Trying to debug a build

[baxpr]

Links

• container collection: https://singularity-hub.org/collections/5207
• GitHub repository or recipe file: https://github.com/baxpr/gf-fmri

Version of Singularity

Tried with two builders: 2.5-100 and 3.2.1-100

Behavior when Building Locally

Builds fine locally with singularity 2.6.1 in an ubuntu 18.04 docker container

Error on Singularity Hub

The matlab binary https://github.com/baxpr/gf-fmri/blob/master/matlab/bin/spm12 doesn't run correctly during the build. In the recipe:

  /opt/gf-fmri/matlab/bin/run_spm12.sh /usr/local/MATLAB/MATLAB_Runtime/v97 function exit

And the error:

/opt/gf-fmri/matlab/bin/spm12: 1: version: not found
/opt/gf-fmri/matlab/bin/spm12: 2: oid: not found
/opt/gf-fmri/matlab/bin/spm12: 3: size: not found

What do you think is going on?

No clue. These error messages are obviously specific to the particular binary so I don't suppose they're much help. But are there some settings that could be different between the shub builders and my local environment? I turned the shub extra debug info on but didn't get any extra info that looked related.

[baxpr]

...one thing I do know is that running this binary uncompresses about 220 MB of files into a new folder. That is why it needs to be run in the first place - to generate these files in the container initially, because they can't be generated at runtime because the container is read-only.

[baxpr]

One more thing, here is a very similar situation where the build worked fine with ~800 MB of files unpacked:
https://github.com/baxpr/cersuit
https://singularity-hub.org/collections/4797

[vsoch]

@baxpr thank you for the detailed links! I haven't used spm in maybe 10 years so I'm feeling out of my base here - it does seem like there is some issue with the binary. Is the run_spm12.sh script made an executable, is it using the right interpreter, and then is spm12 what you think it is (e.g., it's not a dump from GitHub LFS saying "Sorry over quota!" or something like that. I would do some sanity checks - so remove piping to dev null where spm or matlab are concerned, and then when you download a text file or similar, cat the content (and for any downloads maybe check the digest) to be sure you are dealing with what you think you are.

@yarikoptic have you used SPM or Matlab more recently so you might know some ideas?

[baxpr]

Oh for heaven's sake. That's exactly what it is. I had marked the spm12 binary for LFS along with the actual large file spm12.ctf, but failed to download it so the container must be getting the github pointer instead of the file. Let me fix that and see if the problem doesn't go away.

[vsoch]

Fingers crossed! 🤞

[baxpr]

Yep, that was it. Thanks very much for the extra eyes!