PLplot

On 2006-05-17 20:27-0400 Jim Dishaw wrote:

>
> On 2006-05-16 23:41-0700 Alan W. Irwin wrote:
>
>> On 2006-05-16 23:15-0400 Jim Dishaw wrote:
>>>
>>> The line of interest is
>>>
>>> /usr/bin/install: cannot stat `/opt/build/plplot-cvs/pkgcfg//misc/media/software/source/scientific/plplot-cvs/pkgcfg/README':No such file or directory
>>>
>>> Notice the '//' in the error.
>>
>> Something really peculiar is going on.  The correlation between your results
>> and mine is low.
>>
>
> I think it is an artifact of using a VPATH build.  If I have time, I may
> try to trace the problem in autotools.
>
>> What is the complete set of options you are specifying to ./configure? I
>> suspect your --prefix option is screwed up.  That directory location must be
>> absolute (start with a "/" as in --prefix=/usr/local/plplot) and must be
>> separate from your source/build tree.
>
> /misc/media/software/source/scientific/plplot-cvs/configure \
>  --prefix=/opt/sci --libdir=/opt/sci/lib/i686-pc-linux-gnu --enable-f95 \
>  --with-x \
>  CC=/usr/local/gcc-cvs/bin/gcc \
>  F77=/usr/local/gcc-cvs/bin/gfortran \
>  FC=/usr/local/gcc-cvs/bin/gfortran \
>  CXX=/usr/local/gcc-cvs/bin/g++ \
>  FFLAGS=-I. \ FCFLAGS=-I. \ --enable-ltdl-convenience
>

We have often traced inexplicable errors to inconsistent versioning
of autotools.  So if you want to debug this further, use the same autotools
versions that we do.  If you still get the problem, then you may have found
a bug in the way we support libdir.  What happens if you try a build with
the correct autotools versions and without --libdir being set?

Alan
__________________________
Alan W. Irwin

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the FreeEOS equation-of-state implementation
for stellar interiors (freeeos.sf.net); PLplot scientific plotting software
package (plplot.org); the Yorick front-end to PLplot (yplot.sf.net); the
Loads of Linux Links project (loll.sf.net); and the Linux Brochure Project
(lbproject.sf.net).
__________________________

Linux-powered Science
__________________________

On 2006-05-17 18:23-0700 Alan W. Irwin wrote:

> On 2006-05-17 20:27-0400 Jim Dishaw wrote:
>
>> 
>> On 2006-05-16 23:41-0700 Alan W. Irwin wrote:
>> 
>>> What is the complete set of options you are specifying to ./configure?...
>> 
>> /misc/media/software/source/scientific/plplot-cvs/configure \
>>  --prefix=/opt/sci --libdir=/opt/sci/lib/i686-pc-linux-gnu --enable-f95 \
>>  --with-x \
>>  CC=/usr/local/gcc-cvs/bin/gcc \
>>  F77=/usr/local/gcc-cvs/bin/gfortran \
>>  FC=/usr/local/gcc-cvs/bin/gfortran \
>>  CXX=/usr/local/gcc-cvs/bin/g++ \
>>  FFLAGS=-I. \ FCFLAGS=-I. \ --enable-ltdl-convenience
>> 
>
> We have often traced inexplicable errors to inconsistent versioning
> of autotools.  So if you want to debug this further, use the same autotools
> versions that we do.  If you still get the problem, then you may have found
> a bug in the way we support libdir.  What happens if you try a build with
> the correct autotools versions and without --libdir being set?

As part of my tests for the new install location of the fortran modules I
tried the following:

./configure --prefix=/usr/local/plplot --disable-static --enable-f95
--libdir=/tmp/plplot >& configure.outa

That --libdir option (first time I ever tried that!) worked fine not only in
build and install, but also the build and execution of the installed
examples.  So this verifies that the new install location for the fortran
modules is working fine even for --libdir.

