[ET Trac] [Einstein Toolkit] #1898: Testsuites in QuasiLocalMeasures failing
Einstein Toolkit
trac-noreply at einsteintoolkit.org
Wed May 25 11:09:34 CDT 2016
#1898: Testsuites in QuasiLocalMeasures failing
-----------------------------------+----------------------------------------
Reporter: knarf | Owner: knarf
Type: defect | Status: new
Priority: major | Milestone: ET_2016_05
Component: EinsteinToolkit thorn | Version: development version
Keywords: |
-----------------------------------+----------------------------------------
At the time of writing, the issue has been fixed. However, I open (and
close) this ticket to have some place to save information about it, in
case it is of use later.
Recently, QuasiLocalMeasures showed its testsuites failing on 2 processes.
The problem, as it turns out, was that the output was generated not
independent of the number of processes.
This is the log from the commit that now fixes the immediate problem:
(https://bitbucket.org/einsteintoolkit/einsteinanalysis/commits/76e24130c79d/)
Change testsuites to use compact ASCII output and to not output
information from within ghostzones. This makes the output independent
from the number of processes, and allows the 2-process testsuites
to succeed, even with the output generated using 1 process.
The testsuites did succeed before this change, but only on one process.
They did fail on more than one, due to completely different output
format then. However, the testsuite mechanism didn't pick up this
failure until only lately, due to multiple factors. One was (and still
is) that Cactus has issues detecting errors when values are zero, due to
relative errors being non-trivial to define. Ticket #1889 is about that.
The other issue was that, while those lines were skipped, the different
length of the output files was not detected until a fix for that was
introduced recently, uncovering the problem here.
--
Ticket URL: <https://trac.einsteintoolkit.org/ticket/1898>
Einstein Toolkit <http://einsteintoolkit.org>
The Einstein Toolkit
More information about the Trac
mailing list