bpo-40679: Fix _PyEval_EvalCode() if qualname is NULL

[vstinner]

_PyEval_EvalCode() now uses co->co_name for error messages if
qualname is NULL (ex: when PyEval_EvalCodeEx() is called).

https://bugs.python.org/issue40679

[cjerdonek]

Rather than putting this "fix" in _PyEval_EvalCode(), what about the idea of documenting that the qualname (or code_name) argument can't be NULL (and asserting that), and updating the callers instead? There are only a couple callers, and it will put the messiness closer to the source of the issue.

[vstinner]

qualname is NULL in PyEval_EvalCodeEx. I don't think that it would be possible to pass co->co_name as qualname in such case.

[vstinner]

qualname parameter of PyEval_EvalCode() is passed to PyCoro_New() for example. Extract of its code:

 if (name != NULL) gen->gi_name = name; else gen->gi_name = ((PyCodeObject *)gen->gi_code)->co_name; Py_INCREF(gen->gi_name); if (qualname != NULL) gen->gi_qualname = qualname; else gen->gi_qualname = gen->gi_name; Py_INCREF(gen->gi_qualname); 

Note: In PyEval_EvalCodeEx(), name is also NULL.

Or maybe we can set name and qualname to a more useful value than NULL in PyEval_EvalCodeEx(), I don't know.

[cjerdonek]

Yeah, that's why I was saying "(or code_name)" in my comment (i.e. _PyEval_EvalCode()'s argument can be changed to code_name). I'm not sure the "fix" is needed outside of PyEval_EvalCodeEx(), which is why I was suggesting it may be better there.

[cjerdonek]

(I don't know the implications for PyCoro_New() though.)

[vstinner]

Even if it's "private" (_Py) prefix, _PyEval_EvalCodeWithName() is exposed in the C API and can be called with qualname=NULL. So I don't think that it's worth it to attempt to leave _PyEval_EvalCode() unchanged: I consider that _PyEval_EvalCode() is the right place to fix the issue.

Anyway, I wrote PR #20615 which tries to change less code.

[cjerdonek]

Thanks for discussing with me. I approved the other PR, but either is fine (not sure which you prefer at this point).

[vstinner]

I close this PR in favor of PR #20615.