So you learn something new every day.  Geoffrey, you should have a look at
./configure --help (especially the section entitled "Fine tuning of the
installation directories").  Your disambiguation concerns may all be addressed
there.

Also, Jim, since --libdir works for me, then your problems may be due to (a)
autotools versions that are inconsistent with our requirements, (b) some
separate build tree bug (although Andrew often tests that), (c) some other
bug on our part, or (d) some other silly error such as stale files,
permission problems, etc.

Good luck finding out what it is.

Alan
__________________________
Alan W. Irwin

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the FreeEOS equation-of-state implementation
for stellar interiors (freeeos.sf.net); PLplot scientific plotting software
package (plplot.org); the Yorick front-end to PLplot (yplot.sf.net); the
Loads of Linux Links project (loll.sf.net); and the Linux Brochure Project
(lbproject.sf.net).
__________________________

Linux-powered Science
__________________________

"Alan W. Irwin" <irwin@...> writes:

> On 2006-05-17 20:27-0400 Jim Dishaw wrote:
>
>>
>> On 2006-05-16 23:41-0700 Alan W. Irwin wrote:
>>
>>> On 2006-05-16 23:15-0400 Jim Dishaw wrote:
>>>>
>>>> The line of interest is
>>>>
>>>> /usr/bin/install: cannot stat `/opt/build/plplot-cvs/pkgcfg//misc/media/software/source/scientific/plplot-cvs/pkgcfg/README':No such file or directory
>>>>
>>>> Notice the '//' in the error.
>>>
>>> Something really peculiar is going on.  The correlation between your results
>>> and mine is low.
>>>
>>
>> I think it is an artifact of using a VPATH build.  If I have time, I may
>> try to trace the problem in autotools.
>>
>>> What is the complete set of options you are specifying to ./configure? I
>>> suspect your --prefix option is screwed up.  That directory location must be
>>> absolute (start with a "/" as in --prefix=/usr/local/plplot) and must be
>>> separate from your source/build tree.
>>
>> /misc/media/software/source/scientific/plplot-cvs/configure \
>>  --prefix=/opt/sci --libdir=/opt/sci/lib/i686-pc-linux-gnu --enable-f95 \
>>  --with-x \
>>  CC=/usr/local/gcc-cvs/bin/gcc \
>>  F77=/usr/local/gcc-cvs/bin/gfortran \
>>  FC=/usr/local/gcc-cvs/bin/gfortran \
>>  CXX=/usr/local/gcc-cvs/bin/g++ \
>>  FFLAGS=-I. \ FCFLAGS=-I. \ --enable-ltdl-convenience
>>
>
> We have often traced inexplicable errors to inconsistent versioning
> of autotools.  So if you want to debug this further, use the same autotools
> versions that we do.  If you still get the problem, then you may have found
> a bug in the way we support libdir.  

I think that the offending line is in pkgcfg/Makefile.am, specifically
(annotated with ******)

install-data-hook:
if with_pkg_config
        $(mkinstalldirs) $(DESTDIR)$(docdir)
******  $(INSTALL) `pwd`/$(top_srcdir)/pkgcfg/README $(DESTDIR)$(docdir)/README.pkg-config
endif

Using the configure command, as specified above, yields

top_srcdir = /misc/media/software/source/scientific/plplot-cvs/pkgcfg

and the error

/misc/media/software/source/scientific/plplot-cvs/cf/install-sh -d /opt/sci/share/doc/plplot
/usr/bin/install -c `pwd`//misc/media/software/source/scientific/plplot-cvs/pkgcfg/README /opt/sci/share/doc/plplot/README.pkg-config
/usr/bin/install: cannot stat `/opt/build/plplot-cvs/pkgcfg//mnt/media/software/source/scientific/plplot-cvs/pkgcfg/README': No such file or directory
make[3]: *** [install-data-hook] Error 1
make[3]: Leaving directory `/opt/build/plplot-cvs/pkgcfg'
make[2]: *** [install-data-am] Error 2
make[2]: Leaving directory `/opt/build/plplot-cvs/pkgcfg'
make[1]: *** [install-am] Error 2
make[1]: Leaving directory `/opt/build/plplot-cvs/pkgcfg'
make: *** [install-recursive] Error 1

