codehound 1.2.0__tar.gz → 1.4.0__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (33) hide show
  1. {codehound-1.2.0 → codehound-1.4.0}/PKG-INFO +91 -20
  2. {codehound-1.2.0 → codehound-1.4.0}/README.md +90 -19
  3. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/__init__.py +5 -4
  4. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/__init__.py +20 -0
  5. codehound-1.4.0/src/codehound/checks/async_property.py +54 -0
  6. codehound-1.4.0/src/codehound/checks/bare_except.py +88 -0
  7. codehound-1.4.0/src/codehound/checks/collections_abc_import.py +83 -0
  8. codehound-1.4.0/src/codehound/checks/discarded_future.py +93 -0
  9. codehound-1.4.0/src/codehound/checks/floating_process.py +130 -0
  10. codehound-1.4.0/src/codehound/checks/lru_cache_on_method.py +79 -0
  11. codehound-1.4.0/src/codehound/checks/removed_asyncio_task_methods.py +66 -0
  12. codehound-1.4.0/src/codehound/checks/removed_getargspec.py +58 -0
  13. codehound-1.4.0/src/codehound/checks/unclosed_socket.py +92 -0
  14. codehound-1.4.0/src/codehound/checks/unprotected_lock.py +105 -0
  15. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/cli.py +20 -16
  16. codehound-1.4.0/src/codehound/sarif.py +67 -0
  17. codehound-1.4.0/src/codehound/terminal.py +57 -0
  18. {codehound-1.2.0 → codehound-1.4.0}/tests/test_checks.py +424 -0
  19. codehound-1.4.0/tests/test_output_formats.py +74 -0
  20. {codehound-1.2.0 → codehound-1.4.0}/.gitignore +0 -0
  21. {codehound-1.2.0 → codehound-1.4.0}/LICENSE +0 -0
  22. {codehound-1.2.0 → codehound-1.4.0}/pyproject.toml +0 -0
  23. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/asyncio_run_in_loop.py +0 -0
  24. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/blocking_async.py +0 -0
  25. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/datetime_utcnow.py +0 -0
  26. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/floating_task.py +0 -0
  27. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/floating_thread.py +0 -0
  28. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/get_event_loop.py +0 -0
  29. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/loop_closure_capture.py +0 -0
  30. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/mutable_defaults.py +0 -0
  31. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/resource_leak.py +0 -0
  32. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/checks/unawaited_coroutine.py +0 -0
  33. {codehound-1.2.0 → codehound-1.4.0}/src/codehound/core.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.5
2
2
  Name: codehound
3
- Version: 1.2.0
3
+ Version: 1.4.0
4
4
  Summary: An AST-based static analyzer that hunts real correctness and async-safety bugs in Python code.
5
5
  Project-URL: Homepage, https://github.com/kratos0718/codehound
6
6
  Project-URL: Issues, https://github.com/kratos0718/codehound/issues
@@ -24,7 +24,7 @@ Description-Content-Type: text/markdown
24
24
 
25
25
  <h1 align="center">codehound</h1>
26
26
 
27
- **An AST-based static analyzer that hunts *real* bugs in large Python codebases — seven of the ten rules are backed by a bug that was actually found and merged into a major open-source AI framework; the other three are hardening rules verified against real false positives instead.**
27
+ **An AST-based static analyzer that hunts *real* bugs in large Python codebases — twenty checks, seven backed by a bug that was actually found and merged into a major open-source AI framework, the rest hardening rules verified against real false positives across a ~20-framework validation corpus instead of just reasoned about.**
28
28
 
29
29
  [![CI](https://github.com/kratos0718/codehound/actions/workflows/ci.yml/badge.svg)](https://github.com/kratos0718/codehound/actions/workflows/ci.yml)
30
30
  [![PyPI](https://img.shields.io/pypi/v/codehound.svg)](https://pypi.org/project/codehound/)
@@ -34,7 +34,7 @@ Description-Content-Type: text/markdown
34
34
 
35
35
  Most linters flag style. `codehound` flags the *subtle correctness and async-safety bugs* that slip past code review and only bite in production — event-loop stalls, shared mutable state, leaked file descriptors, fire-and-forget tasks that get garbage-collected mid-run.
36
36
 
37
- Each of the six checks below isn't theoretical. **I wrote it after finding — and fixing, via a merged pull request — that exact bug in a real, popular framework** (agno 25k⭐, crewAI 30k⭐, mem0, huggingface_hub).
37
+ Most of the checks below aren't theoretical. **I wrote them after finding — and fixing, via a merged pull request — that exact bug in a real, popular framework** (agno 25k⭐, crewAI 30k⭐, mem0, llama_index, accelerate).
38
38
 
