[petsc-dev] Registration implicitly collective on COMM_WORLD

[Matthew Knepley]

On Tue, Feb 5, 2013 at 8:42 AM, Jed Brown <jedbrown at mcs.anl.gov> wrote:

>
> On Tue, Feb 5, 2013 at 7:32 AM, Matthew Knepley <knepley at gmail.com> wrote:
>
>>    4. XXIntializePackage() is called automatically for all default PETSc
>> classes when using --with-single-library=1
>>
>> I am guessing that default call will use PETSC_COMM_WORLD.
>>
>
> One way to do this is to make a "constructor" function for each shared
> library that registers its *InitializePackage function to be called from
> PetscInitialize. (Constructors are called before main so they can't be
> collective.) Then we wouldn't have a source-level reverse dependency and it
> would get loaded even for single-library=0. I'm not aware of any platforms
> that don't support some constructor mechanism, either via an attribute or
> C++ (which should not require linking libstdc++).
>

It sounds like the most likely thing to break on porting, however its
exactly what we have for dynamic libraries.

   Matt

-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20130205/24df5dc8/attachment.html>