Message-Id: <200008111015.GAA18203@delorie.com>
Date: Fri, 11 Aug 2000 12:51:40 +0200
To: tr AT eth DOT net
X-Mailer: Emacs 20.6 (via feedmail 8.2.emacs20_6 I) and Blat ver 1.8.5b
From: "Eli Zaretskii" <eliz AT is DOT elta DOT co DOT il>
CC: djgpp-workers AT delorie DOT com, meyering AT ascend DOT com
In-reply-to: <Pine.LNX.4.21.0007250651010.823-100000@midpec.com> (message from
	Prashant TR on Tue, 25 Jul 2000 06:59:03 +0530 (IST))
Subject: Re: Sh-utils 2.0j sources uploaded
References:  <Pine DOT LNX DOT 4 DOT 21 DOT 0007250651010 DOT 823-100000 AT midpec DOT com>
Reply-To: djgpp-workers AT delorie DOT com
Errors-To: nobody AT delorie DOT com
X-Mailing-List: djgpp-workers AT delorie DOT com
X-Unsubscribes-To: listserv AT delorie DOT com
Precedence: bulk

[CC: to Jim Meyering, since some issues go beyond DJGPP-specific aspects.]

> Date: Tue, 25 Jul 2000 06:59:03 +0530 (IST)
> From: Prashant TR <tr AT eth DOT net>
> 
> The DJGPP port of the Sh-utils pretest is available.
> 
> ftp://ftp.simtel.net/pub/simtelnet/gnu/djgpp/v2gnu/alphas/shl20js.zip

I finally found time to download and test this beta version.  The
package built for me flawlessly, both on plain DOS and on Windows 98,
both in the source directory and outside it.  Good work!

I have only a few minor comments:

  1) When the configure script runs, it prints an error message from
     `ln' (when it tests whether `lstat' dereferences a symlink with a
     trailing slash).  This is expected behavior, but it might alarm
     users.  I suggest either to redirect stderr to /dev/null or at
     least mention the message in djgpp/README.

  2) The version of `depcomp' supplied with sh-utils-2.0j does not
     work correctly for DOS/Windows-style pathnames with drive
     letters.  The following Sed script breaks if .deps/*.Tpo files
     include absolute file names with drive letters:

      sed 's/^[^:]*: / /' < "$tmpdepfile" >> "$depfile"

     This happens because on DOS/Windows, "gcc -MD" produces the
     following in, e.g., lib/.deps/addext.Tpo:

       e:/src/gnu/sh-util2.0j/lib/addext.o: \
       e:/src/gnu/sh-util2.0j/lib/addext.c ../config.h \
       e:/src/gnu/sh-util2.0j/lib/backupfile.h c:/bin/gcc/include/limits.h \
       c:/bin/gcc/include/sys/types.h c:/bin/gcc/include/sys/djtypes.h \
       c:/bin/gcc/include/string.h c:/bin/gcc/include/sys/movedata.h \
       c:/bin/gcc/include/unistd.h

     The first line above is supposed to be removed by the Sed script,
     but that fails because the file name includes a colon.

     Note that this problem is only visible when building Sh-utils
     from another drive, because only then the drive letter appears in
     the file name.

     The patch to depcomp which fixes this is below.  Jim, where
     should I post it?  Also, is \{0,1\} supported in all versions of
     Sed?

  3) The test suite fails in several ways, unless the package is
     configured, built, and tested in the source directory.  This
     happens because djgpp/edtest.bat is not written to work except in
     that case.  Either edtest.bat should be augmented, or a note to
     this effect should be added to djgpp/README.

     In any case, I think that the DJGPP source distribution should
     come with the files under `tests/' already edited and converted
     to the DOS-style CRLF EOL format (where appropriate).  (If you
     decide to accept this advice, you need to take into consideration
     that someone might run config.bat again, so the conversion Sed
     scripts should be prepared for multiple editing.)

  4) lib/Makefile.in has a rule to produce C files from Yacc grammar
     sources.  This rule will not work on plain MS-DOS, because file
     names like y.tab.c are not allowed, and because the DJGPP port of
     Yacc produces y_tab.c instead of y.tab.c when it runs on plain
     DOS machine.  This problem will pop up if someone edits
     getdate.y, for example.

     The offending rule comes from Automake, so to correct this once
     and for all, Automake should be changed.  What is the correct
     address to report this?  Is it automake-bugs AT gnu DOT org?

     For now, I suggest to add some magic to djgpp/config.sed so that
     it edits this:

.y.c:
	$(YACC) $(AM_YFLAGS) $(YFLAGS) $< && mv y.tab.c $*.c
	if test -f y.tab.h; then \
	if cmp -s y.tab.h $*.h; then rm -f y.tab.h; else mv y.tab.h $*.h; fi; \
	else :; fi

     into this:

.y.c:
	$(YACC) $(AM_YFLAGS) $(YFLAGS) $<
	-test -f y.tab.c && mv y.tab.c y_tab.c
	-test -f y.tab.h && mv y.tab.h y_tab.h
	mv y_tab.c $*.c
	if test -f y_tab.h; then \
	if cmp -s y_tab.h $*.h; then rm -f y_tab.h; else mv y_tab.h $*.h; fi; \
	else :; fi

  5) A note: since `true' and `false' are now programs, you need to
     make sure shl20b.dsm includes commands to uninstall the old shell
     scripts by those names before installing the binaries.

  6) Finally, it looks like the DJGPP source distribution will require
     to run the configure script before building.  If that is true,
     I'd like to suggest to make any reasonable effort so that the
     sources could come already configured.  AFAIK, every other ported
     package does that; we should try not to require users to have all
     those additional utilities installed just to rebuild a package.

     I'm guessing that this depcomp thingy is one of the major
     obstacles to ship the sources configured.  If so, let's see what
     exactly in depcomp prevents you from doing so, and let's try to
     solve those problems, either by submitting patches to depcomp
     maintainers, or by augmenting the DJGPP-specific scripts.

Last but not least, thanks to Jim for a wonderful package, and to
Prashant for the good work porting it.


Here's the patch for depcomp:


2000-08-11  Eli Zaretskii  <eliz AT is DOT elta DOT co DOT il>

	* depcomp (gcc): Support DOS-style absolute file names with drive
	letters.

--- depcomp.~0	Mon Oct  4 07:56:46 1999
+++ depcomp	Thu Aug 10 20:49:02 2000
@@ -65,7 +65,7 @@
   fi
   rm -f "$depfile" 
   echo "$object : \\" > "$depfile"
-  sed 's/^[^:]*: / /' < "$tmpdepfile" >> "$depfile"
+  sed 's/^\([A-z]:\/\)\{0,1\}[^:]*: / /' < "$tmpdepfile" >> "$depfile"
 ## This next piece of magic avoids the `deleted header file' problem.
 ## The problem is that when a header file which appears in a .P file
 ## is deleted, the dependency causes make to die (because there is