Date: Wed, 15 Oct 1997 13:33:16 +0200 (IST)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
To: Nate Eldredge <eldredge AT ap DOT net>
cc: djgpp AT delorie DOT com
Subject: Re: DOS ugliness affects grep
In-Reply-To: <199710150220.TAA04074@adit.ap.net>
Message-ID: <Pine.SUN.3.91.971015133251.17868T-100000@is>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Precedence: bulk


On Tue, 14 Oct 1997, Nate Eldredge wrote:

> If you use `grep' on a binary file (or on `*' when there are binary files in
> the directory), and it finds a matching "line" which contains a ^Z, it will
> give a misleading error message about ENOSPC because the count was short. It
> will also only print the first part of the "line", before the ^Z.

Thanks for reporting.  This is a known problem.  It is not limited to
`grep', though: any program that reads binary files and prints parts
thereof to the screen (e.g., `tail' from Textutils), will have the
same problem.

> This will not occur if you redirect stdout, for obvious reasons.

The reason that this doesn't happen with redirected stdin is that
`grep' switches it to binary mode, which makes ^Z not special to DOS
I/O.  `grep' avoids switching the console to binary mode, because
doing so has undesirable side-effects (e.g., Ctrl-C doesn't work, so
you can't easily interrupt a runaway program).

> I realize this is DOS's fault and not DJGPP's, and no workarounds come
> immediately to mind.

A workaround is of course to redirect the output to a file or a pipe
and examine redirected output with a browser (such as Less) which can
display binary files.

Other than that, the only solution to this problem would involve a lot
of code in low-level library functions, because we do want `_write' to
return ENOSPC when it fails to write everything (otherwise, a program
that copies files to a floppy would loop forever when the floppy fills
up).  I'm not sure that working around this is worthwhile, but if
somebody feels like working on that, please do.