Security/safe pickle loader

[sanowl]

Overview

This PR eliminates the only medium-severity Bandit findings in the code-base by introducing a restricted-capability pickle loader and replacing import-time pickle.load/loads calls.

Motivation

The repo ships .pickle files that are deserialized at import time.
Unrestricted pickle.load can execute arbitrary code if the file is ever tampered with (CWE-502). While the pickles are trusted in normal use, hardening this path removes a potential supply-chain attack vector.

Key Changes

1. src/alphafold3/common/safe_pickle.py

   • Implements safe_load using a custom Unpickler that only allows harmless built-in container / primitive types (dict, list, tuple, str, int, etc.).
   • Any attempt to unpickle other globals raises pickle.UnpicklingError.

2. Replaced unsafe loads

   • constants/chemical_component_sets.py
   • constants/chemical_components.py
   • constants/converters/chemical_component_sets_gen.py
     These modules now read their pickle files via safe_pickle.safe_load.

3. Security scan result

   $ bandit -r src -ll
   SEVERITY.MEDIUM: 0
   SEVERITY.HIGH:   0
   

   No medium or high findings remain; low-severity items are stylistic.

Testing

• python -m py_compile on all modified files => passes.
• Imported the two constants modules with the generated .pickle files present – loads cleanly.
• Verified safe_pickle.safe_load:
  • Correctly loads a normal dict/tuple pickle.
  • Blocks a pickle containing datetime.datetime with UnpicklingError.
• Re-ran Bandit static analysis (see snippet above).

Impact

• No functional change to model inference or data pipeline; only safer deserialization of bundled constant data.
• Downstream users do not need to modify their code.

Checklist

• New code is lint-clean and typed.
• Bandit scan shows no medium/high findings.
• Added/updated tests or manual verification.
• Commit message follows conventional-commits style.

[Augustin-Zidek]

This file should not be included, same for all of the other pyc files below. (I will add the missing .gitignore file for this project.)