Re: texinfo-5.9.90 pretest available

[Ken Brown]

On 2/27/2015 2:00 AM, Eli Zaretskii wrote:
> > Date: Thu, 26 Feb 2015 21:34:03 +0000
> > From: Gavin Smith <address@hidden>
> > Cc: Eli Zaretskii <address@hidden>, Karl Berry <address@hidden>, Texinfo 
> > <address@hidden>
> >
> > I checked the 5.2 release and it was done differently. The code looked like:
> >
> >    if (*compression_program)
> >      { /* It's compressed, so fclose the file and then open a pipe.  */
> >        char *command = concat (*compression_program," -cd <", *opened_filename);
> >        if (fclose (f) < 0)
> >          pfatal_with_name (*opened_filename);
> >        f = popen (command, "r");
> >        if (f)
> >          *is_pipe = 1;
> >        else
> >          pfatal_with_name (command);
> >      }
> >
> > The ChangeLog entry on 2014-02-19 explains:
> >
> >          * install-info/install-info.c (open_possibly_compressed_file):
> >          Work even if the file name contains arbitrary shell
> >          metacharacters, for example:
> >            install-info --info-dir="/d/a b/info" "/d/a b/info/emacs.info.gz"
> >          Do this by running the decompressor on standard
> >          input, rather than by having the shell open the file.
> >          Return either stdin or a pipe.
> >          Don't bother with IS_PIPE arg; no longer needed.
> >          All callers changed.  Check for freopen failure.
> >
> > I wonder if there is a problem with Cygwin with two processes having
> > the same file open at once, or else.
>
> Ken, could this have something to do with text vs binary I/O mode?
> Perhaps the stdin that gzip inherits is not in binary mode, or
> something like that?

No, it turns out that it has to do with buffered I/O.  See the thread starting 
here:

  https://cygwin.com/ml/cygwin/2015-02/msg00887.html

Replacing the fseek in line 841 by another call to freopen fixes the problem.  I 
don't know if that's the best solution.

Ken