Date:	Tue, 12 Jan 1999 09:47:55 +0100
From:	Laszlo Molnar <laszlo DOT molnar AT eth DOT ericsson DOT se>
To:	Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
Cc:	DJGPP Workers List <djgpp-workers AT delorie DOT com>
Subject:	Re: djgpp 2.02 + perl + glob
Message-ID:	<19990112094755.O29345@duna54>
References:	<19990111094726 DOT F29345 AT duna54> <Pine DOT SUN DOT 3 DOT 91 DOT 990111121821 DOT 14237A-100000 AT is>
Mime-Version:	1.0
X-Mailer:	Mutt 0.94.17i
In-Reply-To:	<Pine.SUN.3.91.990111121821.14237A-100000@is>; from Eli Zaretskii on Mon, Jan 11, 1999 at 12:25:06PM +0200
Reply-To:	djgpp-workers AT delorie DOT com

On Mon, Jan 11, 1999 at 12:25:06PM +0200, Eli Zaretskii wrote:
> > djdev202, I've found a problem. Perl failed on some of its self tests
> > which use glob(). But when I linked the executable with libc.a from
> > djdev201, it worked fine. So I guess, there may be a bug in glob() in
> > 2.02.
> The diffs between glob.c in v2.01 and v2.02 are appended.  I don't see 
> anything that seems like a bug here.  Do you?

No, glob() seems to be perfect.

> Maybe Perl is counting on the behavior in case globbing failed, which now 
> requires a special flag (see the last hunk of changes below)?
> Of course, some of the functions that `glob' calls might be the cause.

Ok, I've found the problem which is in the fsext stuff (perl's globbing
uses this too for builtin globbing). The problem: when a new handle is
allocated, it is not opened, but dup()-ed in djgpp 2.02. It would work
perfectly if the file pointer would be set to 0, ie an lseek(..,0,0)
should be added after dup() IMHO.

Laszlo

- Raw text -