individual non-last
log-delta should be complete, otherwise we have
corruption. ]
The main point of this BugReport is that one has to
distinguish between:
(1) corrupt deltas (e.g. obviously wrong codes,
inexistent bat names, impossible data values). The log
should be sanitized (i.e. we acknowledge having lost
data, but bring back the system in usable state) before
the database is restarted, or one should not allow it
to restart (but then, one has to provide a sanitizing
script -- not very attractive)
(2) incomplete deltas (indicating a crashed database
that did not make commit). These deltas should not be
applied, and the database may be restarted (though some
warning would be nice).
my requests concern these two types of failures, and
their handling:
(1a) please detect lock corruption, report an error,
and sanitize the system (preferable)
(1b) after signalling corruption, make it possible to
get back the log in sane state without losing the
entire repository
(2a) report incomplete deltas in the log
(2b) ensure that incomplete transactions in the log
will not obstruct recovering the log later when other
completed transactions have been appended to it
A possible approach during logger_create() is to
recover all complete transactions from the log, and
then perform a checkpoint on all data bats, and remove
all logfiles (a.k.a. log restart). This should work
even also on the sane part of a later corrupt log.
Secondly, the requirement that all bats should be
mentioned in the logger catalog could be lifted. When a
transaction is logged that menions a new bat, it could
be added silently to the catalog (now this error is
ignored!!), and the state of the catalog should be
marked "dirty". When writing the commit record with a
dirty catalog, the logger should schedule a subcommit
first, to make sure the catalog is ok.
If these suggestions are followed, we can check the
logger_readlog() for errors, and just remove all log
files and clean the entire catalog in response. We will
arrive in a clean state where everything that was
recovereable has been recovered.
## Comment 6379
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1593287 at http://sourceforge.net/support/tracker.php?aid=1593287
Bug has invalid status, setting status to "NEW".
Previous status was "DELETED".
## Comment 14318
Date: 2010-07-09 14:47:21 +0200
From: @sjoerdmullender
(In reply to comment 0)
> If reading the logs fails, logger_create() should
> create an error. Now the return value of
> logger_readlogs() is ignored, see line 800 in logger.mx:
>
> logger_readlogs(logger, fp, filename);
The fact that logger_readlogs()'s result isn't checked is a red herring. logger_readlogs() always returns 0, because logger_readlog() always returns 0.
Also, as the comment in logger_readlogs() states, "we cannot distinguish errors from incomplete transactions".
But perhaps we do need to think about in what circumstances the logger should abort (i.e. exit the server) and when it should just struggle on.
## Comment 14341
Date: 2010-07-09 15:17:29 +0200
From: @sjoerdmullender
Bug was DELETED in SF.
| logger_create does not check for errors => DISASTER | https://api.github.com/repos/MonetDB/MonetDB/issues/1135/comments | 0 | 2020-11-30T09:22:41Z | 2024-06-28T13:31:59Z | https://github.com/MonetDB/MonetDB/issues/1135 | 753,305,899 | 1,135 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-09 11:09:31 +0100
From: @peterboncz
To: @njnes
Version: -- development
Last updated: 2007-12-03 04:20:02 +0100
## Comment 6372
Date: 2006-11-09 11:09:31 +0100
From: @peterboncz
If reading the logs fails, logger_create() should
create an error. Now the return value of
logger_readlogs() is ignored, see line 800 in logger.mx:
logger_readlogs(logger, fp, filename);
the database is then just started as if nothing
happened. With a corrupt log, subsequent updates will
also be lost, because after hitting the corrupt point
in the logfile, logger_readlogs() gives up.
In my case, the logger had been corrupted because my
test repository was old. So my fault, but
nothwitstanding that, the logger should be designed to
handle failures.
I'm actually wondering what will happen if the log
contains an incomplete transaction. That is a legal
situation. According to the rules, such a transaction
is incomplete and thus aborted. In that case, I expect
all previous transactions to be recovered, and in fact
no error should be reported. Maybe this is the reason
that the return value of logger_readlogs() is ignored?
However, the incomplete trasaction (that will cause
error when reading the log) may obstruct future
succesful transactions appended to the log. Is that
indeed a problem?? The way to handle this would be to
start writing in the logfile at the byte position where
the imcomplete transaction started. I do not know
whether that is the case now.
[in any case, I would prefer more flexible logging,
where transactions could log deltas intertwined,
because this releaves lock pressure. This also leads to
a situation where the logger may contain any amount of
incomplete information. Still, each individual non-last
log-delta should be complete, otherwise we have
corruption. ]
The main point of this BugReport is that one has to
distinguish between:
(1) corrupt deltas (e.g. obviously wrong codes,
inexistent bat names, impossible data values). The log
should be sanitized (i.e. we acknowledge having lost
data, but bring back the system in usable state) before
the database is restarted, or one should not allow it
to restart (but then, one has to provide a sanitizing
script -- not very attractive)
(2) incomplete deltas (indicating a crashed database
that did not make commit). These deltas should not be
applied, and the database may be restarted (though some
warning would be nice).
my requests concern these two types of failures, and
their handling:
(1a) please detect lock corruption, report an error,
and sanitize the system (preferable)
(1b) after signalling corruption, make it possible to
get back the log in sane state without losing the
entire repository
(2a) report incomplete deltas in the log
(2b) ensure that incomplete transactions in the log
will not obstruct recovering the log later when other
completed transactions have been appended to it
A possible approach during logger_create() is to
recover all complete transactions from the log, and
then perform a checkpoint on all data bats, and remove
all logfiles (a.k.a. log restart). This should work
even also on the sane part of a later corrupt log.
Secondly, the requirement that all bats should be
mentioned in the logger catalog could be lifted. When a
transaction is logged that menions a new bat, it could
be added silently to the catalog (now this error is
ignored!!), and the state of the catalog should be
marked "dirty". When writing the commit record with a
dirty catalog, the logger should schedule a subcommit
first, to make sure the catalog is ok.
If these suggestions are followed, we can check the
logger_readlog() for errors, and just remove all log
files and clean the entire catalog in response. We will
arrive in a clean state where everything that was
recovereable has been recovered.
## Comment 6373
Date: 2006-11-10 23:23:50 +0100
From: @njnes
Logged In: YES
user_id=43556
The logger cannot distinquish between crashes and incorrect
logs (as these can be caused by crashes!).
The users should have backups to solve hardware problems.
The logger always starts a new log file after a broken log
is read, this makes sure no updates are lost
after a broken log. Also the logger now keeps its changes in
memory before applying (as required by xqueries updates). At
the end of recovering the logger always restarts, ie saves
the current status in bats and starts a new log.
## Comment 6374
Date: 2006-11-11 00:31:54 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Hi Niels,
It is good to know that you restart the log after an
incorrect log. That prevents a number of issues.
However, this is not what I observed. My corrupt log was
kept on multiple session starts, and thus kept failing in
the recovery process.
I think you can distinguish between a corrupt logfile and a
crashed logfile (excluding weird crashes that caused corrupt
file writes to the log or something). A crashed logfile is a
truncated logfile (i.e. cut-off). Any other abnormal logfile
is corrupt.
In any case, it seems the log_delta code should be more
defensive, as data on bats had been logged that did not
appear in the catalog.
Peter
## Comment 6375
Date: 2006-12-02 20:47:42 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: NO
set to "Pending" as we should think of creating tests also for the logger/recovery, just like we should try to bukd tests for memory limitations, etc.
## Comment 6376
Date: 2007-12-03 04:20:02 +0100
From: Pseudo user for Sourceforge import <<sfimport>>
---- Original comment by: sf-robot@users.sourceforge.net ----
Logged In: YES
user_id=1312539
Originator: NO
This Tracker item was closed automatically by the system. It was
previously set to a Pending status, and the original submitter
did not respond within 365 days (the time period specified by
the administrator of this Tracker).
## Comment 6377
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1593271 at http://sourceforge.net/support/tracker.php?aid=1593271
| logger_create does not check for errors | https://api.github.com/repos/MonetDB/MonetDB/issues/1134/comments | 0 | 2020-11-30T09:22:39Z | 2024-06-27T11:21:13Z | https://github.com/MonetDB/MonetDB/issues/1134 | 753,305,863 | 1,134 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-08 04:10:49 +0100
From: Jan Rittinger <<jan.rittinger>>
To: @njnes
Version: -- development
Last updated: 2009-07-28 02:26:09 +0200
## Comment 6360
Date: 2006-11-08 16:10:49 +0100
From: Jan Rittinger <<jan.rittinger>>
Hi,
attached you find a generated mil file that works on
the auction file in pathfinder/benchmarks/XMark and
does strange things when appending a bat with values to
an empty one.
I couldn't get the example smaller and assume that it
is somehow property related.
## Comment 6361
Date: 2006-11-08 16:19:09 +0100
From: @drstmane
Logged In: YES
user_id=572415
Did you run Mserver with default (i.e., --debug-0)?
Could you also try with Mserver --debug=10 ?
What's the result?
## Comment 6362
Date: 2006-11-08 16:25:10 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
sorry -- forgot to try that one. It returns the correct result:
-----------------
h t name
void oid type
-----------------
[ 0@0, 3@0 ]
[ 1@0, 1@0 ]
The info (of the second call -- its in a while loop) does
not look different except for halign, talign, batId,
batCacheid, batStamp, lastUsed, and curStamp. These fields
do look quite innocent or?
## Comment 6363
Date: 2006-11-10 19:32:52 +0100
From: @njnes
Logged In: YES
user_id=43556
problem is that the input too BATappend(a,b) a.tdense isn't
set. Therefor the void,void isn't materialized. The cause is
probably the copy of the kdiff result.
## Comment 6364
Date: 2006-11-10 21:38:17 +0100
From: @njnes
Logged In: YES
user_id=43556
Fixed bug in BATappend(a,b). In case of empty void,void bat
'a' we set the t-seqbase to that of 'b'. But that seqbase
can be nil leading to strange results. This case is now fixed.
Switched to pending as we still need a test script.
## Comment 6365
Date: 2007-11-11 04:20:02 +0100
From: Pseudo user for Sourceforge import <<sfimport>>
---- Original comment by: sf-robot@users.sourceforge.net ----
Logged In: YES
user_id=1312539
Originator: NO
This Tracker item was closed automatically by the system. It was
previously set to a Pending status, and the original submitter
did not respond within 365 days (the time period specified by
the administrator of this Tracker).
## Comment 6366
Date: 2007-11-22 11:35:42 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Switched to pending as we still need a test script.
## Comment 6367
Date: 2008-11-22 03:20:04 +0100
From: Pseudo user for Sourceforge import <<sfimport>>
---- Original comment by: sf-robot@users.sourceforge.net ----
This Tracker item was closed automatically by the system. It was
previously set to a Pending status, and the original submitter
did not respond within 365 days (the time period specified by
the administrator of this Tracker).
## Comment 6368
Date: 2009-02-16 13:49:26 +0100
From: @drstmane
re-opened to remind us that we should consider adding a test
## Comment 6369
Date: 2009-02-17 01:24:54 +0100
From: @drstmane
the attached MIL script does not work with "modern" MonetDB/XQuery any more
## Comment 6370
Date: 2009-07-28 14:26:09 +0200
From: @sjoerdmullender
Added the script with modification (ws_create() gets an argument now): pathfinder/tests/BugTracker/value_append_to_nil.SF-1592689.
Closing.