39
39
  ---
40
40
 
@@ -71,7 +71,7 @@ I was contributing bug fixes to large AI frameworks and noticed the same handful
71
71
  pip install codehound
72
72
  ```
73
73
 
74
- Zero dependencies — it's ~750 lines on top of the standard-library `ast` module, so this installs instantly and runs fully offline, no API key or network call involved.
74
+ Zero dependencies — it's ~2,200 lines on top of the standard-library `ast` module, so this installs instantly and runs fully offline, no API key or network call involved.
75
75
 
76
76
  <details>
77
77
  <summary>From a clone instead (for development)</summary>
@@ -91,6 +91,9 @@ PYTHONPATH=src python -m codehound.cli scan path/to/project
91
91
  # scan a project (skips tests/, docs/, examples/, vendored code by default)
92
92
  codehound scan path/to/project
93
93
 
94
+ # scan multiple files/directories in one invocation (what pre-commit does)
95
+ codehound scan file1.py file2.py src/
96
+
94
97
  # only run specific checks
95
98
  codehound scan path/to/project --select CH001,CH006
96
99
 
@@ -98,6 +101,9 @@ codehound scan path/to/project --select CH001,CH006
98
101
  codehound scan path/to/project --format json
99
102
  codehound scan path/to/project --format csv
100
103
 
104
+ # GitHub Code Scanning (Security tab) can ingest this directly
105
+ codehound scan path/to/project --format sarif > results.sarif
106
+
101
107
  # list every available check
102
108
  codehound list
103
109
  ```
@@ -108,6 +114,29 @@ codehound list
108
114
  - run: codehound scan src # fails the build on a regression
109
115
  ```
110
116
 
117
+ ### GitHub Action
118
+
119
+ ```yaml
120
+ - uses: kratos0718/codehound@v1
121
+ with:
122
+ path: src
123
+ # select: CH001,CH006 # optional, defaults to all checks
124
+ # fail-on-findings: "false" # optional, report without failing the build
125
+ # upload-sarif: "false" # optional, skip the Code Scanning upload
126
+ ```
127
+
128
+ Uploads findings to the repo's **Security → Code Scanning** tab via SARIF, in addition to failing the step (unless `fail-on-findings: "false"`).
129
+
130
+ ### pre-commit
131
+
132
+ ```yaml
133
+ repos:
134
+ - repo: https://github.com/kratos0718/codehound
135
+ rev: v1.4.0
136
+ hooks:
137
+ - id: codehound
138
+ ```
139
+
111
140
  ---
112
141
 
113
142
  ## The checks
@@ -124,15 +153,25 @@ codehound list
124
153
  | **CH008** | `asyncio-run-in-running-loop` | `asyncio.run(...)` called from inside an `async def` — always raises `RuntimeError`, immediately, every time. | hardening rule — zero corpus hits (see below) |
125
154
  | **CH009** | `floating-thread` | A non-daemon `threading.Thread` that's `.start()`ed but never `.join()`ed — the thread analog of CH006. | hardening rule — see below |
126
155
  | **CH010** | `loop-closure-capture` | A `lambda` inside a `for` loop (or comprehension) that's *stored* (appended, assigned, returned) and captures the loop variable by reference — every stored instance ends up sharing the loop's **final** value. | **accelerate** (HuggingFace) — `MegatronEngine.get_module_config`'s `param_sync_func` list, PR #4273 |
156
+ | **CH011** | `lru-cache-on-method` | `@lru_cache`/`@cache` decorating an instance method — the cache holds a strong reference to `self` forever, so every instance that ever calls the method leaks for the process lifetime. | hardening rule — real hits across litellm, vllm, accelerate, marimo, dspy |
157
+ | **CH012** | `floating-process` | A non-daemon `multiprocessing.Process` that's `.start()`ed but never `.join()`ed — the process analog of CH009. | hardening rule |
158
+ | **CH013** | `discarded-future` | `ThreadPoolExecutor`/`ProcessPoolExecutor.submit(...)` called as a bare statement — the returned `Future` (and any exception raised inside the submitted work) is silently discarded. | hardening rule — real hits in litellm, accelerate, langchain |
159
+ | **CH014** | `unprotected-lock-acquire` | `lock.acquire()` outside a `with`, whose matching `.release()` isn't inside a `finally:` — an exception between acquire and release deadlocks every future caller of that lock. | hardening rule — real hits in vllm, accelerate, torchtune |
160
+ | **CH015** | `async-property` | `@property`/`@cached_property` wrapping an `async def` — accessing the attribute hands back an un-awaited coroutine object, not the value. | hardening rule |
161
+ | **CH016** | `unclosed-socket` | `socket.socket(...)` stored without a context manager or matching `.close()` — the socket analog of CH005; leaks the file descriptor. | hardening rule |
162
+ | **CH017** | `collections-abc-import` | `from collections import Mapping` (or `Sequence`, `Iterable`, …) — the ABCs were removed from `collections` itself in Python 3.10; they live in `collections.abc`. | hardening rule |
163
+ | **CH018** | `removed-asyncio-task-methods` | `asyncio.Task.current_task()` / `.all_tasks()` — both removed in Python 3.9; use `asyncio.current_task()` / `asyncio.all_tasks()`. | hardening rule |
164
+ | **CH019** | `removed-getargspec` | `inspect.getargspec(...)` — removed in Python 3.11 after a decade-plus deprecation; use `inspect.signature(...)`. | hardening rule |
165
+ | **CH020** | `bare-except` | A bare `except:` (or unused `except BaseException:`) — also catches `KeyboardInterrupt`/`SystemExit`, so Ctrl-C stops working and `sys.exit()` gets silently absorbed. | hardening rule — real hits in agno, llama_index, marimo, litellm |
127
166
 
128
167
  `codehound list` prints this from the source of truth.
129
168
 
130
- CH007-CH010 don't have found-and-merged bugs behind all of them the way
131
- CH001-CH006 do - three are hardening rules for well-known Python
169
+ CH007-CH020 don't have found-and-merged bugs behind all of them the way
170
+ CH001-CH006 and CH010 do - most are hardening rules for well-known Python
132
171
  correctness gotchas rather than something this project personally
133
172
  tracked down first. CH010 is the exception: it found a genuine, serious
134
- bug on its own, in HuggingFace's `accelerate` - see below. Building all
135
- four surfaced real false positives, each one fixed before shipping:
173
+ bug on its own, in HuggingFace's `accelerate` - see below. Building
174
+ CH007-CH010 surfaced real false positives, each one fixed before shipping:
136
175
 
137
176
  - **CH007** (agno): a bare `self.foo()` call matched against an unrelated
138
177
  same-named `async def foo` on a *different* class (agno's own
@@ -170,14 +209,27 @@ it's very unlikely to survive basic testing; CH007 and CH009 both only
170
209
  match same-file names by design, and most real cases of either are
171
210
  plausibly cross-module.
172
211
 
173
- **One check we built and did not ship: CH011 `exception-chaining`**
174
- (`except X as e: raise Y(...)` with no `from e`, discarding the real
175
- traceback - overlaps flake8-bugbear B904). It worked exactly as designed,
176
- but at a scale that says more about how common the pattern is than about
177
- anything worth flagging: **1,911 hits across the same ~20-framework
178
- corpus**. Shipping a check that fires that often would make every scan
179
- result mostly CH011 noise, undermining the "a finding must be defensible"
180
- standard the rest of this tool holds itself to. Built, measured, and
212
+ **Two checks we built and did not ship.** `exception-chaining` (`except X
213
+ as e: raise Y(...)` with no `from e`, discarding the real traceback -
214
+ overlaps flake8-bugbear B904) worked exactly as designed, but at a scale
215
+ that says more about how common the pattern is than about anything worth
216
+ flagging: **1,911 hits across the same ~20-framework corpus**. Shipping a
217
+ check that fires that often would make every scan result mostly noise,
218
+ undermining the "a finding must be defensible" standard the rest of this
219
+ tool holds itself to.
220
+
221
+ `cancelled-error-swallowed` (`except asyncio.CancelledError: pass` -
222
+ premise: silently swallowing task cancellation is a bug) went further
223
+ than volume alone: the **first two real hits checked**, in two different
224
+ frameworks, were both correct code, not bugs. agno's was `existing_task.
225
+ cancel(); try: await existing_task; except CancelledError: pass` - the
226
+ textbook-correct way to await a task's own cancellation. letta's was an
227
+ explicit, logged recovery path (`except (CancelledError, ...) as e: logger
228
+ .info(...); <continue processing>`) with a comment literally saying it was
229
+ overriding the cancellation on purpose. Unlike the exception-chaining
230
+ volume problem, this one meant the check's core premise was false in a
231
+ large fraction of real occurrences - so it was deleted outright rather
232
+ than kept at a lower confidence tier. Both are: built, measured, and
181
233
  deliberately left out - a real decision, not an oversight.
182
234
 
183
235
  ---
@@ -189,7 +241,9 @@ codehound/
189
241
  ├── core.py # file discovery, AST parsing, the Finding/Check contract,
190
242
  │ # and a child→parent map so checks can ask "what's my
191
243
  │ # enclosing function / am I inside a `with`?"
192
- ├── cli.py # `scan` / `list`, text|json|csv output, CI-friendly exit codes
244
+ ├── cli.py # `scan` / `list`, text|json|csv|sarif output, CI-friendly exit codes
245
+ ├── sarif.py # SARIF 2.1.0 output for GitHub Code Scanning
246
+ ├── terminal.py # colored text output (auto-disabled for non-TTY / NO_COLOR)
193
247
  └── checks/ # one small, independently-tested class per rule
194
248
  ├── blocking_async.py (CH001)
195
249
  ├── mutable_defaults.py (CH002)
@@ -200,12 +254,22 @@ codehound/
200
254
  ├── unawaited_coroutine.py (CH007)
201
255
  ├── asyncio_run_in_loop.py (CH008)
202
256
  ├── floating_thread.py (CH009)
203
- └── loop_closure_capture.py (CH010)
257
+ ├── loop_closure_capture.py (CH010)
258
+ ├── lru_cache_on_method.py (CH011)
259
+ ├── floating_process.py (CH012)
260
+ ├── discarded_future.py (CH013)
261
+ ├── unprotected_lock.py (CH014)
262
+ ├── async_property.py (CH015)
263
+ ├── unclosed_socket.py (CH016)
264
+ ├── collections_abc_import.py (CH017)
265
+ ├── removed_asyncio_task_methods.py (CH018)
266
+ ├── removed_getargspec.py (CH019)
267
+ └── bare_except.py (CH020)
204
268
  ```
205
269
 
