It is possible to construct classes without__firstlineno__:

deff():__firstlineno__=1classX:nonlocal__firstlineno__returnXprint(f().__firstlineno__)# AttributeError

Or:

classY:global__firstlineno__print(Y.__firstlineno__)# AttributeError

This is obviously an extreme edge case though. You could also do this by manipulatinglocals() in the class body or the metaclass.

Is it worth it to makeinspect smarter in the presence of classes without__firstlineno__?

  File "/Users/jelle/py/cpython/insp.py", line 18, in <module>    print(inspect.findsource(X))          ~~~~~~~~~~~~~~~~~~^^^  File "/Users/jelle/py/cpython/Lib/inspect.py", line 1070, in findsource    return lines, object.__firstlineno__ - 1                  ^^^^^^^^^^^^^^^^^^^^^^AttributeError: type object 'X' has no attribute '__firstlineno__'

Good point. I thought that it is the same as for__module__ and__qualname__, but it turned out that they are set implicitly in thetype constructor even when not specified in the class dict. They are set even fortype('A', (), {}).

We can try to set__firstlineno__ to the current line number in the top Python frame, but it is safer to not set it at all. Setting__firstlineno__ for classes created by thetype() call rather than in theclass statement does not help in the only use case for this attribute.

I think that for backward compatibilitygetsource() should convert an AttributeError into OSError which was raised before for classes created by thetype() call.