Use _PyObject_IsUniquelyReferenced instead of `Py_REFCNT(x) == 1` for free-threaded compatibility

[sergey-miryanov]

Feature or enhancement

Proposal:

There are a lot of places where Py_REFCNT(..)==1 used. IIUC it is not correct for no-gil version.

Has this already been discussed elsewhere?

This is a minor feature, which does not need previous discussion elsewhere

Links to previous discussion of this feature:

No response

Linked PRs

• GH-140061: Use _PyObject_IsUniquelyReferenced to check if object is uniquely referenced #140062
• [3.14] gh-140061: Use _PyObject_IsUniquelyReferenced() to check if objects are uniquely referenced (gh-140062) #140157

[cmaloney]

As a heads up this migration led to a subtle free-threading only bug in _PyBytes_Resize (GH-157373). _IsUniquelyReferenced returns false when a reference crosses a thread boundary while Py_REFCNT(x) == 1 returns true. That resulted in an immortal constant object being returned because resize thought the bytes was shared (and thus immutable) in a case where code could still modify the bytes.

I used AI to spot-check the rest of the migrations here (Claude Fable 5.1) and it found one other behavior change worth noting: PySet_Add on a single reference frozenset which crosses thread boundaries now raises SystemError rather than passing.

[vstinner]

I used AI to spot-check the rest of the migrations here (Claude Fable 5.1) and it found one other behavior change worth noting: PySet_Add on a single reference frozenset which crosses thread boundaries now raises SystemError rather than passing.

IMO PySet_Add(frozenset, key) works as expected here: you must not modify a frozenset from a thread different than the one which created it. If you think that the documentation isn't explicit about this point, maybe you can suggest a change for PySet_Add() documentation.