I think the `pwd` is the problem and that just $(top_srcdir) should work
for all configurations--at least it does for mine!

cheers
-jd

Rafael, there is an important autoconf/m4 question for you at the end.
On 2006-06-19 00:44-0400 Jim Dishaw wrote:

> "Alan W. Irwin" <irwin@...> writes:
>
>> On 2006-05-17 20:27-0400 Jim Dishaw wrote:
>>
>>>
>>> On 2006-05-16 23:41-0700 Alan W. Irwin wrote:
>>>
>>>> On 2006-05-16 23:15-0400 Jim Dishaw wrote:
>>>>>
>>>>> The line of interest is
>>>>>
>>>>> /usr/bin/install: cannot stat `/opt/build/plplot-cvs/pkgcfg//misc/media/software/source/scientific/plplot-cvs/pkgcfg/README':No such file or directory
>>>>>
>>>>> Notice the '//' in the error.
>>>>
>>>> Something really peculiar is going on.  The correlation between your results
>>>> and mine is low.
>>>>
>>>
>>> I think it is an artifact of using a VPATH build.  If I have time, I may
>>> try to trace the problem in autotools.
>>>
>>>> What is the complete set of options you are specifying to ./configure? I
>>>> suspect your --prefix option is screwed up.  That directory location must be
>>>> absolute (start with a "/" as in --prefix=/usr/local/plplot) and must be
>>>> separate from your source/build tree.
>>>
>>> /misc/media/software/source/scientific/plplot-cvs/configure \
>>>  --prefix=/opt/sci --libdir=/opt/sci/lib/i686-pc-linux-gnu --enable-f95 \
>>>  --with-x \
>>>  CC=/usr/local/gcc-cvs/bin/gcc \
>>>  F77=/usr/local/gcc-cvs/bin/gfortran \
>>>  FC=/usr/local/gcc-cvs/bin/gfortran \
>>>  CXX=/usr/local/gcc-cvs/bin/g++ \
>>>  FFLAGS=-I. \ FCFLAGS=-I. \ --enable-ltdl-convenience
>>>
>>
>> We have often traced inexplicable errors to inconsistent versioning
>> of autotools.  So if you want to debug this further, use the same autotools
>> versions that we do.  If you still get the problem, then you may have found
>> a bug in the way we support libdir.
>
> I think that the offending line is in pkgcfg/Makefile.am, specifically
> (annotated with ******)
>
> install-data-hook:
> if with_pkg_config
>        $(mkinstalldirs) $(DESTDIR)$(docdir)
> ******  $(INSTALL) `pwd`/$(top_srcdir)/pkgcfg/README $(DESTDIR)$(docdir)/README.pkg-config
> endif
>
> Using the configure command, as specified above, yields
>
> top_srcdir = /misc/media/software/source/scientific/plplot-cvs/pkgcfg
>
> and the error
>
> /misc/media/software/source/scientific/plplot-cvs/cf/install-sh -d /opt/sci/share/doc/plplot
> /usr/bin/install -c `pwd`//misc/media/software/source/scientific/plplot-cvs/pkgcfg/README /opt/sci/share/doc/plplot/README.pkg-config
> /usr/bin/install: cannot stat `/opt/build/plplot-cvs/pkgcfg//mnt/media/software/source/scientific/plplot-cvs/pkgcfg/README': No such file or directory
> make[3]: *** [install-data-hook] Error 1
> make[3]: Leaving directory `/opt/build/plplot-cvs/pkgcfg'
> make[2]: *** [install-data-am] Error 2
> make[2]: Leaving directory `/opt/build/plplot-cvs/pkgcfg'
> make[1]: *** [install-am] Error 2
> make[1]: Leaving directory `/opt/build/plplot-cvs/pkgcfg'
> make: *** [install-recursive] Error 1
>
> I think the `pwd` is the problem and that just $(top_srcdir) should work
> for all configurations--at least it does for mine!

