gh-146448: Add regression test for != followed by a format spec in t-strings

[joshuaswanson]

t'{0!=0:}' used to produce an interpolation whose expression was 0, because the lexer treated the ! of != as the start of a conversion specifier.

The lexer refactor on main already handles this case. _PyLexer_ftstring_punctuation peeks at the character after ! and leaves != alone.

This PR originally added the same lookahead in Parser/lexer/lexer.c. After merging main that change is redundant, so the PR now only adds a regression test in Lib/test/test_tstring.py for t"{0!=0:}" and t"{0!=0:s}".

The branch still contains the NEWS entry from the original fix. I can remove it if a test-only change should skip news.

• Issue: parser: cpython parser bug in template strings t'{0!=0:}' #146448

[Shrey-N]

Hiya @joshuaswanson, Great fix! just a small detail,

Shouldn't we peek at the buffer directly here?
I believe this avoids the overhead and potential lineno desync of tok_nextc/tok_backup if a buffer refill happens between ! and =:

if (c == '!' && (tok->cur >= tok->inp || *tok->cur != '=')) {
    is_punctuation = 1;
}

[github-actions]

This PR is stale because it has been open for 30 days with no activity.