206
270
  Each check receives a parsed `ast` tree plus the precomputed parent map and returns `Finding`s. Adding a rule is one file + one registry line + a test. See [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) for a full walkthrough of the engine, the parent map, and the design decisions.
207
271
 
208
- **False-positive discipline is a feature.** CH005 won't flag a handle that's `return`ed (the caller owns it) or explicitly `.close()`d. CH006 won't flag `TaskGroup.create_task` (the group holds the reference). CH001 only fires when the *enclosing* function is `async`. CH007 scopes `self.foo()` matches to async methods on the *same* class as the call site, and bare `foo()` matches to module-level async functions that aren't shadowed by a same-named parameter. CH009 doesn't flag a thread handed off as *any* object's attribute, not just `self`. CH010 only fires when a lambda is directly stored (appended, assigned, returned), not merely passed as a callback argument that gets consumed on the spot. All four of those guards exist because of real false positives caught while building the checks (see above). The test suite asserts both "bad code is flagged" and "correct code is not."
272
+ **False-positive discipline is a feature.** CH005 won't flag a handle that's `return`ed (the caller owns it) or explicitly `.close()`d. CH006 won't flag `TaskGroup.create_task` (the group holds the reference). CH001 only fires when the *enclosing* function is `async`. CH007 scopes `self.foo()` matches to async methods on the *same* class as the call site, and bare `foo()` matches to module-level async functions that aren't shadowed by a same-named parameter. CH009 doesn't flag a thread handed off as *any* object's attribute, not just `self`. CH010 only fires when a lambda is directly stored (appended, assigned, returned), not merely passed as a callback argument that gets consumed on the spot. CH020 won't flag a `BaseException` handler whose bound name is actually referenced, or whose body re-raises anywhere in its own scope (not counting a nested try/except's own handler) — both real patterns found in agno. All of those guards exist because of real false positives caught while building the checks (see above and [`docs/FINDINGS.md`](docs/FINDINGS.md)). The test suite asserts both "bad code is flagged" and "correct code is not."
209
273
 
