bpo-42064: Offset arguments for PyObject_Vectorcall in the _sqlite module

[encukou]

This weird mechanism allows e.g. methods to be called efficiently by providing space for a "self" argument; see PY_VECTORCALL_ARGUMENTS_OFFSET docs.

https://bugs.python.org/issue42064

[corona10]

Out of curiosity, How efficient? is there any benchmark?

[encukou]

Out of curiosity, How efficient? is there any benchmark?

I'm not aware of one, but the docs say this is encouraged :)

[erlend-aasland]

LGTM, thanks for spotting this! This makes sense for the collation callback, but I can't see how the calls in new_statement_cache and get_statement_from_cache will benefit from this. OTOH, it won't hurt to use the encouraged calling convention :)

[erlend-aasland]

Could you please break lines according to PEP 7? I'm striving to keep all sqlite3 changes PEP 7 compliant :)
Ditto for the other calls.

[erlend-aasland]

That is, either:

Suggested change

	PyObject*inner=PyObject_Vectorcall(
	lru_cache, args+1, 1 | PY_VECTORCALL_ARGUMENTS_OFFSET, NULL);
	PyObject*inner=PyObject_Vectorcall(lru_cache, args+1,
	1 | PY_VECTORCALL_ARGUMENTS_OFFSET,
	NULL);

or:

Suggested change

	PyObject*inner=PyObject_Vectorcall(
	lru_cache, args+1, 1 | PY_VECTORCALL_ARGUMENTS_OFFSET, NULL);
	size_tnargs=1 | PY_VECTORCALL_ARGUMENTS_OFFSET;
	PyObject*inner=PyObject_Vectorcall(lru_cache, args+1, nargs, NULL);

[encukou]

Thanks for the reviews!