Fix internal error when source is modified

[solaluset]

Hello. I develop a project that heavily depends on code modification and monkeypatching: https://github.com/solaluset/pwcp/tree/pytest-broken

In some cases when things get wrong (as replicated in pytest-broken branch), pytest terminates with a long traceback of INTERNALERROR> IndexError: list index out of range. Change in this PR ensures end will never be out of range. With this fix applied, pytest properly shows that the error is TypeError: _maybe_compile() takes 3 positional arguments but 4 were given.

This is probably a niche case, but extra out-of-bounds check won't hurt anyone, right?

[nicoddemus]

Thanks @solaluset!

Usually we request a test, specially to avoid possible regressions in the future. Can you think of a simple test case to add to the test suite with that intent?

[solaluset]

Thank you for your reply. I thought about that, but not sure what would be a minimal reproducible example. I'll look into it.

[solaluset]

Uh, pre commit messed it up

[solaluset]

This is the reason why first line of function must start with non-space or be empty

pytest/src/_pytest/_code/source.py

Lines 205 to 208 in 4253f98

	# If we start with an indented line, put blockfinder to "started" mode.
	block_finder.started = (
	bool(source.lines[start]) and source.lines[start][0].isspace()
	)

Otherwise end gets corrected and doesn't raise IndexError.
This is so tricky, I'm not sure if my solution is optimal.

[solaluset]

@nicoddemus do I need to do anything else?

[Pierre-Sassoulas]

Hey @solaluset, thank you for the PR and the automated test. Shouldn't getstatementrange_ast be fixed at the source instead of special casing a first empty line elsewhere ? Also could you add a changelog, please ?

[solaluset]

Thanks for the response. I'm not sure I get your first point. Do you mean block_finder.started should always be false? As for the changelog, do I use the id of this PR?