## Comment 6371
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1592689 at http://sourceforge.net/support/tracker.php?aid=1592689
| append() translates values into nil | https://api.github.com/repos/MonetDB/MonetDB/issues/1133/comments | 0 | 2020-11-30T09:22:36Z | 2024-06-27T11:21:12Z | https://github.com/MonetDB/MonetDB/issues/1133 | 753,305,830 | 1,133 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-08 01:03:23 +0100
From: @drstmane
To: XQuery devs <<bugs-xquery>>
Version: -- development
Last updated: 2006-11-08 04:21:57 +0100
## Comment 6355
Date: 2006-11-08 13:03:23 +0100
From: @drstmane
configuring Pathfidner with --disable-strict
(default in the "Stable" release branch; nightly
testing currently overrules the defaults and makes the
"Stable" release branches with --enable-strict, but the
"Current" development trunk with --disable-strict)
fails with
configure: error: Pathfinder requires a C99 compliant C
compiler, but your C compiler does not seem to be C99
compliant.
Background:
Without --enable-strict, configure is not allowed to
<quote>(illegally) inject crap in the CFLAGS</quote>;
however, without "-std=c99" (or alike) C compilers
don't identify themselves as C99 compiant.
Since Pathfinder code requires (some) C99 compliance,
we stop instantly with a proper error message as
opposed to failing later on with some (often vague)
compiler error.
## Comment 6356
Date: 2006-11-08 13:04:38 +0100
From: @drstmane
Logged In: YES
user_id=572415
sorry, for got to set "Category" and "Group".
## Comment 6357
Date: 2006-11-08 15:39:51 +0100
From: @grobian
Logged In: YES
user_id=963970
"crap" != absolutely necessary flags needed to be able to
compile the source package.
Injecting -std=c99 obviously is fine, if configure figures
that that works.
## Comment 6358
Date: 2006-11-08 16:21:56 +0100
From: @njnes
Logged In: YES
user_id=43556
we are back to adding c99 options to the CFLAGS.
## Comment 6359
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1592590 at http://sourceforge.net/support/tracker.php?aid=1592590
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-xquery@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| Pathfinder fails to configure with --disable-strict | https://api.github.com/repos/MonetDB/MonetDB/issues/1132/comments | 0 | 2020-11-30T09:22:32Z | 2024-06-27T11:21:11Z | https://github.com/MonetDB/MonetDB/issues/1132 | 753,305,791 | 1,132 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-08 10:56:19 +0100
From: @drstmane
To: @sjoerdmullender
Version: -- development
Last updated: 2006-11-08 01:35:23 +0100
## Comment 6352
Date: 2006-11-08 10:56:19 +0100
From: @drstmane
When configure with --disable-strict
(default in the "Stable" release branch; nightly
testing currently overrules the defaults and makes the
"Stable" release branches with --enable-strict, but the
"Current" development trunk with --disable-strict)
MonetDB fails to compile with icc on FC4 & SuSE 9.3:
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB/.ntv.64.32.d-Fedora4/make-k.out
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB/.ntv.64.64.d-Fedora4/make-k.out
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB/.ntv.64.64.d-SuSE9.3/make-k.out
## Comment 6353
Date: 2006-11-08 13:35:23 +0100
From: @njnes
Logged In: YES
user_id=43556
cwi local icc installation bug. The icc.cfg now contains
fixes for this problem.
## Comment 6354
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1592540 at http://sourceforge.net/support/tracker.php?aid=1592540
| MonetDB: configure --disable-strict => icc-compile fails | https://api.github.com/repos/MonetDB/MonetDB/issues/1131/comments | 0 | 2020-11-30T09:22:30Z | 2024-06-27T11:21:10Z | https://github.com/MonetDB/MonetDB/issues/1131 | 753,305,758 | 1,131 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-08 10:54:54 +0100
From: Jan Rittinger <<jan.rittinger>>
To: @njnes
Version: -- development
Last updated: 2006-11-10 11:16:17 +0100
## Comment 6338
Date: 2006-11-08 10:54:54 +0100
From: Jan Rittinger <<jan.rittinger>>
Peter,
I think the testweb can only work again if you align
Mtest with the pfstart() changes you applied.
## Comment 6339
Date: 2006-11-08 10:59:21 +0100
From: Jan Rittinger <<jan.rittinger>>
Logged In: YES
user_id=993208
In addition it seems to generate an error at *only* some of
the queries.
!ERROR: interpret_params: ^P(param 1): evaluation error.
(This error however does not break the evaluation.)
## Comment 6340
Date: 2006-11-08 11:34:53 +0100
From: @peterboncz
Logged In: YES
user_id=591107
could you be more specific please.
- one example of a failing test
- what is the error message/symptom?
## Comment 6341
Date: 2006-11-08 11:36:58 +0100
From: @drstmane
Logged In: YES
user_id=572415
FYI:
========
$ grep --color .
{buildtools,MonetDB,sql,pathfinder,monetweb}/CVS/Tag
--------
buildtools/CVS/Tag:TMonetDB_4-14
MonetDB/CVS/Tag:TMonetDB_4-14
sql/CVS/Tag:TSQL_2-14
pathfinder/CVS/Tag:TXQuery_0-14
monetweb/CVS/Tag:TMonetDB_4-12_old
========
========
$ find {buildtools,MonetDB,sql,pathfinder,monetweb}/ -type f
| xargs egrep --color '^(|[^].*)(pfstart|xquery_server_start)'
--------
./pathfinder/HowToStart-PF: Mserver
--dbinit="module(pathfinder); mapi_start(); pfstart();"
./pathfinder/HowToStart-PF: Mserver --set
monet_mod_path=<monetdb-prefix>/lib(64)/MonetDB:<pathfinder-prefix>/lib(64)/MonetDB
--dbinit="module(pathfinder); mapi_start(); pfstart();"
./pathfinder/runtime/pathfinder.mx:PROC pfstart() : void { }
deprecated
./pathfinder/runtime/Tests/procs.stable.out:[ "pfstart() :
void"
]
./pathfinder/runtime/Tests/procs.stable.out:[ "pfstart(bit,
lng) : void"
]
./pathfinder/runtime/Tests/procs.stable.out:[
"xquery_server_start() : void"
]
./pathfinder/runtime/Tests/procs.stable.out:[
"xquery_server_start(bit) : void"
]
./pathfinder/runtime/Tests/sigs.stable.out:[ "pfstart(bit,
lng) : void" ]
./pathfinder/tests/XQTS.sh: echo 'Please start it with
(e.g.) `Mserver --dbinit="module(pathfinder); pfstart();"`.'
./pathfinder/tests/BugDay_2005-12-19_0.9.3/Tests/multiple_servers_2.SF-1385152.py:
x += 1; clients(x,"module(mapi);module(pathfinder);
mapi_register(mil_frontend()); pfstart();")
./pathfinder/tests/BugDay_2005-12-19_0.9.3/Tests/multiple_servers_2.SF-1385152.py:
x += 1; clients(x,"module(mapi);module(pathfinder);
pfstart(); mapi_register(mil_frontend());")
./pathfinder/tests/BugDay_2005-12-19_0.9.3/Tests/multiple_servers_2.SF-1385152.py:
x += 1; clients(x,"module(pathfinder);module(mapi);
mapi_register(mil_frontend()); pfstart();")
./pathfinder/tests/BugDay_2005-12-19_0.9.3/Tests/multiple_servers_2.SF-1385152.py:
x += 1; clients(x,"module(pathfinder);module(mapi);
pfstart(); mapi_register(mil_frontend());")
./pathfinder/tests/BugDay_2005-12-19_0.9.3/Tests/multiple_servers_2.SF-1385152.stable.err:server
11 : "module(mapi);module(pathfinder);
mapi_register(mil_frontend()); pfstart();"
[...]
./pathfinder/tests/BugDay_2005-12-19_0.9.3/Tests/multiple_servers_2.SF-1385152.stable.err:server
46 : "module(pathfinder);module(mapi); pfstart();
mapi_register(mil_frontend());", client 4: (MAPI) (MAPI)
(XQUERY) XQUERY
./pathfinder/tests/BugDay_2005-12-19_0.9.3/Tests/multiple_servers_2.SF-1385152.stable.out:server
11 : "module(mapi);module(pathfinder);
mapi_register(mil_frontend()); pfstart();"
[...]
./pathfinder/tests/BugDay_2005-12-19_0.9.3/Tests/multiple_servers_2.SF-1385152.stable.out:server
46 : "module(pathfinder);module(mapi); pfstart();
mapi_register(mil_frontend());", client 4: (MAPI) (MAPI)
(XQUERY) XQUERY
./MonetDB/src/testing/Mtest.py.in: Srvr = '%s
--dbinit="module(pathfinder); pfstart();"' % Srvr
./MonetDB/src/testing/README:
mapi_start()/sql_server_start()/pfstart() is called, but
before (the first)
./MonetDB/src/testing/README: * Mserver --dbname=TSTDB
--set xquery_port=$XQUERYPORT --dbinit="module(pathfinder);
pfstart();" [TST.prologue] &
./monetweb/XQuery/Demo/HelloWorld/content.shtml: Mserver
--dbinit="module(pathfinder); pfstart(); mapi_start();"
./monetweb/XQuery/Documentation/FAQ/content.shtml:have
forgotten to start the MAPI Server module:
<code>pfstart();</code> and/or <code>mapi_start();</code>.
./monetweb/XQuery/Interface/POSIX/content.shtml:
MonetDB> pfstart();
========
## Comment 6342
Date: 2006-11-08 11:44:17 +0100
From: @drstmane
Logged In: YES
user_id=572415
and of course:
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/index_short.html
and
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/index_short.html
(once the results have been collected/analysed; see your
mailbox)
## Comment 6343
Date: 2006-11-08 11:51:12 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Stefan,
this grep does not provide useful information, regrettably.
pfstart() is indeed in the tests but it can remain there (it
does not do anything anymore - that is another matter).
Sure, some proc signatures have disappeared, but apart from
the signature listing, this should not produce significant
differences as these were not used.
The output of my test run, started at 2:00 AM can be browsed at:
http://www.cwi.nl/~boncz/mTests/
while there are clearly some failing/hanging tests, and
logger related errors (see e.g. BugTracker !), the clear
majority of tests work. I.e. it is what I would expect today.
My test run also gets some failing tests. Sometimes. It
depends on the run. It seems to be related with timeouts.
e.g. in the WebSite tests, this error appears:
! Server not ready; skipping attempt to start client!
when I run just WebSite by hand (on a lighter loaded machie,
presumably), it works fine..
Maybe due to the logger initialization, Mserver starts up
slower. From now on, this seems hard to avoid, though. Maybe
the client timeout should be increased.
Peter
## Comment 6344
Date: 2006-11-08 12:29:14 +0100
From: @peterboncz
Logged In: YES
user_id=591107
I examined the test run, disregarding Update (Sjoerd?), XRPC
(Jennie?) and the new W3C use cases.
the following output is ok:
tests/runtime/{sigs,procs,merged_union}
tests/BugDay_2005-10-06_0.9.3/shred_at_depth_258.SF-1174747
?? tests/BugDay_2005-10-06_0.9.3/Entities.SF-1185932
(shredder now says this XML is invalid??)
tests/BugTracker/empty_distinct_values.SF-1431357
tests/BugsViaSourgeforce/ID.1267018.swallowed_error_messages-1
tests/BugsViaSourgeforce/ID.1267018.swallowed_error_messages-2
The Mserver should now load the pathfinder module (not only
MapiClient). This has to be fixed in some tests:
tests/BugDay_2005-11-09_0.9.3/delete_doc.SF-1208129
tests/BugDay_2005-12-19_0.9.3/file_locked_after_shredding.SF-1238352
tests/BugDay_2005-12-19_0.9.3/text_nodes_with_empty_content_seq.SF-1350418
tests/BugDay_2005-12-19_0.9.3/attribute_after_non-attribute_content.SF-1351516
tests/BugDay_2005-12-19_0.9.3/shredding_one_attribute.SF-1376942
I arrive at the following new bugs (other errors seem
identical, correspond to known open bugs):
tests/BugDay_2005-12-19_0.9.3/non-existent_attribute.SF-1324219
tests/BugDay_2005-12-19_0.9.3/oc-order_in_attributes.SF-1328671
tests/runtime/serialization
problem wil xquery() MIL function
tests/Bugday_2005-12-19_0.3.3 multiple_server_2.SF-1385152
logger errors
tests/BugTracker/.mposjoin_void_void.SF-1404023.q1b.src.index.html
other logger errors
tests/BugTracker/Tests/modules_diamond.SF-1510975a.xq
no output
tests/BugTracker/Tests/if-condition_testing_empty_sequence
corrupted DB!!
