bug-4.4.0: missing man page for nces & more...

ocehugo
2014-01-21
2014-01-21
  • ocehugo
    ocehugo
    2014-01-21

    New year resolution: update my nco package.

    Status: Done...with a bug:

    The title says everything, missing a man page link to nces instead of ncea. Also, i really like the rename hehe.

    There is another thing too...running make test on latest OSX Mavericks with gnu-coreutils raise 2 errors:

    nces test 08: Check op with OpenMP............................................................... !!FAILED
    $cmd_rsl_is_num = 1 and $xpc_is_num = 0
    ERR: FAILURE in nces failure: Check op with OpenMP
    ERR::EXPLAIN: Result: [] != Expected: [n2 = 1]

    and

    ncra test 21: Check op with OpenMP............................................................... !!FAILED
    $cmd_rsl_is_num = 1 and $xpc_is_num = 0
    ERR: FAILURE in ncra failure: Check op with OpenMP
    ERR::EXPLAIN: Result: [] != Expected: [n2 = 1]

    Now the question: I think its safe to continue right!? Looks like these tests are not configured well. but its always good to ask:

    Its safe to use this NCO compiled with OpenMP? ./configure --help reports that openmp is default now. Moreover when selecting --disable-openmp, the tests still run these openmp tests (maybe a missing if clause here). Also some consistency need to be addressed in the NCO tests... make check should be removed or be an alias to make test, today make check is different from make test.

    Edit: I just checkout the flags and looks like a miss configured problem...even if --enable-openmp is on, the default behaviour of configure script is to not use openmp compiler directives. (clang and clang++ only uses -g -O2 -- with latest Xcode).

    Also I spotted a cuda folder inside NCO tree...new hot stuff in 2014?

    Cheers!

     
    Last edit: ocehugo 2014-01-21
  • Charlie Zender
    Charlie Zender
    2014-01-21

    Thanks for the reports. ncea manpage issue now fixed, will be in 4.4.1.

    I can get same behavior as you with OpenMP on Linux. It is safe to continue,
    I think, though clearly there is an OpenMP issue that we will look into. If you have a problem with a real use case, work around it by turning off threading with -t 0.

    Not sure about make check vs. make tst. Needs more thought.

    We need more resource to clean-up the OpenMP support. Fingers crossed.

    Cuda folder was an old proof-of-concept. Nothing new there. Great idea for volunteer...

    cz