Make Cactus test suite mechanism independent of Carpet's domain decomposition

Issue #1230 new
Erik Schnetter created an issue

I recently used this line to look for differences:

diff -u <(grep '^[^#]' arrangements/EinsteinEvolve/GRHydro/test/balsara1_2d/grhydro\:\:bcons.y.asc | sort -u) <(grep '^[^#]' TEST/sim/GRHydro/balsara1_2d/grhydro\:\:bcons.y.asc | sort -u)

This removes all comments and empty lines, then sorts the output and removes duplicate lines (from different processes). The result is rather easy to interpret.

If one also deleted column 5 (?) containing the component number, then this would make the self-test independent of Carpet's process decomposition.

Keyword:

Comments (2)

  1. Frank Löffler
    • removed comment

    We tried something similar but more complex recently - directly interpreting the files 'knowning' which columns to use and which to ignore. This is not finished, but something we found is that at least for some testsuites data at the same point (ghost zones) isn't always identical, i.e., ghost zone data isn't always synced. One might argue that this shouldn't be the case in the first place, but apparently this is the case sometimes right now. The example you give wouldn't catch this. Nevertheless I found it quite neat. :)

    Of course - something also nice would be a thorn writing data already independent of the process layout, a bit like PUGH ASCII output. This would scale probably terribly, but it already does so right now anyway.

  2. Erik Schnetter reporter
    • removed comment

    If ghost zones are output, but are undefined (not synced), this is arguably an error in the test case.

  3. Log in to comment