!FATAL: BATpropcheck: BAT tmp_116(-78) has inconsistent
descriptor 32768 (100000)
## Comment 6345
Date: 2006-11-08 12:53:22 +0100
From: @drstmane
Logged In: YES
user_id=572415
Peter,
my "grep' gives information where "pfstart()" is still
present in the code AND *documentation* (and how to aquire
this information); i.e., the places I would have expected to
chance together with any (significant) changes related to
pfstart() like removing it (or actually making it a no-op
--- personally, I would have considered it much "cleaner"
style of programming, if pfstart() had been remove
completely as opposed to making it an empty no-op --- but as
said that just my personal view).
If you do not consider this information "useful", please
feel free to ignore it. I'm sorry.
For xquery (.xq) (as well as SQL (.sql) and MIL MAPI
(.milC)) tests, Mtest basically starts the Mserver using a
kind of "hand-shake" mechanism to ensure the server is
indeed ready before the client is started. This work as follows:
Mserver is called as
Mserver --dbinit="module(pathfinder); pfstart();"
keeping pipes for stdin, stdout, and stderr.
Then, 'printf("\nReady.\n");' is sent to Mserver's stdin
(i.e., console), and its output (stdout) is scanned for the
"Ready." prompt. Once it occurs the server is assumed to be
ready, and the client is called to execute the actual test
script.
In case stdout ends/is closed, before "Ready." has occured,
or Mtest fails to sucessfully ping (cf. "mapi_ping" in
Mtest.py) the server (with/for given port and language),
Mtest assumes the Mserver has failed to start and hence
skips the attempt to start the client. That where "! Server
not ready; skipping attempt to start client!" comes from.
The minimal (default) timeout used in Mtest is 60 seconds (1
minute), hence, there is no timeout involved in case the "!
Server not ready; skipping attempt to start client!" occurs
within less than a minute.
E.g., it occurs within 4 seconds in
http://homepages.cwi.nl/~boncz/mTests/tests/i18n/.good-002.out.html
http://homepages.cwi.nl/~boncz/mTests/tests/i18n/good-002.test.out
In fact, in above example, the "Ready." does occur, hence,
it seem to be the "mapi_ping(port, lang)" that seems to fail
--- can only guess why: because the console thread executes
the 'printf("\nReady.\");' before the Mserver has finish
loading module pathfinder and starting the mapi listerner?
Well if that's the case, Mtest's "hand-shake" mechanism
obviously cannot work any more.
Which would be a pitty, because it used to be very convient
to distinguish really failing (client-)tests from those
cases where client-test were forced to fail as the server
failed to start properly.
Well since I have no clue how 'module(pathfinder);' works
nowadays (sorry for my "ignorance") I unfortunately cannot
tell, why Mtest's "mapi_ping(port,lang)" sanity check does
fail, now.
Stefan
## Comment 6346
Date: 2006-11-08 13:14:42 +0100
From: @drstmane
Logged In: YES
user_id=572415
re: tests/BugDay_2005-10-06_0.9.3/Entities.SF-1185932:
shredder has been saying that this XML is invalid since
Tuesday Aug 15 2006;
cf.
http://sourceforge.net/tracker/index.php?func=detail&aid=1544002&group_id=56967&atid=482468
most (if not all) tests that did fail with the CVS code base
of yesterday (Tue Nov 7 2006) 17:00 CET
are documented in the respective bug reports;
cf.
http://sourceforge.net/tracker/?func=browse&group_id=56967&atid=482468
## Comment 6347
Date: 2006-11-08 13:42:36 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Hi Stefan,
I do hear your (and Jan's) requests for "clean" programming,
but this is a case where I simply had to resign to the fact
that I'm overwhelmed by the sheer amount of features I have
to develop (apart from Sjoerd, I have not received much help
from anybody apart from constructive comments on my code).
So, leaving pfstart() in there is another instance of me
going for an avenue of less development effort. As we have
all seen, it has cost me dearly to make even this deadline,
since the (for me) infamous MonetDB meeting in 2005 where I
was not present, where the target for updates in september
2005 was set (i.e. from 4 months, it became 2 years -- sorry
about that). Just to say, help is welcome and appreciated!
With regards to stabilizing the system, I'm certainly
counting on you (and appreciating the effort).
There should be nothing mysterious about the old function of
pfstart(), it served to start the mapi xquery client. A
simple text search for mapi_listen in pathfinder.mx will
lead you to conclude that currently this is simply done in
the module initialization script. There should be *no*
difference in behavior to 'module(pathfinder);' now and
'module(pathfinder); pfstart();' before.
The only thing now is that we do an additional fork for the
logger, and the logger scans any existing logfile (as
loggers should do). This can take CPU time. BTW the
requirements of the logger mean that we have to initialize
the module in the server session (including checkpoint
thread) otherwide the database maybe corrupt, which made it
more logical to start the mapi client thread as well by
default. As the system is slower, it just takes the
fork(mapi_listen) a bit longer to get ready. Waiting a
couple of seconds more before connecting should do the trick.
Peter
## Comment 6348
Date: 2006-11-08 14:47:17 +0100
From: @drstmane
Logged In: YES
user_id=572415
Peter,
I do acknowledge your (as well as Sjoerd's and Niels') work
on the update features.
Unfortunately, I did not manage to keep up with the pace of
developement to find the niche were I could help
constructively on the features in a non-conflicting way.
Please accept my honest and sincere apologies for that!
(Maybe, I still need someone who tells me excatly what to do...)
Given that, I tried to contribute according to what I though
were my "skills", i.e., by at least monitoring the
correctness/stability of the system, assuming that detecting
bugs and other problems as early after their
"creation"/"introduction" as possible would make it simpler
to analyse debug and fix the problems, as well as by (trying
to) keep a eye on the global picture, i.e., monitoring the
impact of chance of someone on some end of the system/code
on the work of someone else on the other end of the systems.
In retrospect, I have to conclude that I apparently was far
too eager on this, obviously "annoying" people with the
information about recently introduced problems (well, I
actually did try to fix most of them myself at least as far
as I was sure about the intention of the changes), rather
than helping them to solve the problems quickly, and prevent
similar problems in the future.
Thinking about it, I see "communication" (or actually the
lack of that) the major source for all the (not only) recent
"confusion". E.g., I apparenly did never managed to make
clear how I think development and testing should work in an
environtment were many people work on different parts of a
complex system that inherently (though not alsways
obviously) depend on each other i.e., a "small and handy
change" on one side can lead to "total disaster" on the
other end. I hoped, that exhaustive testing could help and
trigger more awareness of each other --- apparently, I was
too naive, here.
Unfortunately, I also never did understand what me fellow
developers do consider a convenient working enviroment in
the given complex context; hence I failed to contribute my
part on improving the situation.
Likewise, I failed (e.g.) to interpret your pfstart()
changes to that extend, that it was your intention to keep
the empty pfstart() as dummy around, instead of removing it
completely, and hence have a "convenient" trigger for all
places why require attentions/fixes related to the pfstart()
changes --- as net effect changing features and their
respective use case inside our system AND test suite would
have given other developers/users an impression (if not
"tutorial") as to whether the changes have impact on their
code and how to adapt/fix that ("instant inlined
documentation by example") --- but again, it's only my
personal "layman's" approach to development and programming.
Concerning the feature freeze deadline:
I proposed to have it yesterday for pure egoistic reasons:
Given my limited skills / efficiency when working on too
many topics at the same time, I wanted to get my
"MonetDB-desk" clean before doing other work (reviews that
are due by the end of this week). However, I didn't mean to
appear as "dictator" wrt. the deadline. I had known that
postponing the feature freeze deadline by whatever time
would a have prevented the confusion of last night (oops,
there's the "communication" again) and today I would have
had no objections at all (in fact, given the "distraction"
of today, I haven't even started with the work I wanted/need
to do ...).
Concenrning stabilizing the system:
As already mentioned on Monday, I'll be available for that
as of next week Tuesday. I tried to leave all my findings so
far well documented.
Concerning Mtest & client tests:
As explained before there is now timeout or timed wait
involved; Mtest uses a simplyfied/layman's "hand-shake" to
find out whether the server is ready for clients.
Apparently, this does not work any more. I don't have an
idea how to fix this properly, yet. Adding a delay it IMHO
not good solution --- if the delay is too long testing will
take unnecessarily long increasing the threshold for using
it (even) more; if the delay is too short, it does not help
--- these are actually the very reason why the "hand-shake"
mechanism has been introduces in the first place. But of
course, I'm open to any suggestion --- I can only learn from
them.
Kind Regards,
Stefan
## Comment 6349
Date: 2006-11-08 14:57:32 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Hi Stefan.
Thanks for the acks, but I do not accept any apologies and
self-dequalifications since they are unwarranted, and I
certainly did not (mean to) ask for these kind of
statements. Only for some help towards finishing this
release. Reading your post, there should be no problem and
you have my thanks in anticipation.
I am fine with removing pf_start() but thought it wiser
first to try to expose and fix the "real" bugs that are out
there. That's what I'm doing right now, actually.
So let us just move forward.
best,
Peter
## Comment 6350
Date: 2006-11-10 23:16:17 +0100
From: @njnes
Logged In: YES
user_id=43556
pfstart() is gone. And removed from the Mtest scripts (and
we introduced dbinit scripts). Allthough many tests still
fail this is not
caused by the pfstart() changes.
## Comment 6351
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1592538 at http://sourceforge.net/support/tracker.php?aid=1592538
| Testweb: testweb broken because of pfstart()? | https://api.github.com/repos/MonetDB/MonetDB/issues/1130/comments | 0 | 2020-11-30T09:22:26Z | 2024-06-27T11:21:09Z | https://github.com/MonetDB/MonetDB/issues/1130 | 753,305,713 | 1,130 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-08 10:52:15 +0100
From: Jan Rittinger <<jan.rittinger>>
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-11-10 09:43:46 +0100
## Comment 6335
Date: 2006-11-08 10:52:15 +0100
From: Jan Rittinger <<jan.rittinger>>
(Some) compilers expect that BATprepareHash copes with
its return value (see pf_support.mx lines 6189, 6190).
