On Thu, 3 Mar 2011, John Peterson wrote:

> I confess that even with dbg builds of libmesh, I use the optimized
> PETSc libraries... therefore the PETSC_ARCH and PETSC_DIR that gets
> set at configure time is truly constant in my builds. One reason for
> this is that even optimized PETSc builds have some debugging
> information available in them...  Another is that dbg petsc builds can
> run significantly slower, and you aren't really debugging petsc after
> all (I hope) but the inputs to it from libmesh.

This is how I operate, and mostly for the same reasons.

But the main reason I never even considered switching on the fly
between different underlying PETSc builds is that I didn't know for
sure it was *possible*.  With libMesh our debug mode isn't
ABI-compatible with our opt mode (due to the glibcxx debug STL stuff),
and I bet there are lots configure-time options that have similar
effects due to structure packing changes, linking dependency changes,
etc.  I didn't want to bother finding out whether PETSc had any
similar "gotchas" between different configurations.

On the other hand, if PETSc really does support that kind of library
swapping, there's no reason we should make it harder in libMesh.  We
definitely don't want to not store PETSC_ARCH/PETSC_DIR *at all*,
since such a change would break anyone who has those env variables set
at configure time but not at build / run time... but maybe we could do
some ?= tricks in the Makefiles to allow overriding the defaults?
---
Roy

------------------------------------------------------------------------------
Free Software Download: Index, Search & Analyze Logs and other IT data in 
Real-Time with Splunk. Collect, index and harness all the fast moving IT data 
generated by your applications, servers and devices whether physical, virtual
or in the cloud. Deliver compliance at lower cost and gain new business 
insights. http://p.sf.net/sfu/splunk-dev2dev 
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users

Reply via email to