[petsc-dev] ugliness due to missing lapack routines

[Jed Brown]

On Fri, Feb 8, 2013 at 12:05 AM, Barry Smith <bsmith at mcs.anl.gov> wrote:

>    Regardless of the craziness of my plan, I think we can all agree that
> the more structure, consistency, maintainability, and ability to introspect
> on the code of the PETSc source code we have the better off we are. So any
> work (ok so most work) I do on upgrading the PETSc source code in pursuit
> of the whale means better code.


I agree, but I think we need to keep things in context. I don't think that
completely eradicating CPP is feasible or even particularly beneficial.
Even without macros, there are way too many things that happen in C to do
these far-reaching transformations. Even if you could do the
transformations safely, you still have to understand their performance and
to explain why one seemingly-innocuous change to the source file causes
radically different transformations to be performed.

My experience with Haskell is that you end up with very confusing
performance effects unless you really understand how the compiler is
implemented.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20130208/8f439c7b/attachment.html>