As of August 2007, the make check runs the suite of tests much faster and with simpler output. It produces one line of output (pass or fail) for each test program. The tests also run much faster than before, primarily because the number of iterations is reduced to one (occasionally 2 or 3). Also, in some cases smaller example matrices are used.

Some LinBox functionality only emerges on large data, specifically the exact behaviour of hybrid algorithms. This is not currently tested -- and may not need to be, per se. Re iterations: there is not need to run tests 10 times and such (Edit here if you disagree). Another remark: compilation time can be cut in half by removing -O2 from the Makefile.

The individual test programs themselves produce no output stream, merely returning 0 for success and nonzero for failure. To get output use test-foobar - to get commentator ouput of the test on standard out, or test-foobar rep to get it written to file 'rep.

The test-blas-domain has been used as a benchmark. That has been moved to benchmark-blas-domain, which will thus give results that can be compared to previous test-blas-domain runs (provided -O2 is used).

Further modifications to the testing program are planned. In particular the make check action should serve as a sanity check after a configure/make sequence for LinBox installation. A different target, say make regression-check, will provide the more thorough regression testing for developers who are about to check-in a change to LinBox. Thus make check can become a run of benchmark suites for dense matrix operations (the benchmark-blas-domain), for blackbox operations on sparse matrices, for field/ring arithmetic performance, and perhaps for a few other things. The current suite of tests shall be refined and extended -- at the least every maintained function should be called once -- and used routinely as the regression-check target to avoid unintended side effects of LinBox modifications.