210
274
  ---
211
275
 
@@ -227,10 +291,17 @@ Every check has paired tests: the buggy pattern *is* flagged, and the idiomatic
227
291
  - [x] `asyncio.run()` inside a running loop — CH008
228
292
  - [x] Non-daemon thread started without a join — CH009 (the thread analog of CH006)
229
293
  - [x] Loop-variable closure capture in lambdas — CH010
294
+ - [x] Pre-commit hook — `.pre-commit-hooks.yaml`
295
+ - [x] GitHub Action — `action.yml`, uploads SARIF to Code Scanning
296
+ - [x] SARIF output — `--format sarif`
297
+ - [x] Colored terminal output (auto-disabled for non-TTY / `NO_COLOR`)
298
+ - [x] Multi-path `scan` invocation (what the pre-commit hook needs)
299
+ - [x] 20 checks — memory leaks (`lru_cache` on methods), floating processes,
300
+ discarded futures, unprotected locks, async properties, unclosed
301
+ sockets, removed-in-3.9/3.10/3.11 stdlib APIs, bare `except:` — CH011-CH020
230
302
  - [ ] Cross-module resolution for CH007/CH009 (currently same-file only)
231
303
  - [ ] Sync HTTP clients constructed inside async request handlers
232
304
  - [ ] `--fix` for the mechanical rules (CH002, CH003, CH004)
233
- - [ ] Pre-commit hook
234
305
 
235
306
  ---
236
307
 
@@ -4,7 +4,7 @@
4
4
 
5
5
  <h1 align="center">codehound</h1>
6
6
 
7
- **An AST-based static analyzer that hunts *real* bugs in large Python codebases — seven of the ten rules are backed by a bug that was actually found and merged into a major open-source AI framework; the other three are hardening rules verified against real false positives instead.**
7
+ **An AST-based static analyzer that hunts *real* bugs in large Python codebases — twenty checks, seven backed by a bug that was actually found and merged into a major open-source AI framework, the rest hardening rules verified against real false positives across a ~20-framework validation corpus instead of just reasoned about.**
8
8
 
9
9
  [![CI](https://github.com/kratos0718/codehound/actions/workflows/ci.yml/badge.svg)](https://github.com/kratos0718/codehound/actions/workflows/ci.yml)
10
10
  [![PyPI](https://img.shields.io/pypi/v/codehound.svg)](https://pypi.org/project/codehound/)
@@ -14,7 +14,7 @@
14
14
 
15
15
  Most linters flag style. `codehound` flags the *subtle correctness and async-safety bugs* that slip past code review and only bite in production — event-loop stalls, shared mutable state, leaked file descriptors, fire-and-forget tasks that get garbage-collected mid-run.
16
16
 
17
- Each of the six checks below isn't theoretical. **I wrote it after finding — and fixing, via a merged pull request — that exact bug in a real, popular framework** (agno 25k⭐, crewAI 30k⭐, mem0, huggingface_hub).
17
+ Most of the checks below aren't theoretical. **I wrote them after finding — and fixing, via a merged pull request — that exact bug in a real, popular framework** (agno 25k⭐, crewAI 30k⭐, mem0, llama_index, accelerate).
18
18
 
