gh-90370: avoid temporary tuple creation for vararg in AC

[skirpichev]

Current patch partially address issue, working in case when all arguments either positional-only or vararg.

Objects/setobject.c and Modules/gcmodule.c adapted. This fixes slight performance regression for set methods, introduced by #115112, e.g.:

Benchmark	ref	patch
set().update(s1, s2)	354 ns	264 ns: 1.34x faster

Benchmark code:

import pyperf
s1, s2 = {1}, {2}
runner = pyperf.Runner()
runner.bench_func('set().update(s1, s2)', set().update, s1, s2)

• Issue: Avoid temporary varargs tuple creation in argument passing #90370

[erlend-aasland]

This also converts gcd(), lcm() and hypot() functions of the math module to the Argument Clinic. Fix issue #101123.

I'd prefer if we could keep those changes separate. Let's nok mix features (adapt methods/functions to Argument Clinic) and performance improvements (avoid temporary tuple creation for varargs in Argument Clinic).

[skirpichev]

I'd prefer if we could keep those changes separate.

Ok, reverted. Edit: description adjusted.

CI is broken

[erlend-aasland]

Re f179557: yes, you want make clinic-tests 😎

[erlend-aasland]

I'll let this sit a couple of days to give @pablogsal a chance to chime in on the gc module change.

[skirpichev]

gc module change.

I think this one is mostly mechanical change. Temporary tuple creation was moved from AC-generated code to gc_get_referrers_impl/gc_get_referents_impl.

[erlend-aasland]

I think this one is mostly mechanical change. Temporary tuple creation was moved from AC-generated code to gc_get_referrers_impl/gc_get_referents_impl.

Yes, but still it is good practice to give code owners a chance to have a look.

[skirpichev]

The rest of changes (math module): #126235