bpo-44032: Move pointer to code object from frame-object to frame specials array.

[markshannon]

The f_code field is the last field in the frame object that cannot be lazily evaluated if we want to lazily create frame objects, or ignore them in the interpreter.

https://bugs.python.org/issue44032

[markshannon]

Skipping news as this is just making f_code consistent with f_globals and f_builtins.

[ericsnowcurrently]

LGTM

I left one minor suggestion for simplifying the GDB tooling. Otherwise this looks ready to go.

[ericsnowcurrently]

I suppose it doesn't matter much but it's too bad you can't use self._f_specials() here. With a small adjustment to _f_specials() this code becomes simpler and more uniform.

For example, let the caller pass the wrapper to use (and default to PyObjectPtr):

Suggested change

	f_valuestack = self.field('f_valuestack')
	code = f_valuestack[FRAME_SPECIALS_CODE_OFFSET - FRAME_SPECIALS_SIZE]
	return PyCodeObjectPtr.from_pyobject_ptr(code)
	return self._f_specials(FRAME_SPECIALS_CODE_OFFSET, PyCodeObjectPtr)

...or let _f_specials() decide based on the index (since this code is already so specific and tightly coupled):

Suggested change

	f_valuestack = self.field('f_valuestack')
	code = f_valuestack[FRAME_SPECIALS_CODE_OFFSET - FRAME_SPECIALS_SIZE]
	return PyCodeObjectPtr.from_pyobject_ptr(code)
	return self._f_specials(FRAME_SPECIALS_CODE_OFFSET)

[nedbat]

I'm confused about how I'm meant to access the code for a frame now? The struct member is gone, and the accessor function has a leading underscore. Is there a supported way to get to the code for a frame?

[nedbat]

Oh, sorry, PyFrame_GetCode, got it :)