C fromisoformat() accepts stray text between the time and the UTC offset

[fitness-trener]

Bug report

Bug description:

The C implementation of time.fromisoformat() and datetime.fromisoformat() accepts stray text between the time and a UTC offset or Z. The pure-Python implementation rejects the same strings:

>>> from datetime import time, datetime
>>> time.fromisoformat('12:30:45 +02:00')
datetime.time(12, 30, 45, tzinfo=datetime.timezone(datetime.timedelta(seconds=7200)))
>>> time.fromisoformat('12:30x+05:00')
datetime.time(12, 30, tzinfo=datetime.timezone(datetime.timedelta(seconds=18000)))
>>> time.fromisoformat('1230:Z')
datetime.time(12, 30, tzinfo=datetime.timezone.utc)
>>> time.fromisoformat('12:30:45.123456 junk Z')
datetime.time(12, 30, 45, 123456, tzinfo=datetime.timezone.utc)
>>> datetime.fromisoformat('2020-01-01T12:30x+05:00')
datetime.datetime(2020, 1, 1, 12, 30, tzinfo=datetime.timezone(datetime.timedelta(seconds=18000)))

>>> import _pydatetime   # Lib/_pydatetime.py from main
>>> _pydatetime.time.fromisoformat('12:30:45 +02:00')
ValueError: Invalid isoformat string: '12:30:45 +02:00'

parse_hh_mm_ss_ff() in Modules/_datetimemodule.c returns 1 ("not at end") in two cases:

• one character remains before tstr_end after HH, MM or SS;
• any text follows a fraction of 6 or more digits.

It also returns 1 for every valid string with an offset, because it reads the designator itself as c. So parse_isoformat_time() can only treat rv == 1 as an error when there is no offset. When an offset follows, the stray text is accepted silently.

A comment on gh-130959 reported the '...05.600000 +02:30' form, but that issue was closed after the pure-Python fix, and the C side is unchanged. gh-107779 ('20230808120000Z' losing a digit) goes through the same rv == 1 path.

Checked on 3.11.15 with the C accelerator (output above), and _pydatetime from main (7eada7c). The C code on main was read, not built; the code path is unchanged.

Suggested fix: have parse_hh_mm_ss_ff() return 0 when c is the terminator or the designator (p > p_end) and 1 when it is a stray character (p == p_end), and return p != p_end after the fraction. parse_isoformat_time() can then reject rv == 1 whether or not an offset follows. I have a draft patch with tests. It is not compiled yet; I checked it through a Python transliteration of the C code against the fromisoformat vectors in datetimetester.py, and every example there still parses. I can open a PR if this is accepted.

CPython versions tested on:

3.11, CPython main branch

Operating systems tested on:

Windows