19
19
  ---
20
20
 
@@ -51,7 +51,7 @@ I was contributing bug fixes to large AI frameworks and noticed the same handful
51
51
  pip install codehound
52
52
  ```
53
53
 
54
- Zero dependencies — it's ~750 lines on top of the standard-library `ast` module, so this installs instantly and runs fully offline, no API key or network call involved.
54
+ Zero dependencies — it's ~2,200 lines on top of the standard-library `ast` module, so this installs instantly and runs fully offline, no API key or network call involved.
55
55
 
56
56
  <details>
57
57
  <summary>From a clone instead (for development)</summary>
@@ -71,6 +71,9 @@ PYTHONPATH=src python -m codehound.cli scan path/to/project
71
71
  # scan a project (skips tests/, docs/, examples/, vendored code by default)
72
72
  codehound scan path/to/project
73
73
 
74
+ # scan multiple files/directories in one invocation (what pre-commit does)
75
+ codehound scan file1.py file2.py src/
76
+
74
77
  # only run specific checks
75
78
  codehound scan path/to/project --select CH001,CH006
76
79
 
@@ -78,6 +81,9 @@ codehound scan path/to/project --select CH001,CH006
78
81
  codehound scan path/to/project --format json
79
82
  codehound scan path/to/project --format csv
80
83
 
84
+ # GitHub Code Scanning (Security tab) can ingest this directly
85
+ codehound scan path/to/project --format sarif > results.sarif
86
+
81
87
  # list every available check
82
88
  codehound list
83
89
  ```
@@ -88,6 +94,29 @@ codehound list
88
94
  - run: codehound scan src # fails the build on a regression
89
95
  ```
90
96
 
97
+ ### GitHub Action
98
+
99
+ ```yaml
100
+ - uses: kratos0718/codehound@v1
101
+ with:
102
+ path: src
103
+ # select: CH001,CH006 # optional, defaults to all checks
104
+ # fail-on-findings: "false" # optional, report without failing the build
105
+ # upload-sarif: "false" # optional, skip the Code Scanning upload
106
+ ```
107
+
108
+ Uploads findings to the repo's **Security → Code Scanning** tab via SARIF, in addition to failing the step (unless `fail-on-findings: "false"`).
109
+
110
+ ### pre-commit
111
+
112
+ ```yaml
113
+ repos:
114
+ - repo: https://github.com/kratos0718/codehound
115
+ rev: v1.4.0
116
+ hooks:
117
+ - id: codehound
118
+ ```
119
+
91
120
  ---
92
121
 
93
122
  ## The checks
@@ -104,15 +133,25 @@ codehound list
104
133
  | **CH008** | `asyncio-run-in-running-loop` | `asyncio.run(...)` called from inside an `async def` — always raises `RuntimeError`, immediately, every time. | hardening rule — zero corpus hits (see below) |
105
134
  | **CH009** | `floating-thread` | A non-daemon `threading.Thread` that's `.start()`ed but never `.join()`ed — the thread analog of CH006. | hardening rule — see below |
106
135
  | **CH010** | `loop-closure-capture` | A `lambda` inside a `for` loop (or comprehension) that's *stored* (appended, assigned, returned) and captures the loop variable by reference — every stored instance ends up sharing the loop's **final** value. | **accelerate** (HuggingFace) — `MegatronEngine.get_module_config`'s `param_sync_func` list, PR #4273 |
136
+ | **CH011** | `lru-cache-on-method` | `@lru_cache`/`@cache` decorating an instance method — the cache holds a strong reference to `self` forever, so every instance that ever calls the method leaks for the process lifetime. | hardening rule — real hits across litellm, vllm, accelerate, marimo, dspy |
137
+ | **CH012** | `floating-process` | A non-daemon `multiprocessing.Process` that's `.start()`ed but never `.join()`ed — the process analog of CH009. | hardening rule |
138
+ | **CH013** | `discarded-future` | `ThreadPoolExecutor`/`ProcessPoolExecutor.submit(...)` called as a bare statement — the returned `Future` (and any exception raised inside the submitted work) is silently discarded. | hardening rule — real hits in litellm, accelerate, langchain |
139
+ | **CH014** | `unprotected-lock-acquire` | `lock.acquire()` outside a `with`, whose matching `.release()` isn't inside a `finally:` — an exception between acquire and release deadlocks every future caller of that lock. | hardening rule — real hits in vllm, accelerate, torchtune |
140
+ | **CH015** | `async-property` | `@property`/`@cached_property` wrapping an `async def` — accessing the attribute hands back an un-awaited coroutine object, not the value. | hardening rule |
141
+ | **CH016** | `unclosed-socket` | `socket.socket(...)` stored without a context manager or matching `.close()` — the socket analog of CH005; leaks the file descriptor. | hardening rule |
142
+ | **CH017** | `collections-abc-import` | `from collections import Mapping` (or `Sequence`, `Iterable`, …) — the ABCs were removed from `collections` itself in Python 3.10; they live in `collections.abc`. | hardening rule |
143
+ | **CH018** | `removed-asyncio-task-methods` | `asyncio.Task.current_task()` / `.all_tasks()` — both removed in Python 3.9; use `asyncio.current_task()` / `asyncio.all_tasks()`. | hardening rule |
144
+ | **CH019** | `removed-getargspec` | `inspect.getargspec(...)` — removed in Python 3.11 after a decade-plus deprecation; use `inspect.signature(...)`. | hardening rule |
145
+ | **CH020** | `bare-except` | A bare `except:` (or unused `except BaseException:`) — also catches `KeyboardInterrupt`/`SystemExit`, so Ctrl-C stops working and `sys.exit()` gets silently absorbed. | hardening rule — real hits in agno, llama_index, marimo, litellm |
107
146
 
108
147
  `codehound list` prints this from the source of truth.
109
148
 
110
- CH007-CH010 don't have found-and-merged bugs behind all of them the way
111
- CH001-CH006 do - three are hardening rules for well-known Python
149
+ CH007-CH020 don't have found-and-merged bugs behind all of them the way
150
+ CH001-CH006 and CH010 do - most are hardening rules for well-known Python
112
151
  correctness gotchas rather than something this project personally
113
152
  tracked down first. CH010 is the exception: it found a genuine, serious
114
- bug on its own, in HuggingFace's `accelerate` - see below. Building all
115
- four surfaced real false positives, each one fixed before shipping:
153
+ bug on its own, in HuggingFace's `accelerate` - see below. Building
154
+ CH007-CH010 surfaced real false positives, each one fixed before shipping:
116
155
 
117
156
  - **CH007** (agno): a bare `self.foo()` call matched against an unrelated
118
157
  same-named `async def foo` on a *different* class (agno's own
@@ -150,14 +189,27 @@ it's very unlikely to survive basic testing; CH007 and CH009 both only
150
189
  match same-file names by design, and most real cases of either are
151
190
  plausibly cross-module.
152
191
 
153
- **One check we built and did not ship: CH011 `exception-chaining`**
154
- (`except X as e: raise Y(...)` with no `from e`, discarding the real
155
- traceback - overlaps flake8-bugbear B904). It worked exactly as designed,
156
- but at a scale that says more about how common the pattern is than about
157
- anything worth flagging: **1,911 hits across the same ~20-framework
158
- corpus**. Shipping a check that fires that often would make every scan
159
- result mostly CH011 noise, undermining the "a finding must be defensible"
160
- standard the rest of this tool holds itself to. Built, measured, and
192
+ **Two checks we built and did not ship.** `exception-chaining` (`except X
193
+ as e: raise Y(...)` with no `from e`, discarding the real traceback -
194
+ overlaps flake8-bugbear B904) worked exactly as designed, but at a scale
195
+ that says more about how common the pattern is than about anything worth
196
+ flagging: **1,911 hits across the same ~20-framework corpus**. Shipping a
197
+ check that fires that often would make every scan result mostly noise,
198
+ undermining the "a finding must be defensible" standard the rest of this
199
+ tool holds itself to.
200
+
201
+ `cancelled-error-swallowed` (`except asyncio.CancelledError: pass` -
202
+ premise: silently swallowing task cancellation is a bug) went further
203
+ than volume alone: the **first two real hits checked**, in two different
204
+ frameworks, were both correct code, not bugs. agno's was `existing_task.
205
+ cancel(); try: await existing_task; except CancelledError: pass` - the
206
+ textbook-correct way to await a task's own cancellation. letta's was an
207
+ explicit, logged recovery path (`except (CancelledError, ...) as e: logger
208
+ .info(...); <continue processing>`) with a comment literally saying it was
209
+ overriding the cancellation on purpose. Unlike the exception-chaining
210
+ volume problem, this one meant the check's core premise was false in a
211
+ large fraction of real occurrences - so it was deleted outright rather
212
+ than kept at a lower confidence tier. Both are: built, measured, and
161
213
  deliberately left out - a real decision, not an oversight.
162
214
 
163
215
  ---
@@ -169,7 +221,9 @@ codehound/
169
221
  ├── core.py # file discovery, AST parsing, the Finding/Check contract,
170
222
  │ # and a child→parent map so checks can ask "what's my
171
223
  │ # enclosing function / am I inside a `with`?"
172
- ├── cli.py # `scan` / `list`, text|json|csv output, CI-friendly exit codes
224
+ ├── cli.py # `scan` / `list`, text|json|csv|sarif output, CI-friendly exit codes
225
+ ├── sarif.py # SARIF 2.1.0 output for GitHub Code Scanning
226
+ ├── terminal.py # colored text output (auto-disabled for non-TTY / NO_COLOR)
173
227
  └── checks/ # one small, independently-tested class per rule
174
228
  ├── blocking_async.py (CH001)
175
229
  ├── mutable_defaults.py (CH002)
@@ -180,12 +234,22 @@ codehound/
180
234
  ├── unawaited_coroutine.py (CH007)
181
235
  ├── asyncio_run_in_loop.py (CH008)
182
236
  ├── floating_thread.py (CH009)
183
- └── loop_closure_capture.py (CH010)
237
+ ├── loop_closure_capture.py (CH010)
238
+ ├── lru_cache_on_method.py (CH011)
239
+ ├── floating_process.py (CH012)
240
+ ├── discarded_future.py (CH013)
241
+ ├── unprotected_lock.py (CH014)
242
+ ├── async_property.py (CH015)
243
+ ├── unclosed_socket.py (CH016)
244
+ ├── collections_abc_import.py (CH017)
245
+ ├── removed_asyncio_task_methods.py (CH018)
246
+ ├── removed_getargspec.py (CH019)
247
+ └── bare_except.py (CH020)
184
248
  ```
185
249
 
