gh-86542: New C-APIs to simplify module attribute declaration

[tiran]

• Check for reference leaks
• verify refcounts.dat
• add CAPI tests

Signed-off-by: Christian Heimes christian@python.org

https://bugs.python.org/issue42376

• Issue: Add helpers to populate modules in C #86542

[vstinner]

I'm not sure how to make this member future-proof in term of stable ABI.

We should try to put a max_align_t inside, but this type requires C11. GCC defines it with:

typedef struct{long long __max_align_ll __attribute__((__aligned__(__alignof__(long long)))); long double __max_align_ld __attribute__((__aligned__(__alignof__(long double)))); /* _Float128 is defined as a basic type, so max_align_t must be sufficiently aligned for it. This code must work in C++, so we use __float128 here; that is only available on some architectures, but only on i386 is extra alignment needed for __float128. */ #ifdef __i386__ __float128 __max_align_f128 __attribute__((__aligned__(__alignof(__float128)))); #endif } max_align_t; 

Maybe we can at least put long long and long double:

// Unused members added to make PyModuleConst_Def large enough // to get a stable ABI and support future additions. long long m_long_long; long double m_long_double; 

[encukou]

This is the only issue I found here :)
The stable ABI doesn't have a good story around evolving structs, and I think we should design a general mechanism for that rather than try to future-proof individual structs.
To move this PR forward, could we exclude PyModule_AddConstants & co. from the limited API for the time being?

[vstinner]

Could you start by adding these two functions in a separated PR, to make this PR shorter?

[encukou]

I don't really think that's necessary; the PR is not that big.

[shihai1991]

nitpick: static PyObject * in here.

[shihai1991]

How about using another sepeareted PR to do this converation operation?

[github-actions]

This PR is stale because it has been open for 30 days with no activity.

[encukou]

This just needs a bit of polish, and checking the refleaks/refcounts as the OP suggests.
@tiran, do you plan to work on this? Should I take over?

[encukou]

Suggested change

	The values for *type* and the definition of the *value* union are
	internal implementation details. Use any of the ``PyModuleConst_`` macros
	to define entries. The array must be terminated by an entry with name
	set to ``NULL``.
	.. c:member::constchar *name
	Attribute name.
	.. c:member::int type
	Attribute type.
	.. c:member::void *value
	Value of the module constant definition, whose meaning depends on
	*type*.
	Definition of a module constant.
	The members of this struct are internal implementation details.
	To define entries, use only the ``PyModuleConst_`` macros below,
	and ``{NULL}`` to mark the end of the array.

[encukou]

These two functions need tests.

[encukou]

Hm, I see these return borrowed references (held by the module object). That saves a DECREF if you don't need the new object, but it'll be problematic for HPy.

[encukou]

I'll close this as it's not likely to get updated any time soon.
If anyone wants to pick this up, see the issue.