Switch to heap types

[MilesCranmer]

This is an issue to track implementation of the following helpful suggestion by @ZeroIntensity on discourse

Alternatively, it’d be nice if Julia didn’t have to use the PyTypeObject structure at all. I’m assuming you guys are using it to define native types? If so, I think it’d be a lot easier to switch to heap types (PyType_Spec), which use an array of slots instead of structure fields to avoid ABI compatibility issues.

cc @cjdoris

Hotfix is now master, to be released soon, though it’s just a temporary fix to stop the bleeding until a full switch can be implemented.

[ZeroIntensity]

Please feel free to ask me for clarification on anything! Heap types haven't been documented very well until recently.

[cjdoris]

Please feel free to ask me for clarification on anything! Heap types haven't been documented very well until recently.

Will do thanks! But I have read the docs and it all seems pretty clear.

[cjdoris]

https://github.com/JuliaPy/PythonCall.jl/actions/runs/16301726336

We're now failing with a bus error on older versions of Python, I think because we dereference PyTypeObjects in a few places, and so we're trying to access out-of-bounds memory.

So we should really switch to heap types ASAP and then remove fields from the end of PyTypeObject that we don't need.

[cjdoris]

Re my previous comment, I have a PR to hotfix that issue too.

[MilesCranmer]

Oops 😆 #644

Though @cjdoris shouldn't we completely remove the PyTypeObject? See my PR for the spots that need to be updated

[cjdoris]

Haha, snap!

Not using the struct for allocation was the big culprit. But yeah if we can remove PyTypeObject entirely that would be good - in case CPython completely changes the layout of this struct one day. Can you make a PR?