raforg/rawhide

find files using pretty C expressions

rawhide's Issues

.re hard to use reliably

rh -e '"\.txt$".re'

Matches on a file called foo.txt but also on one called $'foo.txt\nbar.exe' as the PCRE2_MULTILINE (m) flag is enabled by default.

rh -e '"(?-m)\.txt$".re'

Helps but that still matches on a file called $'foo.txt\n' as the PCRE2_DOLLAR_ENDONLY flag is not enabled.

The usual work around (also needed in commands such as perl-based rename implementations) is to use \z in place of $ which always matches at the end of the subject even if PCRE2_MULTILINE is enabled and/or PCRE2_DOLLAR_ENDONLY is not enabled:

rh -e '"\.txt\z".re'

But then I find that it still matches on a file called $'foo.txt\x80bar.exe'. Possibly linked to the PCRE2_MATCH_INVALID_UTF flag.

That sounds like it could be a bug in PCRE2 but I can't reproduce it with pcre2grep -U.

To match on files whose name ends in .txt, I find that I need:

rh -e '"\.txt\z$".re'

Or:

RAWHIDE_PCRE2_NOT_UTF8_DEFAULT=1 rh -e '"\.txt\z".re'

It seems to me pcre2 should be invoked with different default options when it's meant to match on lines (like in .rebody) where using PCRE2_MULTILINE only would make sense and when meant to match on whole strings (like in .re) where PCRE2_DOTALL|PCRE2_DOLLAR_ENDONLY would be more useful (and with PCRE2_EXTENDED|PCRE2_EXTENDED_MORE|PCRE2_NO_AUTO_CAPTURE|PCRE2_UTF|PCRE2_MATCH_INVALID_UTF in common as required).

---

Note that:

rh -e '"*.txt"'

Would fail to match on $'foo\x80bar.txt' in a UTF-8 locale with the fnmatch() of older versions of glibc which you'd work around by using LC_ALL=C rh -e '"*.txt"' (same as for find -name (or find -regex where supported)). It matches in newer versions but I believe it's because/thanks of that bug.

---

Context

Here tested with the current git HEAD of rawhide (a1c44e2) on Debian GNU/Linux amd64 with

libc6 2.37-12
libpcre2-8-0 10.42-4