Message-ID: <B0FEA00E82A7D1118BFB00A0CC99027821323B@ARGON>
From: Michel de Ruiter <Michel AT smr DOT nl>
To: "'Eli Zaretskii'" <eliz AT is DOT elta DOT co DOT il>
Cc: "'DJGPP workers'" <djgpp-workers AT delorie DOT com>
Subject: RE: fflush question
Date: Tue, 6 Apr 1999 13:34:44 +0200 
X-Mailer: Internet Mail Service (5.5.2448.0)
Reply-To: djgpp-workers AT delorie DOT com

> > This returns 0, unless the fsync is turned on. Is this allowed and
> > intended this way?
> Yes.

I see.

> > The ANSI C9x draft (January 18, 7.19.5.2, page 306)
> > says about fflush's functionality in this case:
> >      ... the fflush function causes any unwritten data ... to be
> >      delivered to the host environment to be written to the file ...
> If this is interpreted like you seem to suggest, then why 
> does Unix need the `fsync' system call?  Obviously, `fflush'
> doesn't deliver the data to the disk, but to the system buffers
> (aka disk cache), which then write the data at its discretion.

But if the file is read again (via another handle) the operating system
knows what to do: read from the system buffers. No?

> > I cannot find anything about one file being open as two 
> > streams (input and output) simultaneously, except (7.19.3,
> > page 301):
> The test program opens the file on ANOTHER HANDLE, and assumes
> the data will be available.  IMHO, this is non-portable. This
> is DOS that interferes here, we cannot do anything to correct it.

I understand. So it is Autoconf that is using unportable behaviour here. It
is in the check whether fflush(NULL) works.

> DJ, does Posix say anything about it?
> If not, I think this should be reported to Autoconf 
> maintainers.  They should call `fsync'.

Right.

> > Oh, now we are at it, ANSI C9x (Annex J.2, page 557) says trying to
> > fflush an input stream causes undefined behaviour
> I don't see how this is relevant, since the code you posted 
> fflush'ed an output stream, not an input stream.

I know. It's another problem.

> >     f->_flag &= ~_IOUNGETC;
> > I assume this *can* have effect on an input stream. 
> Yes.  Please submit a change for the docs.

As soon as I have time again...