[petsc-dev] broken builds

[Jed Brown]

On Tue, Aug 30, 2011 at 14:37, Barry Smith <bsmith at mcs.anl.gov> wrote:

>   Back to your proposal, which you have to admit is pretty ugly
>
>    --download-package      gets the tarball if needed and builds
>   --download-package=something.tar.gz   use this tarball and build
>   --download-package=rebuild   reconfigure and rebuild package (even if it
> thinks it doesn't need to)
>
>   I don't like the fact that what comes after the = is disparate;  a
> keyword or a filename
>
>
>    Don't we also need
>   --download-package=update   get the latest tarball and build from that?
>

Well, --update-package.

But I'm not sure we should try to bolt it on to the current interface. Why
do we have to declare every dependent package up-front? Why not be able to
start with a minimal configuration and add packages? There could be a
shorthand for doing a single configuration line, but I don't think that
adding Umfpack should need to involve a full reconfigure and recompile. We
should be able to install Umfpack, recompile matregis.c (maybe we can avoid
this), compile umfpack.c, relink, and be done. If we have some influence on
dependency tracking, I think this could even be done in a maintainable way.
Not a project for this week, however.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110830/3dd03b04/attachment.html>