Fix re.findall reporting a group that did not match - #8563

re.findall hands back the matched text rather than a match object, so a group that took no part in the match is reported as an empty string. With one group it was reported as None:

>>> re.findall(r"(a)?b", "b ab")
[None, 'a']          # CPython: ['', 'a']
>>> re.findall(r"(x)?", "a")
[None, None]         # CPython: ['', '']

Anything reading the result as a list of strings gets a TypeError from "".join(...), from max(..., key=len), from a .strip() in a loop. With two groups it was already right, because that branch passes "" to Match.groups as the default while the one-group branch fell through to None. The two halves of the same function disagreed.

While pinning that down, the same line turned up a second one. The default is built as a str whatever the pattern is, so a bytes pattern comes back with str mixed into it:

>>> re.findall(rb"(a)|(b)", b"ab")
[(b'a', ''), ('', b'b')]        # CPython: [(b'a', b''), (b'', b'b')]
>>> [type(y).__name__ for x in re.findall(rb"(a)|(b)", b"ab") for y in x]
['bytes', 'str', 'str', 'bytes']

That one is on the two-group branch, so it is there on main today and is not a consequence of the first fix. Both come from the same decision, so the empty value is now built once from zelf.isbytes, using the idiom sub_impl already uses a few lines up, and both branches take it.

Match.groups() keeps reporting None, which is what CPython does and is the reason the two are easy to confuse. re.split keeps None as well. Both are covered by the tests so a later change cannot quietly pull them along.

Checked against CPython 3.14.7 over 41 cases: one group participating and not, named groups, zero groups, two and three groups, str and bytes, the compiled-pattern form, and the neighbours that share the match machinery (finditer, group, groups, groupdict, split, sub, expand). All 41 agree now. Without the change 20 of them differ.

Lib/test/test_re.py::test_re_findall only uses groups that participate, or two groups, so the suite passes either way. It still passes here, 166 tests. The new cases are in extra_tests/snippets/stdlib_re.py, and they also assert the result types, since that is where the bytes half went wrong.

Summary by CodeRabbit

• Bug Fixes

  • Fixed re.findall so unmatched capturing groups consistently return an empty value matching the pattern type: "" for text patterns and b"" for byte patterns.
  • Preserved existing behavior for participating groups, patterns without groups, and other matching APIs.

• Tests

  • Added coverage for unmatched groups across text and byte patterns, including single- and multi-group results.