[C API] Private functions used by mypyc

[cdce8p]

I've been working on adding support for 3.13 to mypyc. After updating the codebase, there are still some functions left we depend on which are now private after #106320. Would it make sense to export these again? That would make adopting 3.13 a bit easier. /CC @vstinner

• _PyBytes_Join defined in pycore_bytesobject.h
• _PyObject_CallMethodIdObjArgs defined in pycore_call.h
• _PyType_CalculateMetaclass defined in pycore_object.h
• _PyUnicode_EQ defined in pycore_unicodeobject.h
• _PyUnicode_FastCopyCharacters defined in pycore_unicodeobject.h (this could be replaced with PyUnicode_CopyCharacters but it's used in a simplified copy of _PyUnicode_JoinArray which also uses _PyUnicode_FastCopyCharacters).

Other functions include these two. However, as they are inlined, it isn't a problem to include them.

• _PyObject_CallMethodIdNoArgs and _PyObject_CallMethodIdOneArg defined in pycore_call.h

It would be fine to keep all of them in internal/pycore_* and just add PyAPI_FUNC.

/CC @hauntsaninja

Linked PRs

• gh-121489: Export private _PyBytes_Join() again #122094
• [3.13] gh-121489: Export private _PyBytes_Join() again #122266
• gh-121489: Export private _PyBytes_Join() again #122267
• [3.13] gh-121489: Export private _PyBytes_Join() again (GH-122267) #122287

[vstinner]

_PyBytes_Join defined in pycore_bytesobject.h

You can call something like:

PyObject *empty = PyBytes_FromString("");
PyObject *res = PyObject_CallMethod(empty, "join", ...);
Py_DECREF(empty);

_PyObject_CallMethodIdObjArgs defined in pycore_call.h

We are trying to get rid of _Py_IDENTIFIER() API. You should use your own mecanism for constant strings.

You can call PyObject_CallMethodObjArgs().

_PyType_CalculateMetaclass defined in pycore_object.h

For this one, I don't know. You might reimplement it, use the internal C API, or we should expose it as a public C API.

_PyUnicode_EQ defined in pycore_unicodeobject.h

You can use PyUnicode_Compare(). Maybe we should expose it.

_PyUnicode_FastCopyCharacters defined in pycore_unicodeobject.h (this could be replaced with PyUnicode_CopyCharacters but it's used in a simplified copy of _PyUnicode_JoinArray which also uses _PyUnicode_FastCopyCharacters).

Would you mind to elaborate your use case? Can you use the new https://docs.python.org/dev/c-api/unicode.html#pyunicodewriter API instead?

[cdce8p]

_PyUnicode_FastCopyCharacters defined in pycore_unicodeobject.h (this could be replaced with PyUnicode_CopyCharacters but it's used in a simplified copy of _PyUnicode_JoinArray which also uses _PyUnicode_FastCopyCharacters).

Would you mind to elaborate your use case? Can you use the new https://docs.python.org/dev/c-api/unicode.html#pyunicodewriter API instead?

It's used here https://github.com/python/mypy/blob/4e3346ee1dee83868adc2411c4a0f4050cf2f95a/mypyc/lib-rt/str_ops.c#L120 in a function which basically reimplements _PyUnicode_JoinArray(). Guess it would work if we replaced it with PyUnicode_CopyCharacters.

[vstinner]

It's used here https://github.com/python/mypy/blob/4e3346ee1dee83868adc2411c4a0f4050cf2f95a/mypyc/lib-rt/str_ops.c#L120 in a function which basically reimplements _PyUnicode_JoinArray().

IMO PyUnicodeWriter would be a good fit for this function, especially if you can compute the exact output string length in advance.

[vstinner]

cc @erlend-aasland @serhiy-storchaka

[vstinner]

_PyBytes_Join defined in pycore_bytesobject.h

I created issue gh-121645: [C API] Add PyBytes_Join() function.