[Carpet] Test suites

[Thomas Radke]

Ian Hinder wrote:
> Hi,
> 
> It looks to me like the Cactus test suite mechanism does not correctly
> handle running a test suite on a different number of processors than
> the test was originally run on with Carpet.  Because the CarpetIOASCII
> 1D and 2D output files are different depending on the processor
> decomposition, the diffing that happens to verify results is not
> enough.  We get test suite failures reported when they should not be.
> 
> Do people just rely on norms for testing that results are independent
> of the number of processors?

I guess so. Some thorns (of the few which provide testsuites at all) 
have separate ones for a specific number of processors (eg. 
CarpetIOHDF5/test/).

The CST perl script which compares output files was written for simple 
PUGH-based testsuites, it doesn't know any of the more specialised 
Carpet ASCII output file formats (such as chunked output). Either one 
adds an option to CarpetIOASCII to output in unchunked mode, or the perl 
script is made more intelligent. Both tasks are non-trivial to implement 
(for me).

-- 
Cheers, Thomas.