[petsc-dev] Registration implicitly collective on COMM_WORLD

[Jed Brown]

On Mon, Feb 4, 2013 at 10:53 PM, Matthew Knepley <knepley at gmail.com> wrote:

> I think this was a hack to begin with (and I did it), so calling the
> library loads up front does not bother me. I don't
> think anyone in the world is interested an a lightweight, partial PETSc.
>

I think this depends if (a) we ever develop an active plugin ecosystem and
(b) the memory per _process_ remains somewhat constraining. If a process
has only around 1GB, a couple unused plugins could eat up sizable chunks,
especially if implemented using templates. Since it's so much nicer to have
those "autoloads" available at run-time than to have to re-link, I think
there is a decent case for not killing late loading entirely.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20130204/8f2d35fe/attachment.html>