Re: PATCH to top level: don't use dirname in configure

Re: PATCH to top level: don't use dirname in configure

[Jeffrey A Law]

  In message < 200007220919.FAA12005@indy.delorie.com >you write:
  > > Date: Fri, 21 Jul 00 17:29:14 CDT
  > > From: msokolov@ivan.Harhan.ORG (Michael Sokolov)
  > > 
  > > ! topsrcdir=`cd \`echo ${progname} | sed 's:/configure$::'\`; pwd`
  > 
  > This assumes that (a) $(progname) always expands to "configure", and
  > (b) that the last delimiter is '/'.  I don't know whether the first
  > assumption is true on each supported system, but the second might be
  > wrong on DOS/Windows where a backslash might be used instead.
I believe we can use something like this to deal with the first problem:

## this sed command emulates the dirname command
dstdir=`echo $dst | sed -e 's,[^/]*$,,;s,/$,,;s,^$,.,'`

The second problem exists in many places in the various configure & install
scripts and probably should be attacked and fixed en-masse.

Michael -- can you verify the sed command I provided above does the right
thing for you?

Also I'd like to have a comment which references dirname (like the one above)
to make these things easier to find (either when updating them to handle
DOS filesystems in the future, or when we need to find a suitable sed command
to emulate dirname elsewhere).

Thanks,
jeff

Re: PATCH to top level: don't use dirname in configure

[Michael Sokolov]

Eli Zaretskii <eliz@delorie.com> wrote:

> This assumes that (a) $(progname) always expands to "configure", and
> (b) that the last delimiter is '/'.  I don't know whether the first
> assumption is true on each supported system, but the second might be
> wrong on DOS/Windows where a backslash might be used instead.

Both of these assumptions are already being made extensively by the rest of the
very same Cygnus configure script. The use of dirname there was actually out of
line with respect to the rest of the script, and my patch makes it the same as
the rest of the script.

> In contrast, `dirname' (when it exists) will always DTRT, because it
                          ^^^^^^^^^^^^^^
> is tailored to the host system.

That's where the problem lies. As Alexandre Oliva <aoliva@redhat.com> wrote,

: Unfortunately, `dirname' is too far from being portable.  We must not
: depend on its existence.

GCC, Binutils, and GDB must be usable as add-ons to existing UNIX systems. At
least one of these systems, 4.3BSD, doesn't have dirname. Therefore, the
current top-level configure behaviour is unacceptable for GCC, Binutils, and
GDB.

My patch fixes an existing problem that must be fixed, and it cannot introduce
new problems that didn't exist before (as it does the same thing the rest of
the script does), therefore, it should be checked in.

--
Michael Sokolov		Harhan Engineering Laboratory
Public Service Agent	International Free Computing Task Force
			International Engineering and Science Task Force
			615 N GOOD LATIMER EXPY STE #4
			DALLAS TX 75204-5852 USA

Phone: +1-214-824-7693 (Harhan Eng Lab office)
E-mail: msokolov@ivan.Harhan.ORG (ARPA TCP/SMTP) (UUCP coming soon)
From msokolov@ivan.Harhan.ORG Sat Jul 22 20:07:00 2000
From: msokolov@ivan.Harhan.ORG (Michael Sokolov)
To: binutils@sources.redhat.com, gcc-patches@gcc.gnu.org, gdb-patches@sources.redhat.com
Subject: Re: PATCH to top level: set MAKE=make on 4.3BSD
Date: Sat, 22 Jul 2000 20:07:00 -0000
Message-id: <0007230304.AA20301@ivan.Harhan.ORG>
X-SW-Source: 2000-07/msg00235.html
Content-length: 660

> 2000-07-21  Michael Sokolov  <msokolov@ivan.Harhan.ORG>
>
>         * configure.in (per-host: vax-*-bsd*): New special case.
>         * config/mh-vaxbsd: New file.

I withdraw this patch. The Makefile framework in the entire Cygnus tree is
broken beyond my ability to fix it, I had to give up and install GNU make.

--
Michael Sokolov		Harhan Engineering Laboratory
Public Service Agent	International Free Computing Task Force
			International Engineering and Science Task Force
			615 N GOOD LATIMER EXPY STE #4
			DALLAS TX 75204-5852 USA

Phone: +1-214-824-7693 (Harhan Eng Lab office)
E-mail: msokolov@ivan.Harhan.ORG (ARPA TCP/SMTP) (UUCP coming soon)
From eliz@delorie.com Sun Jul 23 04:01:00 2000
From: Eli Zaretskii <eliz@delorie.com>
To: msokolov@ivan.Harhan.ORG
Cc: binutils@sources.redhat.com, gcc-patches@gcc.gnu.org, gdb-patches@sources.redhat.com
Subject: Re: PATCH to top level: don't use dirname in configure
Date: Sun, 23 Jul 2000 04:01:00 -0000
Message-id: <200007231101.HAA13289@indy.delorie.com>
References: <0007221542.AA29888@ivan.Harhan.ORG>
X-SW-Source: 2000-07/msg00236.html
Content-length: 1152

> Date: Sat, 22 Jul 00 10:42:11 CDT
> From: msokolov@ivan.Harhan.ORG (Michael Sokolov)
> 
> Eli Zaretskii <eliz@delorie.com> wrote:
> 
> > This assumes that (a) $(progname) always expands to "configure", and
> > (b) that the last delimiter is '/'.  I don't know whether the first
> > assumption is true on each supported system, but the second might be
> > wrong on DOS/Windows where a backslash might be used instead.
> 
> Both of these assumptions are already being made extensively by the
> rest of the very same Cygnus configure script.

Perhaps all of those places should be fixed not to use a literal
"configure".

> My patch fixes an existing problem that must be fixed, and it cannot
> introduce new problems that didn't exist before (as it does the same
> thing the rest of the script does), therefore, it should be checked
> in.

I didn't say the patch shouldn't be checked in (it's something for the
maintainers to decide), I just pointed out a possible regression that
could happen, in case it would be missed.  This is what these forums
are for.  If there's no regression, or if the regression is deemed not
important, it's fine with me.