## Comment 6336
Date: 2006-11-08 13:25:56 +0100
From: Jens Teubner <<teubner>>
Logged In: YES
user_id=731390
gcc -DHAVE_CONFIG_H -I. -I. -I.. -I. -I/opt/MonetDB/include/MonetDB -I/opt/MonetDB/include/MonetDB/common -I/opt/MonetDB/include/MonetDB/gdk -I/
opt/MonetDB/include/MonetDB/monet -I/opt/MonetDB/include/MonetDB/plain -I/opt/MonetDB/include/MonetDB/contrib -I/opt/MonetDB/include/MonetDB/mapi
-I/opt/MonetDB/include/MonetDB/C -I/usr/include/libxml2 -I../compiler/include -I./../compiler/include -DLIBPF_SUPPORT -g -O2 -std=c99 -Wall -W -
Werror-implicit-function-declaration -Werror -Wno-unused-function -Wno-unused-label -DEMBEDDED -DMT -DNO_CGI -DNO_AUTH -DIO_MAX=10240 -c
pf_support.c -fPIC -DPIC -o .libs/lib_pf_support_la-pf_support.o
cc1: warnings being treated as errors
pf_support.mx: In function 'pf_support_prelude':
pf_support.mx:6189: warning: value computed is not used
pf_support.mx:6190: warning: value computed is not used
make[1]: *** [lib_pf_support_la-pf_support.lo] Error 1
make[1]: Leaving directory `/home/teubner/pf/cvs-head/runtime'
make: *** [all] Error 2
## Comment 6337
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1592537 at http://sourceforge.net/support/tracker.php?aid=1592537
| XQuery: Use of BATprepareHash breaks compilation | https://api.github.com/repos/MonetDB/MonetDB/issues/1129/comments | 0 | 2020-11-30T09:22:23Z | 2024-06-27T11:21:08Z | https://github.com/MonetDB/MonetDB/issues/1129 | 753,305,679 | 1,129 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-08 09:42:44 +0100
From: @grobian
To: @njnes
Version: -- development
Last updated: 2006-11-08 10:32:23 +0100
## Comment 6321
Date: 2006-11-08 09:42:44 +0100
From: @grobian
cc1: warnings being treated as errors
/Volumes/Scratch/monetdb/current/monetdb/src/modules/plain/bat.mx:
In function 'local_itoa':
/Volumes/Scratch/monetdb/current/monetdb/src/modules/plain/bat.mx:1253:
warning: format '%d' expects type 'int', but argument 4
has type 'ssize_t'
make[7]: *** [lib_bat_la-bat.lo] Error 1
make[7]: *** Waiting for unfinished jobs....
make[6]: *** [all] Error 2
make[5]: *** [all-recursive] Error 1
make[4]: *** [all] Error 2
make[3]: *** [all-recursive] Error 1
make[2]: *** [all] Error 2
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2
% gcc --version
gcc (GCC) 4.0.1 (Apple Computer, Inc. build 5341)
Copyright (C) 2005 Free Software Foundation, Inc.
This is free software; see the source for copying
conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE
% echo $MCFLAGS
-mcpu=G5 -mtune=G5 -mpowerpc64 -mpowerpc-gpopt -pipe -g
-W -Wall
% echo $MX_CFLAGS
-Werror
(no -Ox flags involved)
## Comment 6322
Date: 2006-11-08 09:50:35 +0100
From: @drstmane
Logged In: YES
user_id=572415
Which version of MonetDB? Branch or Head?
I.e., configure default --disable-strict (branch) or
--enable-strict (head)?
## Comment 6323
Date: 2006-11-08 09:54:01 +0100
From: @grobian
Logged In: YES
user_id=963970
current, which is the head. Sorry, forgot to set it.
% echo $MCONF_EXT_OPTS
--disable-strict
% echo $MMAKE_EXT_OPTS
-j2
## Comment 6324
Date: 2006-11-08 10:01:06 +0100
From: @grobian
Logged In: YES
user_id=963970
I get the impression somehow detection of bitsizes and stuff
goes wrong ("fixing" this issue makes it abort not so much
further in the code), hence I attach my config.log
## Comment 6325
Date: 2006-11-08 10:37:00 +0100
From: @grobian
Logged In: YES
user_id=963970
Not surprisingly it also fails to compile on the
MonetDB_4-14 branch. The only notable difference it that it
takes way longer, but this is because of configure
(illegally) injecting a whole lot of crap in the CFLAGS, of
course.
## Comment 6326
Date: 2006-11-08 15:37:54 +0100
From: @grobian
Logged In: YES
user_id=963970
With a fix from Niels I can get a bit further:
cc1: warnings being treated as errors
/Volumes/Scratch/monetdb/stable/monetdb/src/gdk/gdk_desc.mx:
In function 'BATsavedesc':
/Volumes/Scratch/monetdb/stable/monetdb/src/gdk/gdk_desc.mx:111:
warning: format '%u' expects type 'unsigned int', but
argument 3 has type 'size_t'
/Volumes/Scratch/monetdb/stable/monetdb/src/gdk/gdk_desc.mx:135:
warning: format '%u' expects type 'unsigned int', but
argument 3 has type 'size_t'
## Comment 6327
Date: 2006-11-08 16:13:23 +0100
From: @drstmane
Logged In: YES
user_id=572415
"cc1: warnings being treated as errors" (i.e., -Werror) is
NOT / should NOT be the default in the release branch --- in
case you get despite configuring the release banch with
default settings (in particular without explicite
--enable-strict) or with configuring the release branch or
the development trunk with explicit --disable-strict, then
the default settings resp. the handling of --disable-strict
is broken.
In the release branch, by default, only optimization options
for "known" compilers (possibly not all latest versions are
"known", though) are added to CFLAGS --- this can be disable
by configuring with --disable-optimize, which should yield
the configure default i.e., "-g -O2" for gcc and "-g" for
most other compilers.
Could you please also share with us the complete compiler
call that triggers the error?
Likewise, it would be handy to see the actual CFLAGS and
X_CFLAGS setting from your <builddir>/Makefile.
## Comment 6328
Date: 2006-11-08 16:29:16 +0100
From: @grobian
Logged In: YES
user_id=963970
True, I'm just trying to get it to compile without any
warnings. Luckily the -Werror will not be there in the
release, so these errors may not be fatal. That is actually
pretty nice, and I hope you now see why. Because I'm using
-Werror, I switched back the bug to current/HEAD. By the
way, I do not like the default use of "optimize" in the
release, because it's non-standard.
$CC invocation for current/HEAD
(cd .libs && rm -f lib_constant.la && ln -s
../lib_constant.la lib_constant.la)
/bin/sh ../../../libtool --tag=CC --mode=compile
powerpc-apple-darwin8-gcc -DH
AVE_CONFIG_H -I. -I../../..
-I/Volumes/Scratch/monetdb/current/monetdb/src/modul
es/plain
-I/Volumes/Scratch/monetdb/current/monetdb/src/modules/plain
-I../../co
mmon
-I/Volumes/Scratch/monetdb/current/monetdb/src/modules/plain/../../common
-
I../../gdk
-I/Volumes/Scratch/monetdb/current/monetdb/src/modules/plain/../../gd
k -I../../monet
-I/Volumes/Scratch/monetdb/current/monetdb/src/modules/plain/../
../monet -I/Library/Gentoo/usr/include -DLIBDECIMAL
-mcpu=G5 -mtune=G5 -
mpowerpc64 -mpowerpc-gpopt -pipe -g -W -Wall -Werror
-D_REENTRANT -c -o lib_dec
imal_la-decimal.lo `test -f 'decimal.c' || echo
'/Volumes/Scratch/monetdb/curren
t/monetdb/src/modules/plain/'`decimal.c
% egrep "^(X_|)CFLAGS" $BUILD_ROOT/monetdb/Makefile
CFLAGS = -mcpu=G5 -mtune=G5 -mpowerpc64 -mpowerpc-gpopt
-pipe -g -W -Wall $(X_CFLAGS)
CFLAGS_NO_OPT = -O0
X_CFLAGS = -Werror
The CFLAGS try to use the features of the G5 processor, i.e.
native 64-bits ints handling and CPU sqrt instructions.
## Comment 6329
Date: 2006-11-08 16:39:44 +0100
From: @drstmane
Logged In: YES
user_id=572415
sorry, overlook your config.log...
in looks, as if X_CFLAGS='-Werror' is set despite
--disable-strict; IMHO, that should not be the case.
the sizes look OK to me:
define SIZEOF_CHAR 1
define SIZEOF_SHORT 2
define SIZEOF_INT 4
define SIZEOF_LONG 4
define SIZEOF_VOID_P 4
define SIZEOF_SIZE_T 4
define SIZEOF_SSIZE_T 4
define SIZEOF_PTRDIFF_T 4
define SIZEOF_LONG_LONG 8
define SIZEOF___INT64 0
However I'm wondering why configure checks for C++ --- we
IMHO don't ask it to do so, and it doesn't on my Linux box ...
... maybe, it's related to the fact, that you (or your
environment) calls configure with
"--host=powerpc-apple-darwin8 --build=powerpc-apple-darwin8"
(and our configure cannot handle that properly)?
## Comment 6330
Date: 2006-11-08 16:42:10 +0100
From: @grobian
Logged In: YES
user_id=963970
as I showed in my initial comment, I inject -Werror myself.
With regard to C++, within Gentoo we have a epunt_cxx for
that, so I don't think that's a problem you can solve.
## Comment 6331
Date: 2006-11-08 22:06:00 +0100
From: @grobian
Logged In: YES
user_id=963970
back on stable again, as Niels "fixed" something there:
I'm about to upgrade my compiler, but this one still thinks
there is some unused variable around:
cc1: warnings being treated as errors
/Volumes/Scratch/monetdb/stable/monetdb/src/modules/plain/logger.mx:
In function 'logger_commit':
/Volumes/Scratch/monetdb/stable/monetdb/src/modules/plain/logger.mx:675:
warning: unused variable 'b'
make[7]: *** [lib_logger_la-logger.lo] Error 1
make[7]: *** Waiting for unfinished jobs....
make[6]: *** [all] Error 2
make[5]: *** [all-recursive] Error 1
make[4]: *** [all] Error 2
make[3]: *** [all-recursive] Error 1
make[2]: *** [all] Error 2
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2
## Comment 6332
Date: 2006-11-08 22:32:23 +0100
From: @grobian
Logged In: YES
user_id=963970
w00t! It actually compiled right now! Not going to try sql
or anything else for now!
## Comment 6333
Date: 2006-11-08 22:32:49 +0100
From: @grobian
Logged In: YES
user_id=963970
NO TEST / COMPILATION
## Comment 6334
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1592503 at http://sourceforge.net/support/tracker.php?aid=1592503
| M4: build failure on Darwin | https://api.github.com/repos/MonetDB/MonetDB/issues/1128/comments | 0 | 2020-11-30T09:22:20Z | 2024-06-27T11:21:07Z | https://github.com/MonetDB/MonetDB/issues/1128 | 753,305,643 | 1,128 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-08 08:21:53 +0100
From: @drstmane
To: @njnes
Version: -- development
Last updated: 2006-11-08 01:12:11 +0100
## Comment 6314
Date: 2006-11-08 08:21:53 +0100
From: @drstmane
The following tests fail after yesterday's checkins.
Please check and either fix the bugs or approve the new
output.
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/MonetDB/.mTests103/GNU.64.64.d-Fedora4/tests_BugTracker/leak_of_view_gdk_bbp.SF-1430968.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/MonetDB/.mTests103/GNU.64.64.d-Fedora4/tests_Availability/00_Mserver.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/MonetDB/.mTests103/GNU.64.64.d-Fedora4/tests_suite_08/test_01.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/sql/.mTests103/GNU.64.64.d-Fedora4/src_test_BugDay_2005-12-19_2.9.3/select_from_env.SF-1240701.1242164.out.00.html
## Comment 6315
Date: 2006-11-08 09:30:12 +0100
From: @peterboncz
Logged In: YES
user_id=591107
sorry Stefan, I do not have a ready MonetDB test web; these
outputs seem ok (could you approve them??),
except for the leak_of_view
Martin: did some memory tracing by you slip into CVS?
