FSL missing some components

[verdurin]

The FSL easyconfig fails to create components such as imcp.

This is the situation in the note described near the top of:

https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FslInstallation

This is with 5.0.11.

[ocaisa]

Looking at the installation advice at https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FslInstallation/SourceCode

After building the packages from version 5.0.10 onwards it is necessary to download and build a python environment with the FSL folder in it's final location (it cannot be moved after this). Do this using the provided script:
$FSLDIR/etc/fslconf/post_install.sh -f $FSLDIR

I can't see this in the easyblock, so I'm guessing this hasn't been implemented...

[verdurin]

Yes, and I'm surprised no-one has noticed this before.
Or perhaps I was unlucky that the first test I ran required some of the missing components.
Is this best raised as an easyblock issue then?

[ocaisa]

If you run that script in the installation directory, do things turn out ok? If so then it definitely sounds like an easyblock issue and needs to be fixed there.

[ocaisa]

You could always add it as a post-install step but since the easyblock exists, I think it's most appropriate there. To be honest I think it needs further investigation anyway, it says that it will download a python environment which is not something I like the sound of...

[verdurin]

Yeah, it's pretty messy. Installs everything using Miniconda.
Owing to local deadlines I installed it manually for now, entirely separately from the EB version.
Happy to help with fixing it properly.
The developers are local (well, twenty minutes' walk away) so perhaps I'll be able to work with them directly.

[mboisson]

Can you tell them to ditch conda and use python instead, or at least provide the option to use python instead of conda ? Conda does not work well on clusters.

[mboisson]

One of our staff actually tried to add this script to be run at the end of the installation, and it turns into a mess of Conda trying to write stuff in the installer's account.

[leorojas82]

I also noticed eb does not run the post-installation step from FSL documentation, so fslpython does not get built at the end. As already mentioned, fslpython is a conda env, so it contains pre-compiled binaries. I ran the fslpython build manually, and then attempted to run the test suit provided by Oxford FMRIB (the so called FEEDS), but it failed, since the pre-compiled binaries required HDF5 1.10.1, where as eb installs HDF5 1.10.2 as a dependency of FSL. So at the end I had to build HDF5 1.10.1 and build FSL against it, run the post-installation by hand, All worked fine at the end. I don't know if the easyblock could do all that for us, also running FEEDS at the end will be great as a testing mechanism.

[casparvl]

Any update on this? Anyone has a working solution for now?

I bumped into the same issue. I got FEEDS to run in the end by pulling in the missing I/O scripts from an old source (5.0.10, from before they were moved to the Python-part of the installation). However, I'm not sure what else I might be missing...

@verdurin It might be a really great idea if you talk to them and make them aware of our perspective of things (i.e. as people managing larger systems with many software installations). I think it could really be in their benefit if they understand our challenges with a) installations that locally pull in a Miniconda and b) custom installations that have both a configure-make and python part. A better streamlined installation procedure means more likely that we install this kind of software, meaning more users / impact.

[verdurin]

@casparvl I have re-started the discussion with them, and now have a (long...) list of packages required for fsl_python. So far I've lacked the time to do anything with this. The willingness on their part to help is there.

[casparvl]

@verdurin Ok, great to hear they are willing at least.

Do you have any way you got this working for now? If I just added

$FSLDIR/etc/fslconf/post_install.sh -f $FSLDIR

As a postinstallcmd (as suggested according to https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/FslInstallation/SourceCode). But:

== 2019-10-22 20:44:25,197 run.py:589 INFO parse_log_for_error (some may be harmless) regExp (?<![(,-]|\w)(?:error|segmentation fault|failed)(
?![(,-]|\.?\w) found:
Failed to create FSL Python environment - see /scratch-shared/casparl/eb-Gm_lUn/fslpythonZH0D/fslpython_miniconda_installer.log for details
Post install setup failed

Unfortunately, that tmpdir was already removed before I could check those logs =) But I'm guessing it is the problem @leorojas82 mentioned earlier, of the binaries shipped as part of fslpython are linked to HDF5-1.10.1.

Is there any acceptable workaround for now? Did anyone create an EasyConfig that works (even if it should really be fixed as part of the block in the long run)? Or did you all just do the postinstallstep manually?

[akesandgren]

--disable-cleanup-tmpdir --disable-cleanup-builddir ??