GeezerVison

@markshannon Please take a look.

As the person who reported #134786: I would suggest making the documentation more explicit and replacing should with must here, especially since the code now errors out if Py_TPFLAGS_HAVE_GC is not set.

As the person who reported #134786: I would suggest making the documentation more explicit and replacing should with must here, especially since the code now errors out if Py_TPFLAGS_HAVE_GC is not set.

+1

Additionally, this PR puts this into a WhatsNew entry for 3.15. This is not a new thing, this is a documentation fix + fix for the runtime to not allow these invalid types to be created. So I think the added whatsnew is misleading.

I'm not sure if we should support Py_TPFLAGS_MANAGED_DICT without Py_TPFLAGS_HAVE_GC, because Py_TPFLAGS_MANAGED_DICT implies Py_TPFLAGS_HAVE_GC. However, I have added it, though, to confirm that it is doable.

I think the current version needs some updates and clarification in the docs. First, though, I would like to get some feedback on the overall approach. Is this the right way to do after all?

Please take a look if someone interested.

@markshannon Please take a look.

Maybe we should rename this PR?

I forgot to apply changes :(

Once the remaining comment has been addressed, I'm happy to merge this.

For the record, the requirements may change in the future:
For Py_TPFLAGS_MANAGED_DICT things are clear: Py_TPFLAGS_HAVE_GC must be set.
For Py_TPFLAGS_MANAGED_WEAKREF things are more complex. The current implementation requires that Py_TPFLAGS_HAVE_GC to be set, but there is no fundamental reason why the VM cannot manage the weakrefs of objects that cannot be part of cycles.

For Py_TPFLAGS_MANAGED_WEAKREF things are more complex. The current implementation requires that Py_TPFLAGS_HAVE_GC to be set, but there is no fundamental reason why the VM cannot manage the weakrefs of objects that cannot be part of cycles.

I had such commit 42c0932, we decided to revert it for now. Maybe we should open a separate PR for it?

If it isn't too much work, it is worth adding a test for this.

@kumaraditya303 I have added test.

@markshannon @kumaraditya303 It is ready to review, could you please take a look.

Thanks all!