## Comment 6316
Date: 2006-11-08 09:34:34 +0100
From: @drstmane
Logged In: YES
user_id=572415
a) Simply run testing (at least for the part that you
changed, i.e., for MonetDB, if you changes MonetDB code)
*before* checking in, then you do have the testoutput at hand.
b) See
http://monetdb.cwi.nl/Development/TestWeb/Stable/approve/
## Comment 6317
Date: 2006-11-08 13:02:19 +0100
From: @njnes
Logged In: YES
user_id=43556
The introduced BATdeleteHash caused a memory leak. This is
reverted now, fixing the leak_of_view etc test. The others
are caused by changes in the environment variables. No extra
test needed as this bug allready involves the above listed
tests.
## Comment 6318
Date: 2006-11-08 13:07:39 +0100
From: @drstmane
Logged In: YES
user_id=572415
sorry, re-opened as the changed output due to changes in the
environment variables has not been approved, yet.
## Comment 6319
Date: 2006-11-08 13:12:11 +0100
From: @njnes
Logged In: YES
user_id=43556
okay approved now
## Comment 6320
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1592464 at http://sourceforge.net/support/tracker.php?aid=1592464
| MonetDB & SQL: tests fail after checkins | https://api.github.com/repos/MonetDB/MonetDB/issues/1127/comments | 0 | 2020-11-30T09:22:16Z | 2024-06-27T11:21:06Z | https://github.com/MonetDB/MonetDB/issues/1127 | 753,305,599 | 1,127 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-08 08:11:55 +0100
From: Jens Teubner <<teubner>>
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-11-10 11:14:01 +0100
## Comment 6310
Date: 2006-11-08 08:11:55 +0100
From: Jens Teubner <<teubner>>
gcc -DHAVE_CONFIG_H -I. -I. -I../.. -I. -I../include -
I./../include -I/opt/MonetDB/include/MonetDB -I/opt/
MonetDB/include/MonetDB/common -I/opt/MonetDB/include/
MonetDB/gdk -I/opt/MonetDB/include/MonetDB/monet -I/
opt/MonetDB/include/MonetDB/plain -I/opt/MonetDB/
include/MonetDB/contrib -I/opt/MonetDB/include/MonetDB/
mapi -I/opt/MonetDB/include/MonetDB/C -DLIBMIL -g -O2 -
std=c99 -Wall -W -Werror-implicit-function-declaration
-Werror -Wno-unused-function -Wno-unused-label -c
milprint_summer.c -fPIC -DPIC -o .libs/libmil_la-
milprint_summer.o
cc1: warnings being treated as errors
milprint_summer.c: In function 'translate2MIL':
milprint_summer.c:6412: warning: 'item_ext' may be
used uninitialized in this function
milprint_summer.c:6411: warning: 'rc' may be used
uninitialized in this functionmake[5]: *** [libmil_la-
milprint_summer.lo] Error 1
make[5]: Leaving directory `/home/teubner/pf/cvs-head/
compiler/mil'
make[4]: *** [all] Error 2
make[4]: Leaving directory `/home/teubner/pf/cvs-head/
compiler/mil'
make[3]: *** [all-recursive] Error 1
make[3]: Leaving directory `/home/teubner/pf/cvs-head/
compiler'
make[2]: *** [all] Error 2
make[2]: Leaving directory `/home/teubner/pf/cvs-head/
compiler'
make[1]: *** [all-recursive] Error 1
make[1]: Leaving directory `/home/teubner/pf/cvs-head'
make: *** [all] Error 2
gcc (GCC) 4.1.1 (Gentoo 4.1.1)
My compiler seems to be quite picky here. Assigning
to Peter as he was (I think) the last one who checked
in to milprint_summer.c.
## Comment 6311
Date: 2006-11-08 08:28:26 +0100
From: @drstmane
Logged In: YES
user_id=572415
My compiler (gcc 4.0.2 on FC4/64bit) is just as "picky" (or
simply *correct*) --- at least with optimization enabled.
## Comment 6312
Date: 2006-11-10 23:13:44 +0100
From: @njnes
Logged In: YES
user_id=43556
current version compiles on fedora core 6/gcc-4.1.1-30
## Comment 6313
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1592457 at http://sourceforge.net/support/tracker.php?aid=1592457
| PF: milprint_summer.c does not compile on GCC 4.1.1 | https://api.github.com/repos/MonetDB/MonetDB/issues/1126/comments | 0 | 2020-11-30T09:22:13Z | 2024-06-27T11:21:05Z | https://github.com/MonetDB/MonetDB/issues/1126 | 753,305,547 | 1,126 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-07 01:18:40 +0100
From: @grobian
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2006-11-07 02:18:01 +0100
## Comment 6307
Date: 2006-11-07 13:18:40 +0100
From: @grobian
config.status: WARNING:
/ufs/fabian/scratch/monetdb/current/monetdb/src/tools/monetdb-config.bat.in
seems to ignore the --datarootdir setting
Making and installing MonetDB in
/ufs/fabian/scratch/vtmp/build-current-monetdb-pegasus.ins.cwi.nl/monetdb
mkdir: cannot create directory `.libs': File exists
cc1: warnings being treated as errors
/ufs/fabian/scratch/monetdb/current/monetdb/src/monet/yytree.mx:
In function 'yyprint':
/ufs/fabian/scratch/monetdb/current/monetdb/src/monet/yytree.mx:843:
warning: format '%lx' expects type 'long unsigned int',
but argument 3 has type 'ptr'
make[5]: *** [libmonet_la-yytree.lo] Error 1
make[5]: *** Waiting for unfinished jobs....
make[4]: *** [all] Error 2
make[3]: *** [all-recursive] Error 1
make[2]: *** [all] Error 2
make[1]: *** [all-recursive] Error 1
make: *** [all] Error 2
(pegasus:~/.monet) fabian% gcc --version
gcc (GCC) 4.1.1 (Gentoo 4.1.1-r1)
Copyright (C) 2006 Free Software Foundation, Inc.
This is free software; see the source for copying
conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A
PARTICULAR PURPOSE.
## Comment 6308
Date: 2006-11-07 14:18:01 +0100
From: @drstmane
Logged In: YES
user_id=572415
fixed in CVS:
PTRFMT must not be used without PTRFMTCAST
## Comment 6309
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1591963 at http://sourceforge.net/support/tracker.php?aid=1591963
| M4: compilation error | https://api.github.com/repos/MonetDB/MonetDB/issues/1125/comments | 0 | 2020-11-30T09:22:10Z | 2024-06-27T11:21:04Z | https://github.com/MonetDB/MonetDB/issues/1125 | 753,305,517 | 1,125 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-07 01:12:40 +0100
From: @dkwakkel
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2006-11-07 03:39:49 +0100
## Comment 6302
Date: 2006-11-07 13:12:40 +0100
From: @dkwakkel
When running bootstrap I got the following error:
bash-3.1$ ./bootstrap
File not found - .incs.*
automake 1.10 is 1.7 or newer. Good.
autoconf 2.60 is 2.57 or newer. Good.
libtool 1.5.23a is 1.4 or newer. Good.
Python 2.4.3 is 2.0.0 or newer. Good.
Traceback (most recent call last):
File "/usr/local/bin/autogen.py", line 29, in ?
from autogen import am
ImportError: No module named autogen
bash-3.1$ pwd
/cygdrive/d/CWI/MonetDB
## Comment 6303
Date: 2006-11-07 13:18:14 +0100
From: @drstmane
Logged In: YES
user_id=572415
$ cat buildtools/autogen/README
To build, run
python setup.py build
then to install, run
python setup.py install [--prefix=<install-dir-prefix>]
Use
python setup.py --help
to get more information about the possible options.
If you use the --prefix option or otherwise install autogen in a
different place, you may have to set your PYTHONPATH environment
variable accordingly. The directory which needs to be added is
<install-dir-prefix>/lib(64)/python<python-version>/site-packages,
where
<python-version> is the version of Python used (e.g. 2.4).
You can automatically get the respective directory via
"<install-dir-prefix>/`python -c 'import
distutils.sysconfig; print
distutils.sysconfig.get_python_lib(0,0,"")'`".
On Windows, to follow these instructions, Python must of
course be
found in your PATH. The default location where autogen.py is
installed is the Scripts subdirectory of the Python tree
(i.e. C:\Python24\Scripts if you have installed Python 2.4
in the
default location). Make sure this folder is in your PATH when
building MonetDB and the other packages.
## Comment 6304
Date: 2006-11-07 15:22:13 +0100
From: @dkwakkel
Logged In: YES
user_id=886515
This works, thanks!
But when I followed the README files top-down, the buildtools/README file stated that to build the tools in
one go I just have to do:
./bootstrap; ./configure; make; make install
Is it not possible to do this python stuff automatic? If not I would expect a not in the buildtools/README
file.
## Comment 6305
Date: 2006-11-07 15:39:49 +0100
From: @drstmane
Logged In: YES
user_id=572415
I added
"
In particular, see autogen/README for how to set your
PYTHONPATH in case
./bootstrap of MonetDB, sql, pathfinder, etc. fails to find
the autogen
module with the default settings.
"
to the top-level buildtools README.
[Compilation/usage; hence, no (further) tests required.]
## Comment 6306
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1591961 at http://sourceforge.net/support/tracker.php?aid=1591961
| autogen import problem when building MonetDB | https://api.github.com/repos/MonetDB/MonetDB/issues/1124/comments | 0 | 2020-11-30T09:22:07Z | 2024-06-27T11:21:03Z | https://github.com/MonetDB/MonetDB/issues/1124 | 753,305,478 | 1,124 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-07 01:07:06 +0100
From: @drstmane
To: @sjoerdmullender
Version: -- development
Last updated: 2006-11-07 02:22:25 +0100
## Comment 6298
Date: 2006-11-07 13:07:06 +0100
From: @drstmane
After yesterday's (Monday, Nov 06 2006) checkins,
various Pathfinder tests fail in case of 64-bit OIDs with
!ERROR: interpret: unknown variable 'logger_bats'.
I'm not sure, why/how the problem is related to 64-bit
OIDs, but from experinces, I guess is *might* be
related to the with 64-bit OIDs document storage is
(thought to be) larger, and hence, clean-up of cached
documents kicks in, while it doesn't in case of 32-bit
OIDs (in that case, the error *might* be triggered in
pathfinder.mx's collection_cleanup() proc).
examples of failing tests (basically all tests that did
work yesterday, but fail today; see the respective
testing result email):
almost all MBench tests, e.g.,
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Darwin8.8.0/benchmarks_MBench/count_nodes.out.00.html
various BugTracker tests, e.g.,
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Debian3.1/tests_BugTracker/if-condition_testing_empty_sequence.SF-1589820.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d-Debian3.1/tests_BugTracker/collection-naming.SF-1590782.out.00.html
## Comment 6299
Date: 2006-11-07 14:22:25 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Fixed by moving the declaration of logger_bats further up so
that it is defined before the first possible call to
cleanup_collection.
## Comment 6300
Date: 2006-11-07 16:02:29 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
No test needed since existing tests found this.
## Comment 6301
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1591958 at http://sourceforge.net/support/tracker.php?aid=1591958
| XQ: !ERROR: interpret: unknown variable 'logger_bats'. | https://api.github.com/repos/MonetDB/MonetDB/issues/1123/comments | 0 | 2020-11-30T09:22:03Z | 2024-06-27T11:21:02Z | https://github.com/MonetDB/MonetDB/issues/1123 | 753,305,430 | 1,123 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-07 08:13:39 +0100
From: @drstmane
To: Manuel Mayr <<mayrm>>
Version: -- development
Last updated: 2006-11-07 11:40:39 +0100
## Comment 6291
Date: 2006-11-07 08:13:39 +0100
From: @drstmane
When configured with "--enable-optimize", Pathfinder's
new SQL generation part fails to compile with gcc:
========
gcc -DHAVE_CONFIG_H -I.
