|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
import os |
|
|
import sys |
|
|
import llvm_helper |
|
|
import json |
|
|
import subprocess |
|
|
import dateparser |
|
|
from datetime import timezone |
|
|
|
|
|
bisect_runner_file = os.path.join( |
|
|
os.path.dirname(os.path.abspath(__file__)), "bisect_runner.py" |
|
|
) |
|
|
|
|
|
|
|
|
def is_llvm_functional_change(commit: str) -> bool: |
|
|
commit = commit.strip() |
|
|
changed_files = llvm_helper.git_execute( |
|
|
[ |
|
|
"show", |
|
|
"--name-only", |
|
|
"--format=", |
|
|
commit, |
|
|
"--", |
|
|
"llvm/lib/*", |
|
|
"llvm/include/*", |
|
|
] |
|
|
).strip() |
|
|
return changed_files != "" |
|
|
|
|
|
|
|
|
def bisect_issue(issue): |
|
|
path = os.path.join(llvm_helper.dataset_dir, issue) |
|
|
with open(path) as f: |
|
|
data = json.load(f) |
|
|
if data["bug_type"] == "hang": |
|
|
|
|
|
return |
|
|
if "bisect" in data: |
|
|
return |
|
|
commit = data["bisect"] |
|
|
try: |
|
|
commit_parsed = llvm_helper.git_execute(["rev-parse", commit]).strip() |
|
|
if commit_parsed == commit: |
|
|
return |
|
|
except subprocess.CalledProcessError: |
|
|
print(f"Previous bisect commit({commit}) is invalid. Redo bisect...") |
|
|
pass |
|
|
print(data["issue"]["title"]) |
|
|
try: |
|
|
base_commit = data["base_commit"] |
|
|
base_commit_time = dateparser.parse( |
|
|
llvm_helper.git_execute(["show", "-s", "--format=%ci", base_commit]).strip() |
|
|
).astimezone(timezone.utc) |
|
|
report_time = dateparser.parse(data["knowledge_cutoff"]).astimezone( |
|
|
timezone.utc |
|
|
) |
|
|
print("Report time:", report_time) |
|
|
print("Base commit time:", base_commit_time) |
|
|
if report_time < base_commit_time: |
|
|
commit_before_report = llvm_helper.git_execute( |
|
|
[ |
|
|
"log", |
|
|
"--before", |
|
|
report_time.isoformat(), |
|
|
"--oneline", |
|
|
"-n", |
|
|
"1", |
|
|
"--no-abbrev", |
|
|
"--format=%H", |
|
|
] |
|
|
).strip() |
|
|
if ( |
|
|
base_commit != commit_before_report |
|
|
and subprocess.run( |
|
|
[bisect_runner_file, path, commit_before_report] |
|
|
).returncode |
|
|
== 1 |
|
|
): |
|
|
base_commit = commit_before_report |
|
|
print("Adjusted base commit to", base_commit) |
|
|
else: |
|
|
print("Use original base commit (unavailable)") |
|
|
else: |
|
|
print("Use original base commit (newer)") |
|
|
|
|
|
good_commit = None |
|
|
offset = 100 |
|
|
while offset <= 204800: |
|
|
commit_sha = llvm_helper.git_execute( |
|
|
["rev-parse", f"{base_commit}~{offset}"] |
|
|
).strip() |
|
|
if subprocess.run([bisect_runner_file, path, commit_sha]).returncode == 0: |
|
|
good_commit = commit_sha |
|
|
break |
|
|
offset = int(offset * 1.6) |
|
|
if good_commit is None: |
|
|
raise RuntimeError("Cannot find a good commit") |
|
|
print("BAD", base_commit, "GOOD", good_commit) |
|
|
llvm_helper.git_execute(["bisect", "reset"]) |
|
|
llvm_helper.git_execute( |
|
|
["bisect", "start", "--no-checkout", base_commit, good_commit] |
|
|
) |
|
|
out = subprocess.check_output( |
|
|
[ |
|
|
"git", |
|
|
"-C", |
|
|
llvm_helper.llvm_dir, |
|
|
"bisect", |
|
|
"run", |
|
|
bisect_runner_file, |
|
|
path, |
|
|
], |
|
|
cwd=llvm_helper.llvm_dir, |
|
|
timeout=600.0, |
|
|
).decode() |
|
|
if not out.endswith("bisect found first bad commit\n"): |
|
|
raise RuntimeError("Bisect failed: " + out) |
|
|
pos = out.rfind(" is the first bad commit\n") |
|
|
if pos == -1: |
|
|
raise RuntimeError("Bisect failed") |
|
|
pos2 = out.rfind("\n", 0, pos) |
|
|
if pos2 == -1: |
|
|
raise RuntimeError("Bisect failed") |
|
|
first_bad = out[pos2 + 1 : pos].strip() |
|
|
print(first_bad) |
|
|
data["bisect"] = first_bad |
|
|
except subprocess.TimeoutExpired: |
|
|
data["bisect"] = "Timeout" |
|
|
print("Timeout") |
|
|
except subprocess.CalledProcessError as e: |
|
|
out = e.stdout.decode() |
|
|
key = "The first bad commit could be any of:\n" |
|
|
pos = out.rfind(key) |
|
|
if pos == -1: |
|
|
return |
|
|
pos2 = out.find("We cannot bisect more!", pos) |
|
|
if pos2 == -1: |
|
|
return |
|
|
candidates = out[pos + len(key) : pos2].strip().splitlines() |
|
|
if len(candidates) > 1: |
|
|
candidates = list(filter(is_llvm_functional_change, candidates)) |
|
|
|
|
|
if len(candidates) == 0: |
|
|
data["bisect"] = "Unrelated" |
|
|
elif len(candidates) == 1: |
|
|
first_bad = candidates[0].strip() |
|
|
print(first_bad) |
|
|
data["bisect"] = first_bad |
|
|
except Exception as e: |
|
|
data["bisect"] = str(e) |
|
|
print(e) |
|
|
with open(path, "w") as f: |
|
|
json.dump(data, f, indent=2) |
|
|
|
|
|
|
|
|
task_list = [] |
|
|
if len(sys.argv) == 2: |
|
|
task_list = [sys.argv[1] + ".json"] |
|
|
else: |
|
|
for name in os.listdir(llvm_helper.dataset_dir): |
|
|
if name.endswith(".json"): |
|
|
task_list.append(name) |
|
|
task_list.sort(reverse=True) |
|
|
|
|
|
for idx, task in enumerate(task_list): |
|
|
print("Bisecting", idx + 1, task.removesuffix(".json")) |
|
|
try: |
|
|
bisect_issue(task) |
|
|
except Exception as e: |
|
|
print(e) |
|
|
pass |
|
|
|