Re: Bug in Glibc Regex

[Andreas Pour <pour at mieterra dot com>]

Ulrich Drepper wrote:
> 
> Andreas Pour <pour@mieterra.com> writes:
> 
> > Really?  I don't get that.  Are you sure you used the second attachment
> > I sent (the first one was the one that does work with 2.1.3 -- the one
> > that doesn't has the line:
> >
> >   const char *text = "{test}";
> 
> ??? This cannot work.  Why should it?  { and } are special characters
> and have to be escaped.  The behavior for above expression is
> undefined.  Glibc is correct, your code not.

Well, it worked under prior versions of glibc.  And my understanding of
'{' and '}' is that they are special only when followed by a number;
that is how it is defined in the regex man page:

       A bound is `{' followed by an  unsigned  decimal  integer,
       possibly  followed  by  `,'  possibly  followed by another
       unsigned decimal integer, always  followed  by  `}'.

Notice how other metacharacters do not have the "followed by" anything
description.  For example:

       A  bracket  expression is a list of characters enclosed in
       `[]'.

Moreover, a regex is defined as branches.  A branch as a "piece"
optionally followed by a "bound".  In the example, there is no "piece"
before the "bound".  So the two conditions prescribed for a bound --
preceded by a "piece" and succeeded by an "unsigned decimal integer" --
are not present in the "{test}" case.  Thus there is no reason to
believe that it would be interpreted as a "bound".  Can you give a
reason why it should be?

And that's how I've seen it implemented in other regex engines.  As I
mentioned in my initial report, I don't have the POSIX spec available,
so I don't know if it contradicts the man page and the way everyone else
-- including prior versions of glibc -- implement regexes.

Ciao,

Andreas Pour

http://www.kde.com/ :  Everything KDE
http://apps.kde.com/:  The Latest in KDE Applications