python / python/cpython

Disallow arbitrary types in co_consts

Open
#122,850 2 comments 0 reactions 1 assignee View on GitHub

@encukou is already working on this.

Since Aug 9, 2024.

interpreter-core performance
Dominant language
Python
Stars
77.2k
Forks
35.9k
PR merge metrics
PR metrics pending

Description

co_consts can contain values of any type, including mutable ones. For example:

def f():
    print(())

code = f.__code__
code = code.replace(co_consts=(None, []))
exec(code)

This is dangerous especially in the free-threaded build, which

  • de-duplicates constants and
  • immortalizes constants and relies them to be immortal

The “feature” is untested, as seen from the above snippet failing an assertion in free-threaded build.

Mutable subclasses of immutable built-in types are especially “interesting” here.

A solution would be to only allow built-in immutable marshallable types (booleans, integers, floating-point numbers, complex numbers, strings, bytes, tuples, frozensets, code objects, None, Ellipsis, StopIteration), excluding sublasses. That's a backwards-incompatible change.

Note that CPython now already recursively analyzes co_consts; there's little cost to doing additional inspection.

My current plan is:

  • Disallow “bad” values in free-threaded builds in 3.14, as this build is still experimental and this breaks vital invariants.
  • Deprecate “bad” values in regular build, and initially plan to disallow them in 3.16. (Minimal PEP-387 deprecation period chosen because this “feature” is very likely to interfere with optimizations, and unlikely to get proper test coverage.)

Until then, please avoid assuming that code constants are “sane”. (cc @brandtbucher @markshannon @colesbury)

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.