Date: Mon, 17 Jan 2000 13:08:54 +0200 (IST)
From: Eli Zaretskii <eliz AT is DOT elta DOT co DOT il>
X-Sender: eliz AT is
To: Laurynas Biveinis <lauras AT softhome DOT net>
cc: djgpp-workers AT delorie DOT com
Subject: Re: Bash 2.03 prerelease now available
In-Reply-To: <388208F5.9D20B750@softhome.net>
Message-ID: <Pine.SUN.3.91.1000117130831.22412C-100000@is>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
Reply-To: djgpp-workers AT delorie DOT com
Errors-To: dj-admin AT delorie DOT com
X-Mailing-List: djgpp-workers AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk


On Sun, 16 Jan 2000, Laurynas Biveinis wrote:

> Eli Zaretskii wrote:
> > Btw, I understand that the configure script was produced by the DJGPP
> > port of Autoconf.  If so, that port needs more work, because the
> > produced configure script still has some of the Unix-centric
> > constructs that break on DOS/Windows.  For example, the test for
> > absolute name near ac_given_INSTALL doesn't support drive letters.  (I
> > can post a full list of problems, if someone is interested.)
> 
> I would be interested, because my first step in porting any autoconf'ed
> program to DJGPP is regenerating configure script.

With the current Bash port and its TEST_FINDS_EXE feature, why is it
even necessary to generate a configure script?  The ported Autoconf
doesn't seem to do much more than handle the test -f problem.

Anyway, here's the list of additional gotcha's in configure scripts
and Makefile's they produce:

 1) The configure script includes fragments 

      case "$ac_given_srcdir" in
      .)  srcdir=.
	  if test -z "$ac_dots"; then top_srcdir=.
	  else top_srcdir=`echo $ac_dots|sed 's%/$%%'`; fi ;;
      /*) srcdir="$ac_given_srcdir$ac_dir_suffix"; top_srcdir="$ac_given_srcdir" ;;
      *) # Relative path.
	srcdir="$ac_dots$ac_given_srcdir$ac_dir_suffix"
	top_srcdir="$ac_dots$ac_given_srcdir" ;;
      esac

      case "$ac_given_INSTALL" in
      [/$]*) INSTALL="$ac_given_INSTALL" ;;
      *) INSTALL="$ac_dots$ac_given_INSTALL" ;;
      esac

    The "/*)" and "[/$]*)" wildcards don't support DOS-style file
    names with drive letters.  This breaks the script if you try to
    configure the package from a directory on a different drive.

 2) The following line also breaks when a drive letter is present in
    file names:

      ac_file_inputs=`echo $ac_file_in|sed -e "s%^%$ac_given_srcdir/%" -e "s%:% $ac_given_srcdir/%g"`

    To make it work, the order of the two -e commands passed to Sed
    should be reversed: first remove the (artificially added) colon,
    *then* put back the leading directories.

 3) The file-name changes such as config.h.in -> config.h-in,
    Makefile.in.in -> Makefile.in-in, etc. must be done in the
    Makefile's as well (to prevent problems like the one we just
    discussed).  One way of doing this is to add some Sed commands to
    the part of the configure script that edits Makefile.in into
    Makefile.

 4) Some Makefile's, notably those in the test suite supplied with the
    package, set PATH before invoking programs, like this:

      foo: bar
	   PATH=$(here):"$PATH" make_foo_from_bar

    You need to edit : into ";" (note the quotes!).  There's a similar
    problem with TEXINPUTS (in the doc subdirectory).

    These two are actually Automake problems, but my attempts to get
    Automake changed, backed by some very helpful people like Alain
    Magloire and Francois Pinard, didn't have any visible success yet,
    so for now this needs to be done either by Autoconf or manually.

    Again, the way I do it is to add suitable Sed commands to the
    modified configure script, so these changes are done when
    Makefile's are generated from Makefile.in's.