feat: new mypyc primitive for weakref.ref

[BobTheBuidler]

This PR adds a new mypyc primitive for weakref.ref

I wasn't able to figure out what name mypyc expects for weakref.proxy, so I took that out and will keep that for a separate PR later on. ref is more commonly used than proxy anyway.

for later, I tried:

• weakref.proxy
• weakref.ProxyType
• weakref.weakproxy

no luck with those

also for later, I'll need to finish #19145 to add a primitive for weakref.ref.__call__

[BobTheBuidler]

Done. This just needs tests now.

• tests

[BobTheBuidler]

@JukkaL I've cleaned up the commit history and this is now ready for review.

[JukkaL]

Thanks for the PR! Left a suggestion about how to make this more general.

[JukkaL]

Can you also support calls with a single argument, since this is probably quite common?

You'd need to add another function_op with arg_types=[object_rprimitive], and probably something like extra_int_constants=[(0, pointer_rprimitive)], to add an implicit NULL argument.

[BobTheBuidler]

This all makes sense to me, the pointer_rprimitive would become the NULL (i think?) but what does the 0 do in your above example?

[BobTheBuidler]

Even though I don't understand it, it seems to be working. Is it an index? as in 'for extra int constant in slot 0, use pointer_rprimitive'?

[JukkaL]

0 is the constant value of the argument (zero i.e. NULL).

[BobTheBuidler]

Oh great, I was trying and trying different ways to do this and eventually gave up. Will put that together sometime this week. Any thoughts on how I might determine the correct fullname for weakref.proxy so I can handle that with both cases (1- and 2-arg) at the same time?

…

On Mon, Jun 2, 2025 at 12:12 PM Jukka Lehtosalo ***@***.***> wrote: ***@***.**** commented on this pull request. Thanks for the PR! Left a suggestion about how to make this more general. ------------------------------ In mypyc/primitives/weakref_ops.py <#19099 (comment)>: > @@ -0,0 +1,13 @@ +from mypyc.ir.ops import ERR_MAGIC +from mypyc.ir.rtypes import object_rprimitive +from mypyc.primitives.registry import function_op + +# Weakref operations + +new_ref_op = function_op( + name="weakref.ReferenceType", + arg_types=[object_rprimitive, object_rprimitive], + return_type=object_rprimitive, + c_function_name="PyWeakref_NewRef", Can you also support calls with a single argument, since this is probably quite common? You'd need to add another function_op with arg_types=[object_rprimitive], and probably something like extra_int_constants=[(0, pointer_rprimitive)], to add an implicit NULL argument. — Reply to this email directly, view it on GitHub <#19099 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQ3HIHXND656N5PHKHNLJZ33BRZVZAVCNFSM6AAAAAB5JUBQO2VHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDQOBZGEYDCMZSGQ> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[BobTheBuidler]

@JukkaL I know I said I'd do it later this week, but I've already finished. This is ready for review.

[JukkaL]

Any thoughts on how I might determine the correct fullname for
weakref.proxy so I can handle that with both cases (1- and 2-arg) at the
same time?

You can look at how it's defined in mypy/typeshed/stdlib/weakref.pyi. You can also print the fullname of the relevant mypy AST node e.g. in some visitor (visit_member_expr).

[BobTheBuidler]

hmm. It shows ProxyType in the typeshed though that wasn't working for me before. Let me try this again

[BobTheBuidler]

It didn't work. I'll keep working on proxy in #19217 , this one is ready to merge if you agree

[BobTheBuidler]

I got it to work for weakref.proxy on #19217. The C code generates correctly but I believe I've exposed a bug in mypyc's reference counting as my new pytest cases fail both with the new code and with the original builtin. The weakly-proxied object is being destroyed too early, while there should still be a strong reference to it.