bug#16777: [PATCH] Revert "grep: DFA now uses rational ranges in unibyte

[Aharon Robbins]

Hi Paul.

> > As a point of information, it does happen for gawk.
>
> Could you please say where that happens?  I just now looked at the gawk 
> trunk, and the only two places I saw it calling dfaexec 
> (helpers/testdfa.c and re.c), it passed a nonnull backref argument.

OK, I was wrong and spoke without checking. My point was that the code
in question is buggy for gawk.

I suspect that the reason gawk always passes non-null backref may be
that at one time the pointer had to be valid.

I tried it by passing a NULL pointer and only one of the tests in the
test suite fails; no time right now to investigate more, but it might
be possible for gawk to pass NULL and have it work.

> Obviously this revert does reintroduce the bug with [\]-\]] that you 
> mentioned with Awk regular expressions.  It reintroduces similar bugs 
> for GNU grep.  At the very least we should fix that. I'll follow up with 
> a patch.

Looking forward to this.

Grep has general RRI problems.  I think the grep team needs to decide if
RRI is important for grep or not, and if so whether it is more important
than collating sequences and equivalence characters.

I think everyone knows how I feel about these issues. If not, I will
expound further.

Thanks,

Arnold