Umbrella issue: code review reports from cpython-review-toolkit

[devdanzin]

Bug report

What happened?

This is an umbrella issue for code review reports crafted using cpython-review-toolkit, covering ~350K lines of CPython C code in ~146 files (Modules/, Objects/, Python/).

The idea of this issue is that you read a report, find something you're interested in, and propose an issue based on that. If the issue is considered valid and worth fixing, you can then open a PR.

Some of the FIX-level bugs have already been collected in #146102, be sure to check that issue before picking a bug from the reports.

Edit: Some of the issues were found on alpha 6 and not confirmed in HEAD, be sure to check the issue reproduces on latest revision before investing time in it.

One important thing is that we used fewer agents than available when analyzing some of these files, so if you run the tool on a single file, it's probable it'll find more issues than present in the reports below.

Bug Classification

• FIX: Confirmed code defect requiring a fix (crash, leak, data corruption, wrong behavior)
• CONSIDER: Potential issue that may be acceptable depending on design intent
• POLICY: Code quality observation, not a correctness bug

Directory	Files Analyzed	Lines of Code	FIX-level Bugs
Modules/	~47 modules	~137,000	~120
Objects/	49 files	~117,000	18
Python/	~50 files	~93,000	31
Total	~146 files	~347,000	~170

Methodology

• Automated agents: Specialized refcount-auditor and error-path-analyzer agents examined each file
• Manual verification: Promising findings validated by reading source code
• Crash reproducers: 35+ bugs confirmed with Python-level reproducers
• OOM injection: _testcapi.set_nomemory() used to trigger allocation failures
• ASan builds: AddressSanitizer-instrumented debug build for memory error detection

Tools Used

• cpython-review-toolkit agents: refcount-auditor, error-path-analyzer, null-safety-scanner, memory-pattern-analyzer, gil-discipline-checker
• _testcapi.set_nomemory(N, 0) for persistent OOM injection
• resource.setrlimit(RLIMIT_AS, ...) for non-ASan OOM testing
• ASan-instrumented --with-pydebug build

Report Gist Index

Summary Reports

These offer important information and insights on methodology and results.

Gist	Report
CPython C Code Audit — Overview & Methodology	00_overview
Modules/ Directory Audit — Summary Report	01_modules_summary
Objects/ Directory Audit — Summary Report	02_objects_summary
Python/ Directory Audit — Summary Report	03_python_summary

Modules/ Reports

Gist	Report
_collectionsmodule.c — 2,894 lines	3 FIX, 2 CONSIDER
CPython C Code Exploration Report: Modules/_ctypes/	modules__ctypes
CPython C Code Exploration Report: _cursesmodule.c	modules__cursesmodule
CPython C Code Exploration Report: _datetimemodule.c	modules__datetimemodule
_functoolsmodule.c — 2,017 lines	2 FIX, 1 CONSIDER
CPython C Code Exploration Report: _interpchannelsmodule.c	modules__interpchannelsmodule
CPython C Code Exploration Report: _interpqueuesmodule.c	modules__interpqueuesmodule
CPython C Code Exploration Report: _interpretersmodule.c	modules__interpretersmodule
CPython C Code Exploration Report: _json.c	modules__json
CPython C Code Exploration Report: Modules/_sqlite/	modules__sqlite
_sre/sre.c — 3,493 lines	1 FIX, 2 CONSIDER
CPython C Code Exploration Report: _ssl.c	modules__ssl
CPython C Code Exploration Report: _struct.c	modules__struct
CPython C Code Exploration Report: _threadmodule.c	modules__threadmodule
CPython C Code Exploration Report: _zoneinfo.c	modules__zoneinfo
CPython C Code Exploration Report: arraymodule.c	modules_arraymodule
Batch 5 Small Modules — Combined Report	modules_batch5_small_modules
CPython C Code Exploration Report: itertoolsmodule.c	modules_itertoolsmodule
CPython C Code Exploration Report: mmapmodule.c	modules_mmapmodule
CPython C Code Exploration Report: posixmodule.c	modules_posixmodule
CPython C Code Exploration Report: pyexpat.c	modules_pyexpat
CPython C Code Exploration Report: selectmodule.c	modules_selectmodule
signalmodule.c — 2,088 lines	2 FIX, 2 CONSIDER
CPython C Code Exploration Report: socketmodule.c	modules_socketmodule

Objects/ Reports

Gist	Report
abstract.c + frameobject.c + odictobject.c	3 FIX
Objects/ Batch 4A — Small-Medium Files	4 FIX
Objects/ Batch 4B — Small Files (11 files)	0 FIX
bytesobject.c + bytearrayobject.c	0 FIX (refcount), 2 non-refcount issues noted
codeobject.c + object.c	4 FIX
dictobject.c (8,337 lines)	0 FIX
exceptions.c + memoryobject.c	2 FIX
genobject.c + funcobject.c + descrobject.c	0 FIX
listobject.c + setobject.c	0 FIX
longobject.c (6,988 lines)	0 FIX
obmalloc.c + unicode_format.c + unicode_writer.c + bytes_methods.c	2 FIX
Objects/ Directory Review — Final Summary	objects_summary
typeobject.c (12,843 lines)	3 FIX
typevarobject.c + floatobject.c + moduleobject.c + unicode_formatter.c	0 FIX
unicodeobject.c (14,971 lines)	0 FIX

Python/ Reports

Gist	Report
ceval.c + bltinmodule.c	python_ceval_bltinmodule
Compiler Pipeline (codegen + compile + symtable + flowgraph + assemble)	6 FIX
crossinterp.c + pystate.c + marshal.c	4 FIX
errors.c + _warnings.c + codecs.c + traceback.c + pythonrun.c	1 FIX
hamt.c + context.c + gc.c + getargs.c	2 FIX
import.c + sysmodule.c	2 FIX
pylifecycle.c + initconfig.c + fileutils.c	6 FIX
optimizer + specialize + instrumentation	3 FIX
Small Python/ files batch	4 FIX
Python/ Directory Review — Final Summary	python_summary

These reports were created with Claude Opus 4.6, using the /cpython-review-toolkit:explore [file or directory] all deep command.

CPython versions tested on:

CPython main branch

Operating systems tested on:

Linux

Output from running 'python -VV' on the command line:

Python 3.15.0a7+ (heads/main:e0f7c1097e1, Mar 17 2026, 18:10:52) [Clang 21.1.2 (2ubuntu6)

[skirpichev]

I suspect, that e.g. https://gist.github.com/devdanzin/742eb4647c2f6ecd6c009af02c143f90 was fixed by #145624.

Output from running 'python -VV' on the command line:
Python 3.15.0a7+ (heads/main:e0f7c1097e1, Mar 17 2026, 18:10:52) [Clang 21.1.2 (2ubuntu6)

Gist says: 3.15.0 alpha 6 (that's before merge of the mentioned pr). Are you sure, that all reports are reproducible on e0f7c10?

[devdanzin]

You are correct, some of the bugs were found on alpha 6 and, while some where confirmed on alpha 7, not all of them were. Adding a note about this, thank you for the heads up.