Remove `DELETE_GLOBAL` bytecode instruction

[markshannon]

Feature or enhancement

Proposal:

DELETE_GLOBAL could be removed by making STORE_GLOBAL handle a NULL value and replacing DELETE_GLOBAL with PUSH_NULL; STORE_GLOBAL.

This wouldn't have an effect on performance or code size, as STORE_GLOBAL would also need to do the job of DELETE_GLOBAL but it would free up an opcode for future use. Also, once #145854 and #145855 are implemented, it would make the bytecode for deletion more consistent.

Has this already been discussed elsewhere?

This is a minor feature, which does not need previous discussion elsewhere

Linked PRs

• gh-145857: Replace DELETE_GLOBAL with PUSH_NULL; STORE_GLOBAL #146314

[MazinSharaf]

DELETE_GLOBAL could be removed by making STORE_GLOBAL handle a NULL value and replacing DELETE_GLOBAL with PUSH_NULL; STORE_GLOBAL.

I don't think this would exactly work, and it would probably be worse. I say this because DELETE_GLOBAL removes the global from the dictionary itself, that means its name as well. STORE_GLOBAL doesn’t do that and instead leaves the name there and only removes the value. This would be bad as well because if the names are still there in the dictionary, but have no value, then it's just a waste sitting there.

I may still test this but I'm very skeptical about making this change.

[brijkapadia]

If you make it so that a NULL value for STORE_GLOBAL corresponds to popping the variable from the global dictionary, then you would get the same behavior as DELETE_GLOBAL. The one thing that seems strange is that STORE_GLOBAL would end up removing a key from the global dictionary, despite the name implying it would set that key to have a value of NULL.

[picnixz]

It is not an issue for the naming. PyObject_SetAttr takes a NULL to indicate deletion.

[brijkapadia]

I can try to work on this change as it is a necessary feature to implement before I continue with one of my PRs (namely PR #145983). Additionally, this change is very similar to what I am currently doing in PR #146006, so I don't think I will have trouble implementing this issue once that one gets merged.