On 19/11/15 05:22, Robert Schmidt wrote:

The build directory should still be there if it failed.

I don't think there is an easy way to get the same environment, but it is a good feature.


eb -x will should the build environment as defined by the EB toolchain support, in a way that you can copy-paste it.

Granted, that doesn't include other environment variables that will get set later, but those will be mentioned in the eb -x output too.

See http://easybuild.readthedocs.org/en/latest/Extended_dry_run.html .


regards,

Kenneth


On Wed, Nov 18, 2015, 9:17 PM Ben Roberts <[email protected] <mailto:[email protected]>> wrote:

    Hi all,

    I’ve got a build (of NWChem) that for some reason fails at the
    linking step of the executable.

    I would like to be able to debug this by making changes in the
    build directory and then re-running “make link” (which is done as
    the final part of the build step). However, I’m somewhat thwarted
    by the fact that the build process sets up a lot of environment
    variables and the like that don’t seem to be preserved outside the
    EB process itself.

    The process of building takes between 30 and 60 minutes before it
    gets to linking, so re-running from scratch isn’t a feasible option.

    Is there a way I can somehow tell EasyBuild to generate a “state
    file” or similar that contains all the environment settings, so
    that I can just load this file and it will automatically set up
    the environment ready to run the “make link”? I imagine such a
    thing would be useful when building many applications.

    Also, when building NWChem, I discovered that the value that the
    variable LIBMPI is to be set to (in the nwchem EasyBlock) was
    hardcoded to a value that doesn’t match the output of “mpiifort
    -show”. I’ve changed that line in my version of the EasyBlock, but
    I wonder whether it would be useful to have it set by an automatic
    process as we can’t necessarily guarantee that all compiler
    environments need the exact same set of MPI libraries. Thoughts?

    Thanks,
    Ben


Reply via email to