assert instr.opname.startswith("CALL_") throws on Python 3.11

[lemenkov]

Hello!
Similar to this StackOverflow message I've got the same error while trying to (indirectly) use docopt-ng with Python 3.11.

I've added print(instr.opname) right after the line 941 and the resulting list of instructions indeed doesn't have anything stareted with CALL_ (see file atached). On the contrary here is a move shorter list of instructions when I run the same with Python 3.10:

LOAD_GLOBAL
LOAD_GLOBAL
LOAD_CONST
LOAD_CONST
CALL_FUNCTION_KW

If I remove the following block completely along with output_value_assigned variable then everything seems to work just fine with Python 3.11:

docopt-ng/docopt/__init__.py

Lines 937 to 950 in 130a542

	if more_magic and parent_frame:
	import dis
	instrs = dis.get_instructions(parent_frame.f_code)
	for instr in instrs:
	if instr.offset == parent_frame.f_lasti:
	break
	assert instr.opname.startswith("CALL_")
	MAYBE_STORE = next(instrs)
	if MAYBE_STORE and (
	MAYBE_STORE.opname.startswith("STORE")
	or MAYBE_STORE.opname.startswith("RETURN")
	):
	output_value_assigned = True

[NickCrews]

Thanks for the report!

It was inevitable that something would break with how we are reaching into internals like this. The whole "magic" thing was a mistake of a feature I think, oh well.

I'll take a look and try to come up with some workaround.

[lemenkov]

So, apparently CALL_FUNCTION, CALL_FUNCTION_KW, CALL_METHOD, CALL_METHOD_KW were removed from Python 3.11. See python/cpython#88691 for the details.

[lemenkov]

Thanks for the report!

It was inevitable that something would break with how we are reaching into internals like this. The whole "magic" thing was a mistake of a feature I think, oh well.

I'll take a look and try to come up with some workaround.

Thanks! Looking forward to it!

[NickCrews]

OK, so brownie is using a vendored version of docopt, so if I fix the bug here then you won't see that change unless they pull that in. I still will look for the bug and probably fix it, but FYI you are going to have to go through more steps.

[NickCrews]

OK this is fixed in #51 (by just deleting the problematic feature ;) ). You will now need to get brownie to re-pull this updated version (or perhaps depend on it from PyPI, I'm not sure why they have it vendored, but they are going to keep getting this problem otherwise)

I'll leave this open for now if you have any problems, but will close soon.