We use the `pwd`/$(top_srcdir) and `pwd`/$(top_builddir) constructs several
places in our Makefile.am files to approximate $(abs_top_srcdir) and
$(abs_top_builddir) as documented by "info autoconf".  (The reason why we
use our own constructs rather than $(abs_top_srcdir) and $(abs_top_builddir)
is for unknown reasons the latter aren't implemented by autoconf even though
they are documented.) The only problem with our constructs is they assume
$(top_srcdir) and $(top_builddir) are relative paths rather than absolute
paths.

As a workaround try your configure command above using a relative path, and
I bet your problem will disappear.  For example, you currently invoke
configure using an absolute path of

/misc/media/software/source/scientific/plplot-cvs/configure


Suppose the top of your build tree was /home/dishaw/build/plplot.  Then that
invocation should be changed to the relative path

../../../../misc/media/software/source/scientific/plplot-cvs/configure

Obviously, this is a pretty clumsy workaround, and we should replace our
'pwd' constructs with an implementation of $(abs_top_pl_srcdir) and
$(abs_top_pl_builddir) which works for both relative and absolute paths.

Rafael, can you come up with such an implementation?

Alan
__________________________
Alan W. Irwin

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the FreeEOS equation-of-state implementation
for stellar interiors (freeeos.sf.net); PLplot scientific plotting software
package (plplot.org); the Yorick front-end to PLplot (yplot.sf.net); the
Loads of Linux Links project (loll.sf.net); and the Linux Brochure Project
(lbproject.sf.net).
__________________________

Linux-powered Science
__________________________

* Alan W. Irwin <irwin@...> [2006-06-19 09:20]:

> >I think that the offending line is in pkgcfg/Makefile.am, specifically
> >(annotated with ******)
> >
> >install-data-hook:
> >if with_pkg_config
> >       $(mkinstalldirs) $(DESTDIR)$(docdir)
> >******  $(INSTALL) `pwd`/$(top_srcdir)/pkgcfg/README 
> >$(DESTDIR)$(docdir)/README.pkg-config
> >endif
>
> [...]
>
> Obviously, this is a pretty clumsy workaround, and we should replace our
> 'pwd' constructs with an implementation of $(abs_top_pl_srcdir) and
> $(abs_top_pl_builddir) which works for both relative and absolute paths.
> 
> Rafael, can you come up with such an implementation?

I looked at the specific problem with the pwd in pkgcfg/Makefile.am and
noticed that it was introduced in revision 1.15 of the file:

$ cvs diff -r1.14 -r1.15 Makefile.am
Index: Makefile.am
===================================================================
RCS file: /cvsroot/plplot/plplot/pkgcfg/Makefile.am,v
retrieving revision 1.14
retrieving revision 1.15
diff -u -r1.14 -r1.15
--- Makefile.am 12 May 2006 19:04:37 -0000      1.14
+++ Makefile.am 13 May 2006 15:24:32 -0000      1.15
@@ -62,7 +62,7 @@
 install-data-hook:
 if with_pkg_config
        $(mkinstalldirs) $(DESTDIR)$(docdir)
-       $(INSTALL) README $(DESTDIR)$(docdir)/README.pkg-config
+       $(INSTALL) `pwd`/$(top_srcdir)/pkgcfg/README $(DESTDIR)$(docdir)/README.pkg-config
 endif

 EXTRA_DIST = README


This was done by Andrew Ross:


$ cvs log -r1.15 Makefile.am
RCS file: /cvsroot/plplot/plplot/pkgcfg/Makefile.am,v
Working file: Makefile.am
head: 1.15
branch:
locks: strict
access list:
keyword substitution: kv
total revisions: 15;    selected revisions: 1
description:
----------------------------
revision 1.15
date: 2006/05/13 15:24:32;  author: andrewross;  state: Exp;  lines: +1 -1
Fix installation of README.pkg-config to work with a separate build tree.
=============================================================================


I do not know why Andrew decided to use the pwd command but, actually,
"`pwd`/$(top_srcdir)/" is strictly the same as just "$(top_srcdir)/" when
top_srcdir is relative (which should be the case in normal situations).
When top_srcdir is absolute, the second form should work fine. Therefore,
I did not see reasons for using pwd here.  Andrew?