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/pathfinder/compiler/sql
-I../..
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/pathfinder/compiler/sql
-I../include
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/pathfinder/compiler/sql/../include
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/MonetDB/.GNU.32.32.d-Fedora4/include/MonetDB
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/MonetDB/.GNU.32.32.d-Fedora4/include/MonetDB/common
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/MonetDB/.GNU.32.32.d-Fedora4/include/MonetDB/gdk
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/MonetDB/.GNU.32.32.d-Fedora4/include/MonetDB/monet
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/MonetDB/.GNU.32.32.d-Fedora4/include/MonetDB/plain
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/MonetDB/.GNU.32.32.d-Fedora4/include/MonetDB/contrib
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/MonetDB/.GNU.32.32.d-Fedora4/include/MonetDB/mapi
-I/export/scratch1/monet/monet.GNU.32.32.d.31844/MonetDB/.GNU.32.32.d-Fedora4/include/MonetDB/C
-DLIBSQL -O2 -std=c99 -Wall -W -O6 -fomit-frame-pointer
-finline-functions -falign-loops=4 -falign-jumps=4
-falign-functions=4 -fexpensive-optimizations
-funroll-loops -frerun-cse-after-loop -frerun-loop-opt
-DHWCOUNTERS -DHW_Linux -DHW_i686
-Werror-implicit-function-declaration -Werror
-Wno-unused-function -Wno-unused-label -c
/export/scratch1/monet/monet.GNU.32.32.d.31844/pathfinder/compiler/sql/sql.c
-fPIC -DPIC -o .libs/libsql_la-sql.o
cc1: warnings being treated as errors
/export/scratch1/monet/monet.GNU.32.32.d.31844/pathfinder/compiler/sql/sql.c:
In function 'PFsql_alg_schmitm_conv':
/export/scratch1/monet/monet.GNU.32.32.d.31844/pathfinder/compiler/sql/sql.c:178:
warning: 'ret.type' is used uninitialized in this function
make[5]: *** [libsql_la-sql.lo] Error 1
========
For details, see (e.g.)
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.GNU.64.32.d-Fedora4/make-k.out
## Comment 6292
Date: 2006-11-07 11:27:53 +0100
From: Manuel Mayr <<mayrm>>
Logged In: YES
user_id=1629542
I fixed this, and I will speak with Jens because my compiler
accepts the code,
and i don't see warnings like this one. Maybe he knows some
compiler flags oder another
compiler that helps me to avoid such failures.
I work with gcc:
Using built-in specs.
Target: i686-apple-darwin8
Configured with:
/private/var/tmp/gcc/gcc-5250.obj~20/src/configure
--disable-checking -enable-werror --prefix=/usr
--mandir=/share/man --enable-languages=c,objc,c++,obj-c++
--program-transform-name=/^[cg][^.-]*$/s/$/-4.0/
--with-gxx-include-dir=/include/c++/4.0.0
--build=powerpc-apple-darwin8 --with-arch=pentium-m
--with-tune=prescott --program-prefix=
--host=i686-apple-darwin8 --target=i686-apple-darwin8
Thread model: posix
gcc version 4.0.1 (Apple Computer, Inc. build 5250)
maybe someone know how I can configure it, to see this
errors. I'm thankful for any hints.
Best regards,
Manuel
## Comment 6293
Date: 2006-11-07 11:38:51 +0100
From: @drstmane
Logged In: YES
user_id=572415
Fixed by Manuel in CVS.
Thanks!
Manuel,
as Sjoerd in me already explained in the developers list,
this is an icc specific problem, hence, you only have to
"worry" about it, once nightly testing tells you so (or
someone you just happend to use a stricter compiler that you
did.
We are more than happy, if you (a) ensure that your code
compiles (at least) with in your favorite environment before
you check it in, (b) take care of fixing problems that pop
up during nightly testing (i.e., read the respective mails
and/or check the testweb the next morning), take care of bug
reports assigned to you in a timely manner.
So far, you did great on all three points, hence, nothing to
"worry" about!
## Comment 6294
Date: 2006-11-07 11:39:25 +0100
From: @grobian
Logged In: YES
user_id=963970
You could try to update your compiler to release 5341, but I
think you will get them if you configure pathfinder like
Stefan suggested. How do you invoke configure?
## Comment 6295
Date: 2006-11-07 11:40:39 +0100
From: @drstmane
Logged In: YES
user_id=572415
re-closed after concurrent update conflicts
## Comment 6296
Date: 2006-11-07 11:40:49 +0100
From: @grobian
Logged In: YES
user_id=963970
SourceForge doesn't deal very well with concurrency it seems.
## Comment 6297
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1591831 at http://sourceforge.net/support/tracker.php?aid=1591831
| PF/SQL fails to compile with gcc + optimization | https://api.github.com/repos/MonetDB/MonetDB/issues/1122/comments | 0 | 2020-11-30T09:21:59Z | 2024-06-27T11:21:01Z | https://github.com/MonetDB/MonetDB/issues/1122 | 753,305,375 | 1,122 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-06 05:38:17 +0100
From: @sjoerdmullender
To: Manuel Mayr <<mayrm>>
Version: -- development
Last updated: 2006-11-06 10:51:27 +0100
## Comment 6286
Date: 2006-11-06 17:38:17 +0100
From: @sjoerdmullender
/ufs/sjoerd/src/MonetDB/devel/pathfinder/compiler/compile.c:149:
error: unknown field 'generate_sql' specified in
initializer
It seems to me a change in include/pathfinder.h needs
to be checked in.
## Comment 6287
Date: 2006-11-06 17:43:27 +0100
From: @drstmane
Logged In: YES
user_id=572415
in addition to the above, Intel's icc also complains about:
/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/pathfinder/compiler/sql/sql.c(153):
error 181: argument is incompatible with corresponding
format string conversion
printf( "att = %d\n", att );
^
/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/pathfinder/compiler/sql/sql.c(199):
error 593: variable "test" was set but never used
PFsql_schema_item_t test;
^
compilation aborted for
/net/corona.ins.cwi.nl/export/scratch0/manegold/Monet/Testing/Current/pathfinder/compiler/sql/sql.c
(code 2)
## Comment 6288
Date: 2006-11-06 19:57:09 +0100
From: Manuel Mayr <<mayrm>>
Logged In: YES
user_id=1629542
Pathfinder should compile now. I had problems to compile
since the last checkout
due to the last changes in MonetDB.
regards,
Manuel
## Comment 6289
Date: 2006-11-06 22:51:27 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
The problem seems to be fixed.
## Comment 6290
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1591467 at http://sourceforge.net/support/tracker.php?aid=1591467
| XQ: Pathfinder won't compile | https://api.github.com/repos/MonetDB/MonetDB/issues/1121/comments | 0 | 2020-11-30T09:21:56Z | 2024-06-27T11:21:01Z | https://github.com/MonetDB/MonetDB/issues/1121 | 753,305,343 | 1,121 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-06 10:33:56 +0100
From: @drstmane
To: @sjoerdmullender
Version: -- development
Last updated: 2006-11-06 05:36:27 +0100
## Comment 6282
Date: 2006-11-06 10:33:56 +0100
From: @drstmane
With the THRprintf changes in GDK, all(?) error
messages seems to go to stdout, now as opposed to stderr.
In particular, the correct/expected FATAL errors in tests
tests/BugDay_2005-10-06_4.8/Tests/multiple_servers.SF-914113.MAPI
!FATAL: GDKlockHome: Database lock '.gdk_lock' denied
tests/BugDay_2005-10-06_4.9.3/Tests/monet_daemon.SF-623743
!FATAL: scheduleClient: Daemon mode without startup script
now go to stdout iso. stderr, causing these tests to
"fail".
In case these changes were intended like this, please
approve the output accordingly.
In case these changes were not intnded like this please
fix the respective bug.
## Comment 6283
Date: 2006-11-06 17:36:27 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Fixed by printing again to stderr in GDKfatal.
## Comment 6284
Date: 2006-11-07 16:03:21 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
No test needed.
## Comment 6285
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1591242 at http://sourceforge.net/support/tracker.php?aid=1591242
| MonetDB: FATAL error message to stdout: BUG or Feature? | https://api.github.com/repos/MonetDB/MonetDB/issues/1120/comments | 0 | 2020-11-30T09:21:53Z | 2024-06-27T11:21:00Z | https://github.com/MonetDB/MonetDB/issues/1120 | 753,305,303 | 1,120 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-05 10:58:59 +0100
From: @drstmane
To: XQuery devs <<bugs-xquery>>
Version: -- development
Last updated: 2006-12-17 05:41:47 +0100
## Comment 6278
Date: 2006-11-05 22:58:59 +0100
From: @drstmane
The following error occur with XQuery tests when run
with debugmask 0 (i.e., w/o property checking) which is
the default in the code:
!ERROR: interpret: no matching MIL operator to
'leftfetchjoin(BAT[oid,void], BAT[oid,oid])'.
occurs since Saturday Oct 21 2006, i.e., after/due to
changes made between Sunday Oct 15 & Fridaay Oct 20
2006; debugmask 0 testing only runs during weekends
with tests
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_Music/limit_04.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_Music/limit_05.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_Music_mod_MIL/limit_04.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_Music_mod_MIL/limit_05.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_Music_mod_XQ/limit_04.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_Music_mod_XQ/limit_05.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_Music_opt_MIL/limit_04.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_Music_opt_MIL/limit_05.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_Music_opt_XQ/limit_04.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_Music_opt_XQ/limit_05.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-OK.SF-1562868.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_BugTracker/complex-XQuery-batbat_lng_add_inplace-error-KO.SF-1562868.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q01x.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q03.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q04.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q05o.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q05u.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q10.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q11.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q12.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q12x.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q15.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_W3C_use_cases_XQ_R/Q16.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_WebSite/voc2.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_WebSite/voc3.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_WebSite/voc5.err.00.html
!ERROR: merged_union: tail of second BAT must be sorted.
with tests
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_XMark_alg/q09.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/benchmarks_XMark_alg/q10.out.00.html
!ERROR: merged_union: tail of first BAT must be sorted.
with tests
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/runtime/merged_union2.out.00.html
different attribute order in XML result
with tests
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.64.d-Fedora4/tests_StandOff_video/artists.out.00.html
## Comment 6279
Date: 2006-11-11 15:35:23 +0100
From: @drstmane
Logged In: YES
user_id=572415
see also
http://sourceforge.net/tracker/index.php?func=detail&aid=1591006&group_id=56967&atid=482468
## Comment 6280
Date: 2006-12-17 17:41:47 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
all above links seems to work and my own tests (always with -d0) do not show these problems either anymore.
## Comment 6281
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1591024 at http://sourceforge.net/support/tracker.php?aid=1591024
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-xquery@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| XQ: tests fail with debugmask 0 (i.e. w/o property checking) | https://api.github.com/repos/MonetDB/MonetDB/issues/1119/comments | 0 | 2020-11-30T09:21:50Z | 2024-06-27T11:20:59Z | https://github.com/MonetDB/MonetDB/issues/1119 | 753,305,269 | 1,119 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-05 10:29:22 +0100
From: @drstmane
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-12-14 06:59:24 +0100
## Comment 6272
Date: 2006-11-05 22:29:22 +0100
From: @drstmane
With 32-bit OID (on both 32-bit and 64-bit systems) test
tests/W3C_use_cases/XQUF/AddressBook/Tests/Q1x.xq
fails with
ERROR = !WARNING: BATsetaccess: tmp_1131 has 1 views;
deliver a copy.
cf.
