Re: Fix an error during "./configure; make dist"

[Patrice Dumas]

On Fri, Aug 09, 2024 at 11:08:02PM +0100, Gavin Smith wrote:
> With the attached patch, it cd's to srcdir before running pod2texi.
> This way pod2texi with --subdir works correctly.

Looks like it should work as intended, indeed.

> In order for the change of working directory to work correctly, I had
> to make a few changes.  texi2any_internals.texi is listed as a Makefile
> target without the "$(srcdir)/" prefix.  This still works correctly
> due to the VPATH mechanism.  This allows correct use of $@ in the Makefile
> rule in the new working directory.
> 
> The files in $(texi2any_internals_dependencies) are now listed without the
> "$(top_srcdir)/" prefix; rather, they are given a "../../" prefix.  Again,
> the VPATH mechanism should allow these files to be found in the srcdir.
> The "../../" prefix is still correct for these files after "cd $(srcdir)",
> allowing the list of files to be passed to pod2texi correctly.

I recall that on some platform, a BSD I think, Make only used the VPATH
for the prerequiste (if I recall well), and that's why we systematically
added $srcdir for targets even though it would not have been needed
with VPATH.  But I think that your approach can still work, by not using $@
and having a different path in the target and rule:
$(srcdir)/texi2any_internals.texi: ....
        ......... > texi2any_internals.texi ....

> Finally, I had to use absolute directory names (abs_top_srcdir and
> abs_top_builddir) in MAKEINFO_ENVIRONMENT so that they would still
> be still correct after the change of directory; otherwise, ModulePath.pm
> wouldn't be found.

Indeed.

> It's possible that the behaviour of pod2texi is wrong here

Actually I don't think so.

> but I don't
> have any suggestions as to how it should be changed.

> > Here is a recipe to get differences between in source and out of source
> > distributions.  There are some changes for generated files, which seem
> > innocuous, but may be an issue for reproductibility, as well as the
> > difference for texi2any_internals.texi.
> 
> Thanks, I'll try it.

If we really want to support make dist in out of source builds, I think
that we sould make sure that there are no differences at all between in
source and out of source distributions.  This should not be difficult
for some file we generate, but there are also some gperf generated files
in gnulib that have some srcdir prepended, which may be more challenging
to get right.  Maybe we should report that to gnulib, or Bruno, you
could have a look?

-- 
Pat