I just committed a "fixed" version of pkgcfg/Makefile.am.  Please, test it.
If this works, we can audit the other files where pwd is used and just try
similar fixes.

-- 
Rafael

On Mon, Jun 19, 2006 at 11:23:32PM +0200, Rafael Laboissiere wrote:
> * Alan W. Irwin <irwin@...> [2006-06-19 09:20]:
> 
> > >I think that the offending line is in pkgcfg/Makefile.am, specifically
> > >(annotated with ******)
> > >
> > >install-data-hook:
> > >if with_pkg_config
> > >       $(mkinstalldirs) $(DESTDIR)$(docdir)
> > >******  $(INSTALL) `pwd`/$(top_srcdir)/pkgcfg/README 
> > >$(DESTDIR)$(docdir)/README.pkg-config
> > >endif
> >
> > [...]
> >
> > Obviously, this is a pretty clumsy workaround, and we should replace our
> > 'pwd' constructs with an implementation of $(abs_top_pl_srcdir) and
> > $(abs_top_pl_builddir) which works for both relative and absolute paths.
> > 
> > Rafael, can you come up with such an implementation?
> 
> I looked at the specific problem with the pwd in pkgcfg/Makefile.am and
> noticed that it was introduced in revision 1.15 of the file:
> 
> $ cvs diff -r1.14 -r1.15 Makefile.am
> Index: Makefile.am
> ===================================================================
> RCS file: /cvsroot/plplot/plplot/pkgcfg/Makefile.am,v
> retrieving revision 1.14
> retrieving revision 1.15
> diff -u -r1.14 -r1.15
> --- Makefile.am 12 May 2006 19:04:37 -0000      1.14
> +++ Makefile.am 13 May 2006 15:24:32 -0000      1.15
> @@ -62,7 +62,7 @@
>  install-data-hook:
>  if with_pkg_config
>         $(mkinstalldirs) $(DESTDIR)$(docdir)
> -       $(INSTALL) README $(DESTDIR)$(docdir)/README.pkg-config
> +       $(INSTALL) `pwd`/$(top_srcdir)/pkgcfg/README $(DESTDIR)$(docdir)/README.pkg-config
>  endif
> 
>  EXTRA_DIST = README
> 
> 
> This was done by Andrew Ross:
> 
> 
> $ cvs log -r1.15 Makefile.am
> RCS file: /cvsroot/plplot/plplot/pkgcfg/Makefile.am,v
> Working file: Makefile.am
> head: 1.15
> branch:
> locks: strict
> access list:
> keyword substitution: kv
> total revisions: 15;    selected revisions: 1
> description:
> ----------------------------
> revision 1.15
> date: 2006/05/13 15:24:32;  author: andrewross;  state: Exp;  lines: +1 -1
> Fix installation of README.pkg-config to work with a separate build tree.
> =============================================================================
> 
> 
> I do not know why Andrew decided to use the pwd command but, actually,
> "`pwd`/$(top_srcdir)/" is strictly the same as just "$(top_srcdir)/" when
> top_srcdir is relative (which should be the case in normal situations).
> When top_srcdir is absolute, the second form should work fine. Therefore,
> I did not see reasons for using pwd here.  Andrew?
> 
> I just committed a "fixed" version of pkgcfg/Makefile.am.  Please, test it.
> If this works, we can audit the other files where pwd is used and just try
> similar fixes.

Rafael,

I don't think there was a particular reason for using `pwd`. I seem to
recall just copying the code from one of the other Makefiles. Doesn't
seem entirely sensible to me now... Sorry!