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.32.d-Fedora4/tests_W3C_use_cases_XQUF_AddressBook/Q1x.err.00.html
## Comment 6273
Date: 2006-11-12 12:41:14 +0100
From: @drstmane
Logged In: YES
user_id=572415
for the test results with the latest "Stable" release branch
(via `make check` with debugmask 10) see
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/pathfinder/.mTests103/GNU.64.32.d-Fedora4/tests_W3C_use_cases_XQUF_AddressBook/Q1x.err.00.html
for the test results with the latest "Current" development
trunk (via `Mtest.py -r -d0` with debugmask 0) see
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests03/GNU.64.32.d-Fedora4/tests_W3C_use_cases_XQUF_AddressBook/Q1x.err.00.html
## Comment 6274
Date: 2006-11-13 17:11:22 +0100
From: @drstmane
Logged In: YES
user_id=572415
*might* be related to
[ 1595701 ] XQ: "replace value" concatenates value iso.
replacing it
http://sourceforge.net/tracker/index.php?func=detail&aid=1595701&group_id=56967&atid=482468
## Comment 6275
Date: 2006-12-14 18:59:24 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
fixed (made a copy in attribute action reorder block)
## Comment 6276
Date: 2006-12-14 19:04:09 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Thanks!
... any idea why it occured with 32-bit OIDs, only?
## Comment 6277
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1591013 at http://sourceforge.net/support/tracker.php?aid=1591013
| XQ: BATsetaccess warning/error with 32-bit OIDs | https://api.github.com/repos/MonetDB/MonetDB/issues/1118/comments | 0 | 2020-11-30T09:21:47Z | 2024-06-27T11:20:58Z | https://github.com/MonetDB/MonetDB/issues/1118 | 753,305,240 | 1,118 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-05 10:11:50 +0100
From: @drstmane
To: Stefan Manegold <<Stefan.Manegold>>
Version: 11.3.1 (Apr2011) [obsolete]
CC: @sjoerdmullender
Last updated: 2011-05-03 10:51:20 +0200
## Comment 6265
Date: 2006-11-05 22:11:50 +0100
From: @drstmane
The following tests produce different output with
debugmask 0 (no property checking) than with debugmask
(property checking enabled; default with Mtest to
detect wrongly set properties):
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB/.mTests03/GNU.64.64.d-Fedora4/src_gdk/void.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB/.mTests03/GNU.64.64.d-Fedora4/src_gdk/TMsubcommit.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB/.mTests03/GNU.64.64.d-Fedora4/src_modules_plain/xtables.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB/.mTests03/GNU.64.64.d-Fedora4/src_modules_plain/radix.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB/.mTests03/GNU.64.64.d-Fedora4/tests_BugReports/no.148.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB/.mTests03/GNU.64.64.d-Fedora4/tests_from_Twente_moa/test.out.00.html
The differences are related to properties that "happen
to" hold, but are not set unless (exhaustive) property
checking is enabled.
In these tests, most of the differences are probably
harmless, however, this needs to be checked/verified
and the respective code (or the tests) should be
modified to yield the same output with both debugmask 0
& debugmask 10.
## Comment 6266
Date: 2006-11-11 15:35:12 +0100
From: @drstmane
Logged In: YES
user_id=572415
see also
http://sourceforge.net/tracker/index.php?func=detail&aid=1591006&group_id=56967&atid=482468
## Comment 6267
Date: 2006-11-11 15:47:12 +0100
From: @drstmane
Logged In: YES
user_id=572415
There are IMHO two ways to reduce the differences in
MonetDB's behaviour between -d0 & -d10
1)
make the interpreter call a "light" version of BATpropcheck
(also) in case -d10 is not set; this "light" version
*quickly* checks/sets the properties in "trivial" basically
for BATs with at most one BUN.
2)
keep BATpropcheck from (silently!) setting properties that
do hold but have not been set, yet; i.e., make BATpropcheck
*check* for wrongly set properties only.
this would require a way to (re-)enable the (silent) setting
of properties in cases where this (extra) feature of the
current BATpropcheck implementation is currently exploited
(basically join & outerjoin in gdk_relop.mx)
Note:
with -d8388608 (=2^23=8*1024*1024) BATpropcheck issues a
warning whenever it set a property that does nhold but had
not been set, yet.
## Comment 6268
Date: 2008-07-24 07:55:14 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
New URLs of tests "failing" (i.e., showing differences) without property checking (--debug=0) that work fine with property checking (--debug=10):
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB4/.mTests103/GNU.64.64.d.0-Fedora8/src_modules_plain/xtables.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB4/.mTests103/GNU.64.64.d.0-Fedora8/tests_BugReports/no.148.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB4/.mTests103/GNU.64.64.d.0-Fedora8/tests_from_Twente_moa/list.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB4/.mTests103/GNU.64.64.d.0-Fedora8/tests_from_Twente_moa/test.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/sql/.mTests4103/GNU.64.64.d.0-Fedora8/src_test_Dependencies/Dependencies.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/geom/.mTests4103/GNU.64.64.d.0-Fedora8/src_sql/geom_mbr.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/geom/.mTests4103/GNU.64.64.d.0-Fedora8/src_sql/geom_mbr.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/geom/.mTests5103/GNU.64.64.d.0-Fedora8/src_sql/geom_mbr.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/geom/.mTests5103/GNU.64.64.d.0-Fedora8/src_sql/geom_mbr.err.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d.0-Fedora8/tests_StandOff_video/artists.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora8/runtime/merged_union2.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora8/tests_BugTracker/immune_for_updates.SF-1981852.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora8/tests_StandOff_video/artists.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora8/tests_XRpc/echoAll.out.00.html
## Comment 6269
Date: 2008-10-26 19:06:00 +0100
From: @mlkersten
oct 2008. The difference is not visible from the M5 layers.
## Comment 6270
Date: 2008-10-30 15:59:24 +0100
From: @drstmane
Update:
Some tests "now" seems to produce the same result without property checking (--debug=0) as they do with property checking (--debug=10):
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB4/.mTests103/GNU.64.64.d.0-Fedora8/tests_from_Twente_moa/list.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/geom/.mTests4103/GNU.64.64.d.0-Fedora8/src_sql/geom_mbr.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Stable/geom/.mTests4103/GNU.64.64.d.0-Fedora8/src_sql/geom_mbr.err.00.html
The tests that used to show differences with the M5 back-end currently do not work properly due to other reasons:
http://monetdb.cwi.nl/testing/projects/monetdb/Current/geom/.mTests5103/GNU.64.64.d.0-Fedora8/src_sql/geom_mbr.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/geom/.mTests5103/GNU.64.64.d.0-Fedora8/src_sql/geom_mbr.err.00.html
The remaining M4 tests are not tested with M5:
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB4/.mTests103/GNU.64.64.d.0-Fedora8/src_modules_plain/xtables.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB4/.mTests103/GNU.64.64.d.0-Fedora8/tests_BugReports/no.148.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB4/.mTests103/GNU.64.64.d.0-Fedora8/tests_from_Twente_moa/list.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/MonetDB4/.mTests103/GNU.64.64.d.0-Fedora8/tests_from_Twente_moa/test.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTests103/GNU.64.64.d.0-Fedora8/tests_StandOff_video/artists.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora8/runtime/merged_union2.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora8/tests_BugTracker/immune_for_updates.SF-1981852.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora8/tests_StandOff_video/artists.out.00.html
http://monetdb.cwi.nl/testing/projects/monetdb/Current/pathfinder/.mTestsg103/GNU.64.64.d.0-Fedora8/tests_XRpc/echoAll.out.00.html
## Comment 6271
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1591006 at http://sourceforge.net/support/tracker.php?aid=1591006
## Comment 15666
Date: 2011-03-28 17:34:57 +0200
From: @sjoerdmullender
The Mar2011 version has been released.
## Comment 15784
Date: 2011-05-03 10:51:20 +0200
From: @drstmane
The MonetDB5, SQL, & Geom tests don't show any differences between debugmask 0 & 10 with the Apr2011 release anymore.
| MonetDB: test output differs between debugmask 0 and 10 | https://api.github.com/repos/MonetDB/MonetDB/issues/1117/comments | 0 | 2020-11-30T09:21:43Z | 2024-06-27T11:20:57Z | https://github.com/MonetDB/MonetDB/issues/1117 | 753,305,194 | 1,117 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-05 07:33:28 +0100
From: @drstmane
To: Stefan Manegold <<Stefan.Manegold>>
Version: -- development
Last updated: 2007-11-19 05:47:43 +0100
## Comment 6254
Date: 2006-11-05 19:33:28 +0100
From: @drstmane
All closed bug reports with ID >= 1400000
http://sourceforge.net/tracker/index.php?func=browse&group_id=56967&atid=482468&set=custom&_assigned_to=0&_status=2&_category=100&_group=100&order=artifact_id&sort=DESC&offset=0
as well as
all pending bug reports
http://sourceforge.net/tracker/index.php?func=browse&group_id=56967&atid=482468&set=custom&_assigned_to=0&_status=4&_category=100&_group=100&order=artifact_id&sort=DESC&offset=0
need to be checked, whether a respective test script
has been added to CVS, and if not, a test need to be added.
As of now, we should ensure, that no bug reports gets
closed before a respective test scipt has been added.
[Exceptions: bug reports about existing tests,
compilation problems, documentation, web site, and alike.]
## Comment 6255
Date: 2006-11-05 20:01:54 +0100
From: @drstmane
Logged In: YES
user_id=572415
closed bug reports with IDs 1400000 - 1410000 have been
checked by me;
respective test scrips are present where applicable.
## Comment 6256
Date: 2006-11-06 00:47:44 +0100
From: @drstmane
Logged In: YES
user_id=572415
closed bug reports with IDs 1410000 - 1430000 are checked as
well.
## Comment 6257
Date: 2006-11-29 10:26:37 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
closed bug reports with IDs 1430000 - 1440000 are checked as well.
## Comment 6258
Date: 2006-11-29 11:14:43 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Closed bug reports with IDs 1440000 - 1500000 are checked as well.
This leaves closed bug reports with IDs > 1500000 (as well as all pending bug reports) to be checked.
## Comment 6259
Date: 2006-12-02 18:56:01 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Closed bug reports with IDs 1500000 - 1550000 are checked as well.
## Comment 6260
Date: 2006-12-02 19:09:43 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
Closed bug reports with IDs 1550000 - 1560000 are checked as well.
## Comment 6261
Date: 2006-12-02 21:06:18 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
All remaining (currently) closed bug reports (IDs 1560000 - 1601761) have finally been checked for the existence of respective tests (where applicable).
Please help to ensure that as of now, bug reports are only closed, once a proper test script has been added to CVS (or an "excuse" is given, why a test is not necessary/possible).
We should still check all "Pending", whether we could add the respective tests...
## Comment 6262
Date: 2007-06-05 22:46:32 +0200
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
lowered priority or meta/default bug reports
## Comment 6263
Date: 2007-11-19 17:47:43 +0100
From: @drstmane
Logged In: YES
user_id=572415
Originator: YES
The policy of closing bug reports only if there is a proper test for them (or having a proper "excuse" why no test is needed/possible) seems to work reasonably by now. Hence, we can close this meta bug report.
