Menu

regression testing

Developers
2002-08-22
2013-10-17
  • Charlie Zender

    Charlie Zender - 2002-08-22

    Hi,

    I've created an ftp site which holds the results of
    the ./configure build attempts on varios architectures.
    I'd prefer to put these on ftp rather than email since
    they can be large. The site is

    ftp://dust.ps.uci.edu/pub/zender/nco/rgr

    The files are self-explanatory, and are created by
    the mini-scripts in nco/configure.eg. Please take a
    look at the results and see if they help you diagnose
    any of the remaiing problems.

    Thanks,
    Charlie

     
    • Rorik Peterson

      Rorik Peterson - 2002-08-22

      I like the ftp idea.  These logs are a bounty of information.  I've modified configure a lot based on these.  In particular:

      - - - - - - - - - -
      sparc-sun-solaris2.6
      problem: couldn't use the gcc configure found
      solution: changed default CC list; explicitely setting CC=your_compiler still overrides it.  early versions of gcc fail miserably when using Sun's ld.  Configure catches some of these instances and warns you, but this one seems to slip through.  You can tell configure what ld to use by using ./configure LD=/path_to_gnu_ld/ld if you want to use gcc with GNU's ld.  Otherwise, update gcc; I see this was 2.95.

      - - - - - - - - - -
      powerpc-ibm-aix4.3.3.0
      problem:1506-155 (W) Option langlvl requires suboption(s)
      solution: removed extra CPPFLAGS, this may not work, we'll see; give it a shot again.

      problem: failed to compile ncap.c
        #include file "getopt.h" not found
      solution: added #ifdef HAVE_CONFIG_H to ncap.c
      - - - - - - - - - -
      alphav56-unknown-linux-gnu
      problem: attempted to compile c++ library with valarray
      solution: corrected conditional compilation of c++ library

      - - - - - - - - - -
      mips-sgi-iris6.5
      problem: couldn't link to libnetcdf
      solution: added 64-bit compiler flags for 2nd round of linking. They were not there before.  Give it a shot again.

      - - - - - - - - - -
      i686-pc-linux-gnu
      problem: none apparently.

      - - - - - - - - - -
      When you get a chance, run those test again. 

      rorik

       
      • Charlie Zender

        Charlie Zender - 2002-08-22

        OK, I will try to update those soon.

        I use bash normally, with some csh when I'm in the
        underdeveloped world. is there a difference between

        FOO=blah ./configure

        and

        ./configure FOO=blah   ?

        does it depend on the shell?
        Is one preferred over the other?

         
      • Charlie Zender

        Charlie Zender - 2002-08-23

        Hi Rorik,

        I have performed the new regression tests and uploaded the results to
        the ftp site. My diagnoses are contained in the configure.eg file.

        You have made some great changes, LINUXALPHA now passes the basic tests
        and SUNMP is very close!
        Skim the results because there are in some of them compiler warnings
        that you can probably eliminate (e.g., g++: unrecognized option `-KPIC'
        on SUNMP). Also, I am slowly evolving the ./configure so the features
        tested may have changed. I18N, OMP, and DODS can wait until the basic
        builds are working.

        AIX and SGI are down to ABI=64-related link problems and inability to find bash
        SUN has problems finding bash, too, for make distclean.
        Also, SUN finds and tries to link to the HDF library libmfhdf.
        Looking at the action in bld/Makefile, it's clear that -lmfhdf supercedes
        -lnetcdf when HDF5 is enabled (so that only -lmfhdf should be passed to the
        linker). This brings up the general point that the HDF5 stuff is for development
        only, and adds no features yet, so HDF5 should be disabled by default and
        only enabled if specifically requested. SUN does build when HDF5 is disabled.

        Charlie

         
    • Rorik Peterson

      Rorik Peterson - 2002-08-22

      There shouldn't be a difference in practice.  FOO=blah ./configure sets the environment for the shell that is executing configure, whereas the other way sends arguments to configure.  configure then parses the command line and exports them if they are valid environmental variable names.

      I don't think there is a preferred method, and they _should_ yield the same result.

      rorik

       

Log in to post a comment.