Andrew

* Andrew Ross <andrewross@...> [2006-06-19 23:30]:

> I don't think there was a particular reason for using `pwd`. I seem to
> recall just copying the code from one of the other Makefiles. Doesn't
> seem entirely sensible to me now... Sorry!

Thanks for the confirmation, Andrew.  I think we can go ahead and audit the
other uses of pwd in the Makefiles.  Alain, would you be willing to do this?
I am sort of overswamped in the moment...

-- 
Rafael

On 2006-06-20 02:07+0200 Rafael Laboissiere wrote:

> * Andrew Ross <andrewross@...> [2006-06-19 23:30]:
>
>> I don't think there was a particular reason for using `pwd`. I seem to
>> recall just copying the code from one of the other Makefiles. Doesn't
>> seem entirely sensible to me now... Sorry!
>
> Thanks for the confirmation, Andrew.  I think we can go ahead and audit the
> other uses of pwd in the Makefiles.  Alain, would you be willing to do this?
> I am sort of overswamped in the moment...

Sorry, I don't have much understanding of the circumstances (if any) where
the use of `pwd` would be necessary, and I am short of time as well. Maybe
`pwd` was never necessary, and after an initial mistake we just propagated
it elsewhere.  Perhaps the best thing to do is the empirical approach where
we simply remove it without too much thought, and then just test whether
that change works.

Andrew, would you be willing to test the removal of `pwd`?

find -name Makefile.am |xargs grep pwd

shows there are 6 instances total in two separate Makefile.am files.

Alan
__________________________
Alan W. Irwin

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the FreeEOS equation-of-state implementation
for stellar interiors (freeeos.sf.net); PLplot scientific plotting software
package (plplot.org); the Yorick front-end to PLplot (yplot.sf.net); the
Loads of Linux Links project (loll.sf.net); and the Linux Brochure Project
(lbproject.sf.net).
__________________________

Linux-powered Science
__________________________

* Alan W. Irwin <irwin@...> [2006-06-19 18:26]:

> Sorry, I don't have much understanding of the circumstances (if any) where
> the use of `pwd` would be necessary, and I am short of time as well. Maybe
> `pwd` was never necessary, and after an initial mistake we just propagated
> it elsewhere.  Perhaps the best thing to do is the empirical approach where
> we simply remove it without too much thought, and then just test whether
> that change works.
> 
> Andrew, would you be willing to test the removal of `pwd`?
> 
> find -name Makefile.am |xargs grep pwd
> 
> shows there are 6 instances total in two separate Makefile.am files.


Okay, although I am short in time, I did a quick audit on the two files
found by the find/grep command above.  In test/Makefile.am, the pwd was
introduced in version 1.14 by me.  Here is the cvs diff:


=============================================================================
$ cvs diff -r1.13 -r1.14 Makefile.am
Index: Makefile.am
===================================================================
RCS file: /cvsroot/plplot/plplot/test/Makefile.am,v
retrieving revision 1.13
retrieving revision 1.14
diff -u -r1.13 -r1.14
--- Makefile.am 11 Feb 2004 10:31:17 -0000      1.13
+++ Makefile.am 12 Feb 2004 08:47:28 -0000      1.14
@@ -50,8 +50,13 @@

 examples_SCRIPTS = $(scripts) plplot-test.sh

+# The variables below allow "make check" to work from both the build and
+# install trees.  Notice that the evironment variable EXAMPLES_DIR must be
+# absolute, otherwise test_tcl.sh and test_python.sh will fail.  The
+# construct below works because top_builddir is always relative.
+
 TESTS = plplot-test.sh
-TESTS_ENVIRONMENT = EXAMPLES_DIR=$(top_builddir)/examples $(SHELL)
+TESTS_ENVIRONMENT = EXAMPLES_DIR=`pwd`/$(top_builddir)/examples $(SHELL)

 EXTRA_DIST = test_c.sh             \
              test_cxx.sh           \
=============================================================================


Notice the comments above.  It seems that absolute paths are indeed
necessary and we are stuck here, unless the situation of test_tcl.sh and
test_python.sh have changed in the meanwhile.