186
250
  Each check receives a parsed `ast` tree plus the precomputed parent map and returns `Finding`s. Adding a rule is one file + one registry line + a test. See [`docs/ARCHITECTURE.md`](docs/ARCHITECTURE.md) for a full walkthrough of the engine, the parent map, and the design decisions.
187
251
 
188
- **False-positive discipline is a feature.** CH005 won't flag a handle that's `return`ed (the caller owns it) or explicitly `.close()`d. CH006 won't flag `TaskGroup.create_task` (the group holds the reference). CH001 only fires when the *enclosing* function is `async`. CH007 scopes `self.foo()` matches to async methods on the *same* class as the call site, and bare `foo()` matches to module-level async functions that aren't shadowed by a same-named parameter. CH009 doesn't flag a thread handed off as *any* object's attribute, not just `self`. CH010 only fires when a lambda is directly stored (appended, assigned, returned), not merely passed as a callback argument that gets consumed on the spot. All four of those guards exist because of real false positives caught while building the checks (see above). The test suite asserts both "bad code is flagged" and "correct code is not."
252
+ **False-positive discipline is a feature.** CH005 won't flag a handle that's `return`ed (the caller owns it) or explicitly `.close()`d. CH006 won't flag `TaskGroup.create_task` (the group holds the reference). CH001 only fires when the *enclosing* function is `async`. CH007 scopes `self.foo()` matches to async methods on the *same* class as the call site, and bare `foo()` matches to module-level async functions that aren't shadowed by a same-named parameter. CH009 doesn't flag a thread handed off as *any* object's attribute, not just `self`. CH010 only fires when a lambda is directly stored (appended, assigned, returned), not merely passed as a callback argument that gets consumed on the spot. CH020 won't flag a `BaseException` handler whose bound name is actually referenced, or whose body re-raises anywhere in its own scope (not counting a nested try/except's own handler) — both real patterns found in agno. All of those guards exist because of real false positives caught while building the checks (see above and [`docs/FINDINGS.md`](docs/FINDINGS.md)). The test suite asserts both "bad code is flagged" and "correct code is not."
189
253
 
