python / python/cpython

concurrent.interpreters: Interpreter.call() crashes when the result fails to unpickle in the caller

Open
#151,892 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

3.14 3.15 3.16 extension-modules topic-subinterpreters type-bug type-crash
Dominant language
Python
Stars
77.2k
Forks
35.9k
PR merge metrics
PR metrics pending

Description

When Interpreter.call() returns a value, that value is pickled in the callee interpreter and reconstructed (unpickled) in the calling interpreter on the way out of the cross-interpreter session, inside _PyXI_Exit() (Python/crossinterp.c). If the reconstruction step raises — e.g. the returned object's __reduce__ produces a callable that raises during unpickling — the raised exception is left set while _PyXI_Exit() applies its "preserve failure" handling. On a debug build this trips assert(!_PyErr_Occurred(tstate)) and aborts the process (SIGABRT); on a release build the stale exception is clobbered and the resulting error state is wrong.

Reproduction (debug build)
import sys, os, tempfile, importlib
from textwrap import dedent
from concurrent import interpreters

d = tempfile.mkdtemp(); sys.path.insert(0, d)
with open(os.path.join(d, "b.py"), "w") as f:
    f.write(dedent('''
        def _boom():
            raise ValueError("unpickling failed")
        class Bomb:
            def __reduce__(self):
                return (_boom, ())
        def mk():
            return Bomb()
    '''))
m = importlib.import_module("b")
ip = interpreters.create()
ip.prepare_main(_d=d)
ip.exec("import sys; sys.path.insert(0, _d); import b")
ip.call(m.mk)   # debug build: Assertion failed in _PyXI_Exit -> SIGABRT

This affects only the call() path (and any other _PyXI_Enter/_PyXI_Exit-bracketed result-preserve path). Queue/channel get()/recv() and exec() do not go through this path and are unaffected. Since call() runs the caller's own code in the subinterpreter, this is a robustness / crash bug, not a security issue.

Expected: call() should raise a normal Python exception instead of aborting.

Fix

Have the reconstruction-failure branch consume the live exception and propagate it as the failure, mirroring how the active-session _pop_preserved path already routes errors through capture_session_error. With the fix, call() raises NotShareableError with the underlying error reachable as the cause. A self-contained regression test is added in Lib/test/test_interpreters/test_api.py.

Affected versions

main only. The result-preserve machinery (_PyXI_Preserve / _finish_preserved, gh-133484) and the xi_error_* error-handling rewrite (gh-135369) that this touches both landed after the 3.14.0b1 freeze, so the affected code path does not exist on 3.14 or earlier.

Related

A separate, distinct crossinterp NULL dereference on the queue/channel receive path is tracked in #151862.

Linked PR

A PR follows.

Linked PRs
  • gh-151893

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.

Research direction

Start in Python/crossinterp.c at _PyXI_Exit() and compare the reconstruction-failure handling with the active-session _pop_preserved path and capture_session_error. Then read the regression coverage in Lib/test/test_interpreters/test_api.py and run that test; done means call() raises NotShareableError with the underlying error as its cause instead of aborting.

Written by the indexing model from the issue text.

Assessment

Tech stack
c, python
Domain
compilers
Issue type
Bug
Difficulty
3/5
Estimated time
1-2 days
Activity status
Stale
Clarity
Clearly specified
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.