There are other places in tests/Makefile.am where pwd appears, but they
where perhaps propagated by taking the fist `pwd`/$(top_builddir) as the
wrong pattern.

The other file concerned is examples/java/Makefile.am.  The introduction
of pwd was done by Andrew Ross in revision 1.11 with the following log
message:


=============================================================================
$ cvs log -r1.11 Makefile.am
RCS file: /cvsroot/plplot/plplot/examples/java/Makefile.am,v
Working file: Makefile.am
head: 1.25
branch:
locks: strict
access list:
keyword substitution: kv
total revisions: 25;    selected revisions: 1
description:
----------------------------
revision 1.11
date: 2004/06/30 16:04:53;  author: andrewross;  state: Exp;  lines: +40 -2
Changes to configuration, build and test process for java to enable
"make check"to work with java examples. Java examples can now be built
in situ. Tested with jikes/sablevm and with the Sun JDK.

The environment variable JAVACFLAGS is now honoured for building the java
examples. An immediate use of this is to add the -nowarn flag for jikes
which otherwise generates lots of (slightly) spurious warning messages.
Alan and I are working on it...
=============================================================================


I guess that the pwd commands introduced in this change set are
superfluous but I am not 100% sure.

-- 
Rafael

On Mon, Jun 19, 2006 at 06:26:34PM -0700, Alan Irwin wrote:
> On 2006-06-20 02:07+0200 Rafael Laboissiere wrote:
> 
> > * Andrew Ross <andrewross@...> [2006-06-19 23:30]:
> >
> >> I don't think there was a particular reason for using `pwd`. I seem to
> >> recall just copying the code from one of the other Makefiles. Doesn't
> >> seem entirely sensible to me now... Sorry!
> >
> > Thanks for the confirmation, Andrew.  I think we can go ahead and audit the
> > other uses of pwd in the Makefiles.  Alain, would you be willing to do this?
> > I am sort of overswamped in the moment...
> 
> Sorry, I don't have much understanding of the circumstances (if any) where
> the use of `pwd` would be necessary, and I am short of time as well. Maybe
> `pwd` was never necessary, and after an initial mistake we just propagated
> it elsewhere.  Perhaps the best thing to do is the empirical approach where
> we simply remove it without too much thought, and then just test whether
> that change works.
> 
> Andrew, would you be willing to test the removal of `pwd`?
> 
> find -name Makefile.am |xargs grep pwd
> 
> shows there are 6 instances total in two separate Makefile.am files.

I've audited these. I can see no good reason for the `pwd`. Changes
committed to CVS. This works fine for me, with a separate source and
build tree (always a good test) so I'm fairly sure it should be ok.
Please test on wierd and wonderful setups.

Andrew

On 2006-06-20 18:28+0100 Andrew Ross wrote:

>> Andrew, would you be willing to test the removal of `pwd`?
>>
>> find -name Makefile.am |xargs grep pwd
>>
>> shows there are 6 instances total in two separate Makefile.am files.
>
> I've audited these. I can see no good reason for the `pwd`. Changes
> committed to CVS. This works fine for me, with a separate source and
> build tree (always a good test) so I'm fairly sure it should be ok.
> Please test on wierd and wonderful setups.

My recent tests also tested this change.  Thanks, Andrew, for this correction.

Alan
__________________________
Alan W. Irwin

Astronomical research affiliation with Department of Physics and Astronomy,
University of Victoria (astrowww.phys.uvic.ca).

Programming affiliations with the FreeEOS equation-of-state implementation
for stellar interiors (freeeos.sf.net); PLplot scientific plotting software
package (plplot.org); the Yorick front-end to PLplot (yplot.sf.net); the
Loads of Linux Links project (loll.sf.net); and the Linux Brochure Project
(lbproject.sf.net).
__________________________

Linux-powered Science
__________________________