From: "Juan Manuel Guerrero" <ST001906 AT HRZ1 DOT HRZ DOT TU-Darmstadt DOT De>
Organization: Darmstadt University of Technology
To: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
Date: Sat, 24 Feb 2001 19:13:37 +0200
MIME-Version: 1.0
Content-type: text/plain; charset=US-ASCII
Content-transfer-encoding: 7BIT
Subject: Re: gettext pretest available
CC: Bruno Haible <haible AT ilog DOT fr>, djgpp-workers AT delorie DOT com
X-mailer: Pegasus Mail for Windows (v2.54DE)
Message-ID: <2E9DA485A1B@HRZ1.hrz.tu-darmstadt.de>
Reply-To: djgpp-workers AT delorie DOT com

Date: Sat, 24 Feb 2001 10:21:59 +0200
From: "Eli Zaretskii" <eliz AT is DOT elta DOT co DOT il>

> > Unfortunately I do not understand this objection.
> > The goal of the macro is to *inhibit* the switching of *every* file handle
> > conected to a console.
>
> The key to unlock this puzzle is in setmode's source: it treats handle
> 0 specially, and its test for which handles to put into raw (as
> opposed to cooked) mode is different from the test in isatty.  The
> result is that isatty and setmode don't really agree except for
> stdin/stdout.  The danger of using isatty for anything other than
> stdin/stdout is that it might cause the call to setmode to be bypassed
> when you don't want that.

Thank you for the advice. I really did not know this.
I assume that some day I must read and understand the whole libc sources.


> Whether this side effect (of never switching any handle connected to
> the console into binary mode) is okay or not can only be answered by
> someone who knows what the application needs to do; that's not me.  In
> addition, even if today the application doesn't need that, those needs
> could change in the future.  That is why I suggested a comment near
> the macro's definition to that effect.

Only FYI: there are two occurrences of setmode in the whole package.
1) src/msgfmt.c. Here stdout is switched to O_BINARY and stdout is
   really stdout (1).
2) src/msgunftm.c. Here stdin is switched to O_BINARY. It is really stdin (0).
   In this case, the macro is a *must have*. The reason is the way msgunfmt
   evaluates the command line.
   On gettext 0.10.35 the user had to explicitely write a dash for stdin.
   On gettext 0.10.36 the user can write a dash or *nothing* for stdin.
   This has the concequence that a newbie that types the command:

PROMPT> msgunfmt  CR

   will have to reboot the machine. No arguments to msgunfmt always implies:
   read from stdin. Now, there is no way to terminate the programm.
There is no other occurence of setmode in the sources *for the moment*.
Now it is clear that an appropiate comment *must* and will be added to the
macro in lib/system.h.

Regards,
Guerrero, Juan Manuel