On Tue, Feb 28, 2012 at 9:56 PM, Roy Stogner <roystgnr@ices.utexas.edu> wrote:
We really need some benchmarks that get run and tested for excessive
error levels as part of the basic open source libMesh suite.  "Secret
apps that INL runs" plus "Secret apps that UT runs" plus etcetera is
much better than nothing, but in light of the recent system-specific
(one compiler-specific!) failures of doxygen and xdr on libmesh-users,
it would be nice to have a "make check" target that catches numerical
errors rather than just logic errors.

I'd propose we setup a test directory and start tossing stuff in there and have it run during make check, having the test be responsible for deciding if it passes or fails and return 0 for pass and 1 for not. I'm hacking on the examples right now anyway so if you (or anyone else...) wants to send me a couple of benchmark problems, I'll setup the infrastructure for it, with some detailed instructions on adding more tests.