Re: [grep 2.5.1] odd output for some patterns combining backreference, alternation, and repetition

[Alex Lamey]

Hi Ralf,
Yes, that's right.  I neglected to cc the list when replying to  
Bauke, so see below.  The problem does appear to be platform-specific.
alex

> I'm running Mac OS X/Darwin.  I've confirmed that I get the same  
> output on two different Macs (G5 and Intel).  But other unix-based  
> platforms don't seem to exhibit the error.  I've also reported the  
> bug at http://savannah.gnu.org and http://developer.apple.com/ 
> bugreporter.
> alex
>
> On 2-Oct-07, at 1:22 PM, Bauke Jan Douma wrote:
>
>
> > Alex Lamey wrote on 02-10-07 09:41:
> >
> > > > Not a bug -- assuming the ! is and envisualization of your  
> > > > surprise.
> > > Yes, ! is an error marker as I wrote
> > >
> > > > The [0-9] means just one digit in that range.  So each of these
> > > > two constructs matches at most just one digit.  Hence \1 can  
> > > > only be
> > > > a reference to at most one digit.  Same for \2.  And hence the same
> > > > for \1|\2.
> > > That's what I expected.  (\1|\2) should match exactly 1 digit.   
> > > In my examples, just the digit '8'.
> > >
> > > > Therefore {1} gives you just one digit on the right side of the  
> > > > dash,
> > > > {2} gives you 2, and anything else just doesn't match, because  
> > > > you're
> > > > asking too many: {n} means /exctly/ n matches.
> > > So why does (\1|\2){4} match just one '8'?  Why does (\1|\2){6}  
> > > match two, etc.
> >
> > It doesn't output that here.
> > I am running grep 2.5.1 also.
> >
> > Are you sure that's your output??
> >
> > bjd



On 2-Oct-07, at 2:42 PM, Ralf Wildenhues wrote:

> Hello Bauke, Alex,
>
> * Bauke Jan Douma wrote on Tue, Oct 02, 2007 at 12:10:20AM CEST:
> > Alex Lamey wrote on 30-09-07 22:03:
> > > > $ echo 88-88 | egrep -o '([0-9])([0-9])-(\1|\2){3}'
> > > > 88-88
> > >     !
> [...]
> > Not a bug -- assuming the ! is and envisualization of your surprise.
> >
> > The [0-9] means just one digit in that range.  So each of these
> > two constructs matches at most just one digit.  Hence \1 can only be
> > a reference to at most one digit.  Same for \2.  And hence the same
> > for \1|\2.
> > Therefore {1} gives you just one digit on the right side of the dash,
> > {2} gives you 2, and anything else just doesn't match, because you're
> > asking too many: {n} means /exctly/ n matches.
>
> Maybe I'm being dense here, but the above example (just like the  
> others
> that Alex annotated) asks for three digits after the hyphen, but egrep
> actually matches with two.  Looks like a bug to me, no?
>
> FWIW, I can't reproduce the above with version 2.5.1 on GNU/Linux.
>
> Cheers,
> Ralf