bug-grep
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Patch: non-empty matches following empty ones


From: Charles Levert
Subject: Re: Patch: non-empty matches following empty ones
Date: Tue, 23 Aug 2005 11:22:08 -0400
User-agent: Mutt/1.4.1i

* On Tuesday 2005-08-23 at 16:24:05 +0200, Stepan Kasal wrote:
> 
> Sure, I understood that.  That's what
>       sed 's/a*/X/g'
> or
>       awk 'gsub(/a*/,"X")'
> does.
> 
> Your arguments seem to show that grep should print only lines
> with _nonempty_ matches.  (Empty pattern might be an exception.)
> But that's too late to change.

Hmmm.  I may require more thought before we
commit to anything.  To state explicitly what
you probably already understood (I'm thinking out
loud here), the way grep currently works is:

   -- To first identify which lines contain
      matches (both empty and non-empty) so
      that in the case of a 'a*'-type pattern,
      every line will match at that step.
      I assume that's the part you say is too
      late to change.

   -- Then, since there are no non-matching
      lines in that case anyway, all line are
      screened for matches to display in both
      the --only-matching and --color cases
      (otherwise they are just displayed right
      away).  The question at this step is only
      non-empty matches or all matches regardless
      of their length.

> With current setup, the inconsistency will show up because
>       grep --colour 'a*'
> will then print lot of lines without any colour tags.
> It might be confusing for X-ray robots who expects to see the
> tags...

Especially for non-empty matches.

> I wanted to prefer consistency.  Then I'd have to teach people that they
> shouldn't use patterns which match empty string.

Indeed, that's what users should really do.
Let's warn about it in the docs.

> You wanted to change the specification of --only-matching (--colour),
> so that it prints (colorifies) only non-empty matches.

Yes.  Note that there is in fact no specification
since this is a GNU extension that POSIX or
others don't know or say anything about.
Hence probably the current implementation
which also shows the inconsistency, but with a
difference in degree.

> OK, I no longer object to it.  But please change the decumentation
> (info and manpage) so that it states this clearly.

Ok.  I will put this under --only-matching and --color.




reply via email to

[Prev in Thread] Current Thread [Next in Thread]