Re: grep internationalization

[Paul Eggert]

> From: Isamu Hasegawa <address@hidden>
> Date: Wed, 24 Jan 2001 11:28:44 +0900 (JST)
> 
> However our patch can not handle Equivalence class correctly,
> because POSIX don't supply the function for equivalence class.

Sure it does: it's called 'regcomp' and 'regexec'.  :-) Actually, I'm
only half kidding here, as it should be possible to use the system
regcomp and regexec to come up with a compatible matcher.  It would be
some work, though.

Does your patch handle multicharacter collating elements, e.g. "aa" in
Danish?

> However our regex.c is derived from glibc, so these test
> cases also fail when I use the regex.c from glibc-2.2.

That is a bug in glibc 2.2 regex.c.  One of these days we need to fix
glibc 2.2.  In the meantime we need to used a patched regex.c.  The
latest test version of grep checks to see whether the library regex.c
is buggy, and if so, uses grep's regex.c.

>   echo '{' | grep -e '\{'
> '\{' is invalid basic regular expression, isn't it?

No, it's not invalid.  It has undefined behavior, and this means that
a portable POSIX application cannot rely on '\{' producing an error.
GNU grep defines the BRE '\{' to have the same meaning as '{'; this is
compatible with traditional BSD behavior.  More importantly, GNU egrep
defines the ERE '{' to have the same meaning as '\{'.

> Should we fix these inconvenience?

Yes, that sounds like a good idea.

Two more suggestions:

* Adapt your patch to the latest published test version of GNU grep,
  at <ftp://alpha.gnu.org/gnu/grep/grep-2.5a.tar.gz>.

* Make sure your patch is compatible with the latest draft (Draft 5)
  of the revised POSIX standard, which you can get from
  <http://www.opengroup.org/austin/>.  In particular, the semantics of
  regular expressions have been loosened a bit in d5, which will help
  you somewhat.  However, I should warn you that these changes are
  still a bit controversial and they are not final.