[petsc-dev] ML interface

[Jed Brown]

On Tue, 20 Apr 2010 12:45:07 -0500, Barry Smith <bsmith at mcs.anl.gov> wrote:
> 
>     If ML is being replaced why waste any time on it?

Agreed, but I don't know how long it's going to take to have something
usable.  The Sandia folks are all still using ML because the rewrite
isn't usable yet.

> Why not become a beta tester of the rewrite and make an interface for
> that? (You would possibly be able to influence bad decisions they are
> going to make).

Agreed, it would help if Sandia had public repositories.

>     PETSc already defined CCXX even when using the C compiler. You  
> could introduce SOURCECXX in the makefiles and possibly make mixing in  
> some C++ code relatively easily.

Okay.

> BUT why not just build PETSc with C++, what is the downside to that?

The C compiler is faster and produces better error messages because it
doesn't get confused by things like function overloading.  Since C and
C++ are different languages, it's reasonable for a user to want to build
their own work either way without recompiling the library (and still
using PETSc makefiles).  I'm aware of --with-c-support, but I don't see
any point to building PETSc proper (disregarding Sieve) with a C++
compiler.

Jed