bug#22820: grep: Misleading error message when presenting a badly formed character class

[Jim Meyering]

On Fri, Feb 26, 2016 at 6:51 AM,  <address@hidden> wrote:
> Hi,
>
> I'd like to forward a bug filled by Gunnar Wolf in Debian some time
> ago:
>
> "It seems that whenever egrep finds something it cannot digest inside a
> character class, it spews out the same error string: «Unmatched [ or [^».
> This can be misleading and opens the way for long debugging time,
> specially when trying to understand complex regexes. To illustrate the
> point:
>
> $ echo | egrep -v '[[:digit]]+'
> egrep: Unmatched [ or [^
>
> The brackets _are_ balanced, however the character class is not (it
> lacks a finishing colon)."

Thank you for forwarding that.
The diagnostic was fixed in gnulib via a commit last month:
git.savannah.gnu.org/cgit/gnulib.git/commit/?id=7c6e85cf4eccbd5129
Thus, as long as grep is configured --with-included-regex,
you will now see this:

  $ grep -E '[[:digit]]+'
  grep: Unmatched [, [^, [:, [., or [=

If grep is configure with --without-included-regex, you will
still see the inferior diagnostic, but that string will then be
coming from your system's C library.

Since fixing glibc's copy of regcomp.c is outside the scope
of grep's issue tracker, I'm marking this as "done".