From: Arthur Norman <acn1@ca...> - 2011-08-25 08:11:16
These are EXPERIMENTAL for at least a while, and they assume that you have
build BOTH a CSL and a PSL version of Reduce, so that bin/redcsl and
bin/redpsl will work. But the idea is that
will use both Lisp variants to run the test file for the given package,
leaving log files in csl-times and psl-times directories. And that
will run all tests that packages/package.map is aware of.
When looking at the output be aware that the timing reported at the bottom
of each log is a raw one obtained by going someting like
time bin/redcsl ...
so it includes all overhead in the shell and in starup when trying the
code. For many examples (eg specifically "alg.tst") the overhead is very
very much greater than the time spent actually in the test.
Also the scripts compare output against "reference logs" - but some
divergence is EXPECTED. At the very least anything that involves floating
point can be expected to differ and this is not viewed as erroneous
because the two Lisps use different floating point libraries. And until
and unless at some stage in the future we produce an official "stable"
release rather than just snapshots the continuing development of the code
can lead reference logs and actual output to disagree.
However I hope these will generate interest and will help us move towards
converged results. Anybody who can enchance my scripts to remove aspects
of the logs that are sensitive to PSL vs CSL please join in and send
updates to me (or check them in yourself if you can).
Get latest updates about Open Source Projects, Conferences and News.