190
254
  ---
191
255
 
@@ -207,10 +271,17 @@ Every check has paired tests: the buggy pattern *is* flagged, and the idiomatic
207
271
  - [x] `asyncio.run()` inside a running loop — CH008
208
272
  - [x] Non-daemon thread started without a join — CH009 (the thread analog of CH006)
209
273
  - [x] Loop-variable closure capture in lambdas — CH010
274
+ - [x] Pre-commit hook — `.pre-commit-hooks.yaml`
275
+ - [x] GitHub Action — `action.yml`, uploads SARIF to Code Scanning
276
+ - [x] SARIF output — `--format sarif`
277
+ - [x] Colored terminal output (auto-disabled for non-TTY / `NO_COLOR`)
278
+ - [x] Multi-path `scan` invocation (what the pre-commit hook needs)
279
+ - [x] 20 checks — memory leaks (`lru_cache` on methods), floating processes,
280
+ discarded futures, unprotected locks, async properties, unclosed
281
+ sockets, removed-in-3.9/3.10/3.11 stdlib APIs, bare `except:` — CH011-CH020
210
282
  - [ ] Cross-module resolution for CH007/CH009 (currently same-file only)
211
283
  - [ ] Sync HTTP clients constructed inside async request handlers
212
284
  - [ ] `--fix` for the mechanical rules (CH002, CH003, CH004)
213
- - [ ] Pre-commit hook
214
285
 
215
286
  ---
216
287
 
@@ -1,9 +1,10 @@
1
1
  """codehound - an AST-based static analyzer that hunts real bugs in Python code.
2
2
 
3
- Ten checks. Seven are each backed by a bug that was actually found and
3
+ Twenty checks. Seven are each backed by a bug that was actually found and
4
4
  fixed in a popular open-source AI framework (agno, crewAI, mem0,
5
- llama_index, accelerate). The other three (CH007-CH009) are hardening
6
- rules verified against real false positives instead - see docs/FINDINGS.md.
5
+ llama_index, accelerate). The rest (CH007-CH009, CH011-CH020) are
6
+ hardening rules verified against real false positives across a
7
+ ~20-framework validation corpus instead - see docs/FINDINGS.md.
7
8
  """
8
9
 
9
10
  from __future__ import annotations
@@ -11,7 +12,7 @@ from __future__ import annotations
11
12
  from codehound.checks import ALL_CHECKS, get_checks
12
13
  from codehound.core import Check, Finding, scan_file, scan_path
13
14
 
14
- __version__ = "1.2.0"
15
+ __version__ = "1.4.0"
15
16
 
16
17
  __all__ = [
17
18
  "ALL_CHECKS",
@@ -2,16 +2,26 @@
2
2
 
3
3
  from __future__ import annotations
4
4
 
5
+ from codehound.checks.async_property import AsyncProperty
5
6
  from codehound.checks.asyncio_run_in_loop import AsyncioRunInRunningLoop
7
+ from codehound.checks.bare_except import BareExcept
6
8
  from codehound.checks.blocking_async import BlockingCallInAsync
9
+ from codehound.checks.collections_abc_import import CollectionsAbcImport
7
10
  from codehound.checks.datetime_utcnow import DeprecatedDatetimeUtcnow
11
+ from codehound.checks.discarded_future import DiscardedFuture
12
+ from codehound.checks.floating_process import FloatingProcess
8
13
  from codehound.checks.floating_task import FloatingTask
9
14
  from codehound.checks.floating_thread import FloatingThread
10
15
  from codehound.checks.get_event_loop import DeprecatedGetEventLoop
11
16
  from codehound.checks.loop_closure_capture import LoopClosureCapture
17
+ from codehound.checks.lru_cache_on_method import LruCacheOnMethod
12
18
  from codehound.checks.mutable_defaults import MutableDefaultArgument
19
+ from codehound.checks.removed_asyncio_task_methods import RemovedAsyncioTaskMethods
20
+ from codehound.checks.removed_getargspec import RemovedGetargspec
13
21
  from codehound.checks.resource_leak import UnclosedFileHandle
14
22
  from codehound.checks.unawaited_coroutine import UnawaitedCoroutineCall
23
+ from codehound.checks.unclosed_socket import UnclosedSocket
24
+ from codehound.checks.unprotected_lock import UnprotectedLockAcquire
15
25
  from codehound.core import Check
16
26
 
17
27
  ALL_CHECKS: list[type[Check]] = [
@@ -25,6 +35,16 @@ ALL_CHECKS: list[type[Check]] = [
25
35
  AsyncioRunInRunningLoop,
26
36
  FloatingThread,
27
37
  LoopClosureCapture,
38
+ LruCacheOnMethod,
39
+ FloatingProcess,
40
+ DiscardedFuture,
41
+ UnprotectedLockAcquire,
42
+ AsyncProperty,
43
+ UnclosedSocket,
44
+ CollectionsAbcImport,
45
+ RemovedAsyncioTaskMethods,
46
+ RemovedGetargspec,
47
+ BareExcept,
28
48
  ]
29
49
 
