libxc regression test fails when using RPATH

[migueldiascosta]

this is because the regression test is done before the install step, so the library is not there yet

RPATH in xc-regression does include $ORIGIN:$ORIGIN/../lib:$ORIGIN/../lib64 (I suppose CMAKE is adding these?), but the library in this particular case is at $ORIGIN/../

more generically, easybuild could either not use RPATH with test binaries, or run a sanity check on the any test binaries and make sure all required libraries can be found with the relative $ORIGIN paths?

[boegel]

The $ORIGIN entries are injected by EasyBuild itself, via the RPATH wrapper scripts. These also actively filter out any paths in the build directory, which is the cause of the problem here: the test binaries get built with -L<builddir> for example, and then our RPATH wrapper kicks out that path (to ensure it's not relied on for anything in the final installation). We ran into a very similar problem with FlexiBLAS (see #12476).

We could disable that filtering mechanism during the test step, but then we risk that build directory paths leak into the binaries/libraries that get installed. That's not a disaster though (just unclean), and hopefully the sanity check would catch it, so maybe that's a good way of fixing it?

@pescobar Did you ever run into problems like this, libraries not being found in the test step when building with RPATH enabled?

[pescobar]

@boegel I tried to reproduce the RPATH issues I had some time ago but this time it built ok. I cannot reproduce the issue. Not sure if this is related to changes in the latest easybuild release or changes in the application I am building