Re: eliminating conf.texi

[John W. Eaton]

On  3-Dec-2009, John W. Eaton wrote:

| On  2-Dec-2009, Rik wrote:
| 
| | John W. Eaton wrote:
| |
| | > As I see it, we have a some options.
| | >
| | >   1. Require that HAVE_COLAMD, HAVE_CHOLMOD, HAVE_UMFPACK, and
| | >      HAVE_QHULL are all defined in order to run "make dist" (so that
| | >      all the figures for the manual can be generated).
| | >
| | > I'm leaning toward option 1 since it means that the manual in a
| | > distribution file will always be complete.
| | 
| | I think we should begin by implementing option 1.
| 
| I've done some work on this and I think I have something that is very
| close to passing "make distcheck".
| 
| Could you please hold off on any additional changes until I've
| verified my latest changes and checked them in?

OK, I've checked in my changes.  Building Octave still works for me,
and the distclean target is much closer to working properly.  But we
may not be as close to passing "make distcheck" as I thought.  I'm
having a lot of difficulting writing rules that can properly determine
when the documentation doesn't need to be rebuilt.  Some of the
problems plus other notes:

  * .texi files are generated from .txi files using a program that we
    build, so that program should be listed as a dependency of the
    .texi files.  But doing that means that a fresh build from a
    tar.gz file would always generate the .texi files.  I thought I
    could solve that problem with a clever move-if-change rule, but
    that doesn't work with VPATH builds because once the rule is run
    to make the file in the local directory, Make doesn't look in the
    VPATH direoctry for that file again.

  * The new system distributes a lot of generated files that we
    weren't distributing before.  Things like all the vx- and mx-
    files in liboctave and a bunch of generated header files.  I don't
    think this causes much harm, but I'm not sure that we really
    should be distributing them when they can be easily
    reconstructed.

  * Instead of using Texinfo macros to do @verbatiminclude to insert
    example files, I extended the munge-texi program to recognize
    @EXAMPLEFILE and insert the given example file, similar to the way
    @DOCSTRING is handled.  This seems to work OK and avoids the need
    for the abs_top_srcdir or srcdir macros in conf.texi.

  * I left conf.texi for now, but it only contains the version
    number.  Then I found that automake has some way to manage version
    info for Texinfo files, so we could switch to using that.

  * if you don't have the qhull or sparse libraries, no warning is
    produced telling you that the documentation files won't be
    constructed properly.  They will be constructed, but the figures
    that require those libraries will all be sombreros instead of what
    they really should be.

I'm sure that I've missed some things.  Help in getting all these
details straightened out would be greatly appreciated.

Thanks,

jwe