Summary

This PR addresses OOM-0014 from gh-151763.

It fixes a crash in _interpchannels._channel_id() when memory allocation fails while looking up module state.

Issue

channelsmod__channel_id() obtains the owning module with:

PyObject *mod = get_module_from_owned_type(cls);
assert(mod == self);
Py_DECREF(mod);

get_module_from_owned_type() eventually calls _get_current_module(), which allocates the module name using:

PyUnicode_FromString(MODULE_NAME_STR)

Under out-of-memory conditions this allocation can fail and return NULL with a pending MemoryError.

Before this change, channelsmod__channel_id() assumed the helper could never fail.

As a result:

• Debug builds abort at:

assert(mod == self);

because mod == NULL.

• Release builds compile out the assertion and continue to:

Py_DECREF(mod);

which dereferences NULL and results in an access violation / segmentation fault.

Root Cause

The helper get_module_from_owned_type(cls) returns a new reference on success but can legitimately return NULL when _get_current_module() fails during module-name allocation.

channelsmod__channel_id() did not check for that failure before using the returned object.

Fix

Add an explicit NULL check immediately after obtaining the module:

PyObject *mod = get_module_from_owned_type(cls);
if (mod == NULL) {
    return NULL;
}
assert(mod == self);
Py_DECREF(mod);

This preserves the existing invariant check while correctly propagating the pending MemoryError instead of aborting or crashing.

Validation

I reproduced the issue locally on both debug and release builds.

Before the fix

Debug build:

Assertion failed: mod == self
Modules/_interpchannelsmodule.c
Fatal Python error: Aborted

Release build:

Windows fatal exception: access violation

After the fix

The same OOM path now returns a clean MemoryError and no longer aborts or crashes.

Regression Test

Added a focused subprocess regression test in:

Lib/test/test__interpchannels.py

The test:

• uses _testcapi.set_nomemory(0, 1)
• exercises _interpchannels._channel_id(0)
• verifies that the subprocess exits normally
• verifies that MemoryError is propagated instead of aborting or crashing

The test is skipped under Py_TRACE_REFS, matching the existing limitations of _testcapi.set_nomemory().

Tests

Executed:

PCbuild/amd64/python_d.exe -m test test__interpchannels

Result:

== Tests result: SUCCESS ==

1 test OK.

Total tests: run=74 skipped=5

I also verified:

git diff --check

which reports no patch formatting issues (only existing LF/CRLF working-copy warnings).

NEWS

Added a NEWS fragment describing the crash fix and MemoryError propagation.

Addresses OOM-0014 from gh-151763.