30
50
 
@@ -0,0 +1,54 @@
1
+ """CH015 - ``@property`` on an ``async def`` method returns an unawaited coroutine.
2
+
3
+ Accessing a property never uses ``await`` - ``obj.thing`` is a plain
4
+ attribute access, so Python has no way to know it should await anything
5
+ even if the getter is a coroutine function. ``@property async def thing`` is
6
+ syntactically legal, but every access silently hands back a coroutine
7
+ object instead of the value, which is truthy, doesn't equal what you
8
+ compare it to, and (if never awaited) is a hidden
9
+ ``RuntimeWarning: coroutine 'thing' was never awaited`` waiting to happen
10
+ the first time someone actually uses the ``@property`` the way properties
11
+ are meant to be used.
12
+ """
13
+
14
+ from __future__ import annotations
15
+
16
+ import ast
17
+
18
+ from codehound.core import Check, Finding
19
+
20
+
21
+ def _is_property_decorator(dec: ast.expr) -> bool:
22
+ if isinstance(dec, ast.Name):
23
+ return dec.id == "property"
24
+ if isinstance(dec, ast.Attribute):
25
+ return dec.attr in ("property", "cached_property")
26
+ return False
27
+
28
+
29
+ class AsyncProperty(Check):
30
+ code = "CH015"
31
+ name = "async-property"
32
+ description = "@property wrapping an async def returns an unawaited coroutine, not the value."
33
+
34
+ def run(self, tree: ast.AST, parents: dict, path: str) -> list[Finding]:
35
+ findings: list[Finding] = []
36
+ for node in ast.walk(tree):
37
+ if not isinstance(node, ast.AsyncFunctionDef):
38
+ continue
39
+ if not any(_is_property_decorator(d) for d in node.decorator_list):
40
+ continue
41
+ findings.append(
42
+ Finding(
43
+ path=path,
44
+ line=node.lineno,
45
+ col=node.col_offset,
46
+ code=self.code,
47
+ message=(
48
+ f"`{node.name}` is an `async def` wrapped in `@property` - accessing it "
49
+ f"never awaits anything, so every access returns an unawaited coroutine "
50
+ f"object, not the value."
51
+ ),
52
+ )
53
+ )
54
+ return findings
@@ -0,0 +1,88 @@
1
+ """CH020 - Bare ``except:`` (or unused ``except BaseException:``) catches everything.
2
+
3
+ A bare `except:` (or the explicit `except BaseException:`) doesn't just
4
+ catch the exception you meant to handle - it catches `KeyboardInterrupt`
5
+ and `SystemExit` too, meaning Ctrl-C stops working and `sys.exit()` gets
6
+ silently absorbed instead of actually exiting.
7
+
8
+ Not flagged when the handler either **uses** the bound exception (a real,
9
+ deliberate pattern found while building this: agno's background-thread
10
+ runner, `except BaseException as e: thread_error.append(e)`, surfacing
11
+ the exception to the consumer via a queue afterward - legitimate because
12
+ it's a worker thread reporting failures back, not discarding them) or
13
+ **re-raises anything** (another real pattern found the same way: agno's
14
+ `except BaseException: <reset internal state>; raise` - properly
15
+ propagates the original error to the caller after cleanup, so nothing is
16
+ actually swallowed). A bare `except:` with neither has no way to inspect
17
+ or propagate the exception at all, so it's flagged unconditionally.
18
+ """
19
+
20
+ from __future__ import annotations
21
+
22
+ import ast
23
+
24
+ from codehound.core import Check, Finding
25
+
26
+
27
+ def _name_is_referenced(scope: ast.AST, name: str) -> bool:
28
+ for node in ast.walk(scope):
29
+ if isinstance(node, ast.Name) and node.id == name and isinstance(node.ctx, ast.Load):
30
+ return True
31
+ return False
32
+
33
+
34
+ def _has_raise_in_own_scope(handler: ast.ExceptHandler) -> bool:
35
+ """Any `raise` in the handler's own reachable body - not counting a
36
+ nested try/except's own handler, which is a different scope."""
37
+ found = False
38
+
39
+ def walk(node: ast.AST) -> None:
40
+ nonlocal found
41
+ if found:
42
+ return
43
+ for child in ast.iter_child_nodes(node):
44
+ if isinstance(child, ast.Raise):
45
+ found = True
46
+ return
47
+ if isinstance(child, ast.ExceptHandler):
48
+ continue
49
+ walk(child)
50
+
51
+ walk(handler)
52
+ return found
53
+
54
+
55
+ class BareExcept(Check):
56
+ code = "CH020"
57
+ name = "bare-except"
58
+ description = "Bare `except:` (or unused `except BaseException:`) also catches KeyboardInterrupt/SystemExit."
59
+
60
+ def run(self, tree: ast.AST, parents: dict, path: str) -> list[Finding]:
61
+ findings: list[Finding] = []
62
+ for node in ast.walk(tree):
63
+ if not isinstance(node, ast.ExceptHandler):
64
+ continue
65
+ is_bare = node.type is None
66
+ is_base_exception = isinstance(node.type, ast.Name) and node.type.id == "BaseException"
67
+ if not (is_bare or is_base_exception):
68
+ continue
69
+ if _has_raise_in_own_scope(node):
70
+ continue
71
+ if is_base_exception and node.name is not None:
72
+ if any(_name_is_referenced(stmt, node.name) for stmt in node.body):
73
+ continue
74
+ findings.append(
75
+ Finding(
76
+ path=path,
77
+ line=node.lineno,
78
+ col=node.col_offset,
79
+ code=self.code,
80
+ message=(
81
+ "bare `except:`" if is_bare else "`except BaseException:`"
82
+ )
83
+ + " also catches KeyboardInterrupt/SystemExit, and the exception isn't "
84
+ "used anywhere in the handler; catch the specific exception(s) this "
85
+ "handler is actually meant for.",
86
+ )
87
+ )
88
+ return findings