Datasets:
Card: document bug fixes; update counts to 54,803
Browse files
README.md
CHANGED
|
@@ -47,8 +47,8 @@ ds = load_dataset("Frost2o24/bash-instruct-II-55k", split="train")
|
|
| 47 |
|
| 48 |
| Metric | Value |
|
| 49 |
|---|---|
|
| 50 |
-
| Rows | **54,
|
| 51 |
-
| Distinct requests (`variant_group`s) | 45,
|
| 52 |
| `single` / `pipeline` / `script` | **40% / 35% / 25%** (exact) |
|
| 53 |
| Distinct primary utilities | 183 |
|
| 54 |
| `bash -n` valid | **100%** (0 syntax errors) |
|
|
@@ -122,7 +122,7 @@ jq -r '.utility' bash_dataset.jsonl | sort | uniq -c | sort -rn | head
|
|
| 122 |
|
| 123 |
Three layers, in increasing strength:
|
| 124 |
|
| 125 |
-
1. **`bash -n`** on every command — **0 / 54,
|
| 126 |
2. **shellcheck** (warning level) on a large sample — the few findings are style nits
|
| 127 |
(`SC2010` `ls | grep`, `SC2164` `cd` without check).
|
| 128 |
3. **Real execution** of the runnable subset in a disposable Linux sandbox (Debian 13
|
|
@@ -147,6 +147,23 @@ the filter). They were **removed** (55,000 → 54,957). What was cut:
|
|
| 147 |
- `uname java`, `cp --parents f1 f2`, `lspci -s notes.xml`, `head 20 -3 file`
|
| 148 |
- `echo -e "appdb" | tsort` — odd number of tokens
|
| 149 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 150 |
### Running the validator yourself (`validate.py`)
|
| 151 |
|
| 152 |
Validation is **decoupled** from generation — nothing in `generate.py` calls it.
|
|
@@ -237,7 +254,7 @@ fidelity, `bash -n`, dedup — always hold.
|
|
| 237 |
|
| 238 |
| File | What it is |
|
| 239 |
|---|---|
|
| 240 |
-
| `bash_dataset.jsonl` | The dataset — 54,
|
| 241 |
| `generate.py` | The generator (reproducible, resumable). |
|
| 242 |
| `validate.py` | Standalone validator (strict + `--permissive` execution). |
|
| 243 |
| `argcheck.py` | Argument-fidelity checker (generation gate + standalone auditor). |
|
|
|
|
| 47 |
|
| 48 |
| Metric | Value |
|
| 49 |
|---|---|
|
| 50 |
+
| Rows | **54,803** |
|
| 51 |
+
| Distinct requests (`variant_group`s) | 45,142 |
|
| 52 |
| `single` / `pipeline` / `script` | **40% / 35% / 25%** (exact) |
|
| 53 |
| Distinct primary utilities | 183 |
|
| 54 |
| `bash -n` valid | **100%** (0 syntax errors) |
|
|
|
|
| 122 |
|
| 123 |
Three layers, in increasing strength:
|
| 124 |
|
| 125 |
+
1. **`bash -n`** on every command — **0 / 54,803 syntax failures.**
|
| 126 |
2. **shellcheck** (warning level) on a large sample — the few findings are style nits
|
| 127 |
(`SC2010` `ls | grep`, `SC2164` `cd` without check).
|
| 128 |
3. **Real execution** of the runnable subset in a disposable Linux sandbox (Debian 13
|
|
|
|
| 147 |
- `uname java`, `cp --parents f1 f2`, `lspci -s notes.xml`, `head 20 -3 file`
|
| 148 |
- `echo -e "appdb" | tsort` — odd number of tokens
|
| 149 |
|
| 150 |
+
A second pass (manual analysis + a stricter validator) caught two generator bugs and
|
| 151 |
+
fixed them at the source (55,000 → **54,808**):
|
| 152 |
+
|
| 153 |
+
- **Invalid long-form flag variant.** The answer-diversity generator applied a command's
|
| 154 |
+
short→long flag map across a whole pipeline, leaking e.g. grep's `-n → --line-number`
|
| 155 |
+
onto a downstream `tail -n` → `tail --line-number` (invalid). **146 rows removed**;
|
| 156 |
+
the generator now restricts substitution to the owning command's segment. (`tail --lines`
|
| 157 |
+
/ `head --lines` are valid on GNU and were kept.) Also removed 3 `kill --CONT <name>`.
|
| 158 |
+
- **`find -size` missing its comparison operator.** "larger than N" requests emitted bare
|
| 159 |
+
`-size N` (means *exactly* N). **113 rows repaired** to `-size +N`; the template now sets
|
| 160 |
+
`+`/`-` from the request direction.
|
| 161 |
+
|
| 162 |
+
Re-running the hardened validator (now with `set -o pipefail` + a per-stage stderr scan for
|
| 163 |
+
`invalid/unrecognized option`) then surfaced **5 more** malformed rows the old exit-status
|
| 164 |
+
check had missed — `journalctl <service> <service>` (bare units need `-u`) — which were
|
| 165 |
+
removed (→ **54,803**).
|
| 166 |
+
|
| 167 |
### Running the validator yourself (`validate.py`)
|
| 168 |
|
| 169 |
Validation is **decoupled** from generation — nothing in `generate.py` calls it.
|
|
|
|
| 254 |
|
| 255 |
| File | What it is |
|
| 256 |
|---|---|
|
| 257 |
+
| `bash_dataset.jsonl` | The dataset — 54,803 rows (≈22 MB, no Git LFS needed). |
|
| 258 |
| `generate.py` | The generator (reproducible, resumable). |
|
| 259 |
| `validate.py` | Standalone validator (strict + `--permissive` execution). |
|
| 260 |
| `argcheck.py` | Argument-fidelity checker (generation gate + standalone auditor). |
|