Hi list,
we are trying to push the latest version of numexpr into the Debian main archive and we are experimenting a strange test failure on sparc platforms [1].
Two tests fail: test_changing_nthreads_01_dec and test_rational_expr.
Both failures are related to floating point precision.
In the past we had threading related problems on sparc platforms so we decided to set the default number of threads to 1 on that platform (the patch has been applied to the numexpr tip too).
I guess we should also disable all thread specific tests for consistency.
Also it seems that test failure only happens when one uses the debug interpreter but I can’t understand why this should make some difference.
Do you have an idea about which could be the cause of the problem?
Any hint on how to proceed to fix the issue?
best regards
[1]
https://buildd.debian.org/status/fetch.php?pkg=numexpr&arch=sparc&ver=2.2.2-1&stamp=1381095467
--
Antonio Valentino