## Comment 6264
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1590930 at http://sourceforge.net/support/tracker.php?aid=1590930
| missing test scripts | https://api.github.com/repos/MonetDB/MonetDB/issues/1116/comments | 0 | 2020-11-30T09:21:40Z | 2024-06-28T13:31:59Z | https://github.com/MonetDB/MonetDB/issues/1116 | 753,305,165 | 1,116 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-05 01:25:06 +0100
From: @drstmane
To: @njnes
Version: -- development
Last updated: 2006-11-06 08:09:58 +0100
## Comment 6251
Date: 2006-11-05 13:25:06 +0100
From: @drstmane
Since this checkin
===================================================================
2006/10/20 - nielsnes: src/include/Makefile.ag,1.11
src/server/sql_mvc.mx,1.161
src/storage/Makefile.ag,1.7
src/storage/sql_storage.h,1.27
src/storage/bat/bat_logger.mx,1.41
src/storage/bat/bat_sequence.mx,1.13
src/storage/bat/bat_store.mx,1.106
src/test/BugDay_2005-10-06_2.9.3/Tests/bad_sql_logdir.SF-927371,1.3
src/test/BugDay_2005-10-06_2.9.3/Tests/bad_sql_logdir.SF-927371.bat,1.3
src/test/BugDay_2005-10-06_2.9.3/Tests/bad_sql_logdir.SF-927371.stable.err,1.3
src/test/BugDay_2005-10-06_2.9.3/Tests/bad_sql_logdir.SF-927371.stable.out,1.8
src/test/BugDay_2005-12-19_2.9.3/Tests/select_from_env.SF-1240701.1242164.stable.out,1.4
more cleanup of the logger.
Logger is now a global variable. It should now be
possible to move it into
a M4/5 module and reuse it in Xquery.
one big change is the name of the GDKenv variable
sql_logdir into logdir.
===================================================================
rpmbuild for SQL fails with
========
Processing files: MonetDB-SQL-devel-2.13.1-1_Fedora4.oid64
error: File not found by glob:
/var/tmp/MonetDB-SQL-root/usr/include/sql/include/*.h
error: File not found by glob:
/var/tmp/MonetDB-SQL-root/usr/include/sql/storage/*.h
RPM build errors:
File not found by glob:
/var/tmp/MonetDB-SQL-root/usr/include/sql/include/*.h
File not found by glob:
/var/tmp/MonetDB-SQL-root/usr/include/sql/storage/*.h
make: *** [rpm] Error 1
========
basically, because the respective header files are no
longer installed after above checkin.
BUG? => fix Makefiles to in the header files again.
FEATURE? => fix MonetDB-SQL.spec.in to not require the
header files any more.
## Comment 6252
Date: 2006-11-06 08:09:58 +0100
From: @drstmane
Logged In: YES
user_id=572415
fixed by Niels.
Thanks!
## Comment 6253
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1590786 at http://sourceforge.net/support/tracker.php?aid=1590786
| SQL: rpmbuild fails | https://api.github.com/repos/MonetDB/MonetDB/issues/1115/comments | 0 | 2020-11-30T09:21:36Z | 2024-06-27T11:20:55Z | https://github.com/MonetDB/MonetDB/issues/1115 | 753,305,119 | 1,115 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-05 01:17:26 +0100
From: Wouter Alink <<alink>>
To: Peter Boncz <<boncz>>
Version: -- development
Last updated: 2006-12-02 12:54:27 +0100
## Comment 6247
Date: 2006-11-05 13:17:26 +0100
From: Wouter Alink <<alink>>
I don't know whether the following is expected
behaviour, but I found it confusing, therefore this
bugreport.
mil>shred_doc("/.../test.xml","mies.xml","aap",50LL);
works fine, but then:
mil>shred_doc("/.../test.xml","mies.xml","aap2",50LL);
fails with:
ERROR = !ERROR: shred_doc: document mies.xml alreadt
exists in collection aap (1 such errors)!
afterwards trying:
mil>shred_doc("/.../test.xml","noot.xml","aap2",50LL);
works fine, but again:
mil>shred_doc("/.../test.xml","noot.xml","aap",50LL);
fails with:
ERROR = !ERROR: shred_doc: document mies.xml already
exists in collection aap2 (1 such errors)!
## Comment 6248
Date: 2006-11-05 19:23:46 +0100
From: @drstmane
Logged In: YES
user_id=572415
added test in
pathfinder/tests/BugTracker/Tests/collection-naming.SF-1590782.milS
IMHO the error messages are correct/expected, as (in
MonetDB/XQuery) document names
need to be globally unique (not only per collection) as we
allow to use fn:doc()
also to access documents inside collections by only
specifying the document name.
Hence, I approved the error messages as correct.
Peter,
could you please confirm?
## Comment 6249
Date: 2006-12-02 00:54:27 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
yes, this is correct
## Comment 6250
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1590782 at http://sourceforge.net/support/tracker.php?aid=1590782
| XQ: collection-naming on smallest substring | https://api.github.com/repos/MonetDB/MonetDB/issues/1114/comments | 0 | 2020-11-30T09:21:33Z | 2024-06-28T13:31:58Z | https://github.com/MonetDB/MonetDB/issues/1114 | 753,305,084 | 1,114 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-05 12:58:20 +0100
From: Wouter Alink <<alink>>
To: XQuery devs <<bugs-xquery>>
Version: -- development
Last updated: 2007-05-25 11:56:22 +0200
## Comment 6239
Date: 2006-11-05 12:58:20 +0100
From: Wouter Alink <<alink>>
Using the interactive XQ mode, the query <a/> returns:
a/>: cannot open
## Comment 6240
Date: 2006-11-05 13:07:28 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
How *exactly* did you run your query?
Obviously, you need to quote < and > to the shell, so
MapiClient -lx -s <a/>
is *not* going to work (and it is not a bug that it doesn't).
What does work for me is:
MapiClient -lx -s '<a/>'
## Comment 6241
Date: 2006-11-05 13:32:07 +0100
From: Wouter Alink <<alink>>
Logged In: YES
user_id=621590
in the bug-report i stated the 'interactive' mode...
MapiClient -lxq -odm
and then on the 'xquery>' command line
<a/>
Ctrl-D
## Comment 6242
Date: 2006-11-05 13:49:54 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Ah. That's the redirect feature of MapiClient. Type ? on a
line by itself and you see the "<file - read input from
file" command.
Perhaps we need to sanitize these options in MapiClient.
## Comment 6243
Date: 2006-11-05 14:44:39 +0100
From: Wouter Alink <<alink>>
Logged In: YES
user_id=621590
i expected something like that...
it is probably hard for the MapiClient to distinguish
between the redirect-command '<' and a valid xquery. perhaps
the redirect command for the XQ-console could be changed to
"<<" or some other invalid xquery? Or maybe all MapiClient
specific operations could be prefixed by a '' or something
similar?
By the way, if you look at the output of '?' there are a lot
of SQL-specific commands, that do not apply to MIL or XQuery.
## Comment 6244
Date: 2006-11-05 14:57:17 +0100
From: @mlkersten
Logged In: YES
user_id=490798
The MapiClient interactive commands have been introduced for
convenience. Just chancing their prompt does not solve your
problem.
Instead, we might consider a language dependent set of
commands. This would mean that -lmal and -lsql could remain
mostly intact.
Its up to the Xquery folks to decide on a minimal, yet
convenient set of command line directives.
## Comment 6245
Date: 2007-05-25 23:56:22 +0200
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
fixed
## Comment 6246
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1590774 at http://sourceforge.net/support/tracker.php?aid=1590774
The original assignee of this bug does not have
an account here. Reassigning to the default assignee
for the component, bugs-xquery@monetdb.org.
Previous assignee was nobody@users.sourceforge.net.
| XQ: <a/> interpreted as pipe | https://api.github.com/repos/MonetDB/MonetDB/issues/1113/comments | 0 | 2020-11-30T09:21:30Z | 2024-06-27T11:20:53Z | https://github.com/MonetDB/MonetDB/issues/1113 | 753,305,047 | 1,113 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-04 08:07:49 +0100
From: Wouter Alink <<alink>>
To: @sjoerdmullender
Version: -- development
Last updated: 2006-11-16 06:54:59 +0100
## Comment 6234
Date: 2006-11-04 20:07:49 +0100
From: Wouter Alink <<alink>>
(perhaps related to http://sourceforge.net/support/tracker.php?aid=1590580)
given the document:
<aap>
Hello World
</aap>
shredded with:
shred_doc("/ufs/alink/test.xml","test.xml","",50LL);
the first query :
do insert <noot>Bes</noot>
as first into exactly-one(doc("test.xml")/aap)
executed fine, and resulted in
<aap><noot>Bes</noot>
Hello World
</aap>
then the following query:
let $batch := (<beer/>,<mies/>)
return
do insert $batch
after exactly-one(doc("test.xml")//noot))
also executed fine.
But when looking at the result, something strange happened:
<aap><noot>Bes<mies/></noot><beer/>
Hello World
</aap>
## Comment 6235
Date: 2006-11-05 19:03:01 +0100
From: @drstmane
Logged In: YES
user_id=572415
test added in
pathfinder/tests/BugTracker/Tests/inserting_multiple_elements.SF-1590583.milS
fails as reported
## Comment 6236
Date: 2006-11-16 14:58:25 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
Sjoerd, can you still find some pre-holiday time to look at this?
## Comment 6237
Date: 2006-11-16 18:54:59 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Fixed.
## Comment 6238
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1590583 at http://sourceforge.net/support/tracker.php?aid=1590583
| XQ: incorrectly inserting multiple elements | https://api.github.com/repos/MonetDB/MonetDB/issues/1112/comments | 0 | 2020-11-30T09:21:27Z | 2024-06-27T11:20:52Z | https://github.com/MonetDB/MonetDB/issues/1112 | 753,304,994 | 1,112 |
[
"MonetDB",
"MonetDB"
] | Date: 2006-11-04 07:52:13 +0100
From: Wouter Alink <<alink>>
To: @sjoerdmullender
Version: -- development
Last updated: 2006-11-16 06:54:44 +0100
## Comment 6229
Date: 2006-11-04 19:52:13 +0100
From: Wouter Alink <<alink>>
excuse me for keep submitting bugs, but i'm having fun
with XUF at the moment (so expect more bug reports :))
given the document:
<aap>
<noot/>
Hello World
</aap>
The following query:
let $batch := (<beer/>,<mies/>)
return
do insert $batch
after exactly-one(doc("test.xml")//noot))
returns:
MAPI = monetdb@localhost:50000
QUERY = let $batch := (<beer/>,<mies/>)
ERROR = !ERROR: merged_union: BAT 8 (2 BUNs) must have
the same size as BAT 2 (1 BUNs).
!ERROR: CMDmerged_union: operation failed
## Comment 6230
Date: 2006-11-05 18:53:32 +0100
From: @drstmane
Logged In: YES
user_id=572415
added test in
pathfinder/tests/BugTracker/Tests/insert_multiple.SF-1590580.xq
fails as reported
[ no "excuse" for submitting bug report --- any testing work
is highly appreciated!! ]
[ adding the respective tests script in CVS (if possible)
would be even more appreciated ... ;-)) ]
## Comment 6231
Date: 2006-11-16 14:57:57 +0100
From: @peterboncz
Logged In: YES
user_id=591107
Originator: NO
Sjoerd, can you still find some pre-holiday time to look at this?
## Comment 6232
Date: 2006-11-16 18:54:44 +0100
From: @sjoerdmullender
Logged In: YES
user_id=43607
Originator: NO
Fixed.
## Comment 6233
Date: 2010-05-04 09:32:09 +0200
From: Pseudo user for Sourceforge import <<sfimport>>
This bug was previously known as tracker item 1590580 at http://sourceforge.net/support/tracker.php?aid=1590580
| XQ: insert multiple | https://api.github.com/repos/MonetDB/MonetDB/issues/1111/comments | 0 | 2020-11-30T09:21:24Z | 2024-06-27T11:20:51Z | https://github.com/MonetDB/MonetDB/issues/1111 | 753,304,952 | 1,111 |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.