fix: detect hidden pytorch zip pickles#1043
Merged
mldangelo-oai merged 6 commits intomainfrom Apr 17, 2026
Merged
Conversation
Contributor
Performance BenchmarksCompared
|
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: f4e6c684e3
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
mldangelo-oai
force-pushed
the
mdangelo/codex/fix-pytorch-hidden-pickles
branch
from
f4e6c68 to
a274fc8
Compare
Apply the same refinements this reviewer just landed on PR #1048 to the hidden-pickle discovery path: 1. Aggregate probe-failure INFO checks. An adversarial archive with many members that raise on decompression (for example, unsupported methods or intermittent I/O) previously produced one `Pickle Discovery` INFO finding per member, flooding the checks list. Collect failures into a single summary check carrying the per-member exceptions under `details["entries"]`, with `details["zip_entries"]` and `details["failed_count"]` for quick consumers. `mark_inconclusive_ scan_result` is hoisted out of the loop so the metadata reason is recorded exactly once even if many members fail. 2. Document the identity-based dedup invariant in `_discover_pickle_files`. Both passes iterate the same `safe_entries` list, so `id(ZipInfo)` is stable for the duration of discovery; a future refactor that rebuilds the list from separate `infolist()` walks or fresh `ZipInfo` constructions would silently defeat the dedup. Call that out inline so we don't re-learn it the hard way. 3. Explain the magic thresholds in `_looks_like_binary_pickle_prefix`. The `>= 4` clean-parse and `>= 2` truncation thresholds are tuned to balance tensor-storage false positives against real pickle prefixes; undocumented they read like arbitrary numbers. 4. Add "keep in sync" comments on the standalone picklescan copies of `_looks_like_binary_pickle_prefix` and `_looks_like_proto0_or_1_ pickle` so the duplication (intentional for the standalone package) is at least signposted. Expand the CHANGELOG bullet to describe the always-on second-pass sniff, the fail-closed aggregation behavior, and the standalone-package mirror. Adds a new `test_pytorch_zip_discovery_aggregates_probe_failures_ into_single_check` regression test that proves 5 failing members collapse to one `Pickle Discovery` check with a deduplicated reason and all five per-member records under `details["entries"]`. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 234dfe66d3
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
…torch-hidden-pickles # Conflicts: # CHANGELOG.md
…torch-hidden-pickles # Conflicts: # CHANGELOG.md # modelaudit/scanners/pytorch_zip_scanner.py
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Finding
Fixes finding 1: hidden PyTorch ZIP pickles are skipped after data.pkl.
Validation