Dear John-Paul,

On 08/05/2018 18:47, John-Paul Robinson wrote:
Hi,

We are working on a build of FSL with SLURM support patches.  During the build we found that the FSL easybuild unpacks the upstream source tarball in the installdir and doesn't use the builddir.  This is by design in the fsl easy block. (self.build_in_installdir = True)

I suspect this is an artifact of the pre-built binaries distributed for FSL that would be unpacked in place.  The current easyconfig for fsl, however, downloads the source tarball and appears to do a regular source build.

Despite being the main author of the FSL easyblock, I don't exactly recall when the FSL easyblock does a "build-in-installdir" procedure.

It's probably mostly due to laziness or not knowing what exactly needs to installed. The install_step method is currently empty, we get away with that exactly because we're building in the installation directory...


As far as I can tell, this is still the case for recent FSL easyconfigs too, since they just leverage the FSL easyblock.
Which easyconfig file are you referring to exactly?


There's nothing inherently wrong with building in the install dir, its just that in our environment the installdir is a NFS share. Unpacking the FSL tarball to that location is very slow (~5 min). If we specify a local disk for installdir then the upack is nearly instantaneous.

I'm wondering if this build behavior with the installdir should be maintained or if a more standard fsl build should be pursued.

It would definitely be better to have a cleanly separated build & installation directory, partially for the reason you mention (to speed up the installation when the final destination in a (slow) shared filesystem), but also for other reasons (clean installation, etc.).

And now for the million dollar question: are you planning to work on that yourself? If so, do let us know if you have any questions!

If so, it probably make sense to update the easyblock such that it does a cleanly separated build & install procedure for the latest version of FSL onwards.


regards,

Kenneth


Thanks for any insights,

John-Paul

HPC Architect
U. AL at Birmingham

Reply via email to