can't build afni with R

[andreifoldes]

Greetings,

I'm trying to run the following code:

singularity run docker://repronim/neurodocker:latest generate singularity \
    --pkg-manager yum \
    --base-image fedora:36 \
    --afni method=binaries version=latest install_r_pkgs=true > /cubric/collab/359_curiosity/tamas/misc/afni-binaries-r.Singularityfile

singularity build --remote /cubric/collab/359_curiosity/tamas/images/afni-binaries-r.sif /cubric/collab/359_curiosity/tamas/misc/afni-binaries-r.Singularityfile

but get the following error

[...]
42 files removed
+ rm -rf '/var/cache/yum/*'
++ find / -name 'libgsl.so.??'
find: File system loop detected; '/tmp/build-temp-1350571086/rootfs' is part of the same file system loop as '/'.
++ printf ''
+ gsl_path=/usr/lib64/libgsl.so.25
+ '[' -n /usr/lib64/libgsl.so.25 ']'
++ dirname /usr/lib64/libgsl.so.25
+ ln -sfv /usr/lib64/libgsl.so.25 /usr/lib64/libgsl.so.0
'/usr/lib64/libgsl.so.0' -> '/usr/lib64/libgsl.so.25'
+ ldconfig
+ mkdir -p /opt/afni-latest
Downloading AFNI ...
+ echo 'Downloading AFNI ...'
+ tar -xz -C /opt/afni-latest --strip-components 1
+ curl -fL https://afni.nimh.nih.gov/pub/dist/tgz/linux_openmp_64.tgz
  % Total    % Received % Xferd  Average Speed   Time    Time     Time  Current
                                 Dload  Upload   Total   Spent    Left  Speed
100 1092M  100 1092M    0     0  26.8M      0  0:00:40  0:00:40 --:--:-- 29.5M
+ yum install -y -q R-devel
+ yum clean all
42 files removed
+ rm -rf '/var/cache/yum/*'
+ rPkgsInstall -pkgs ALL
/.post.script: line 47: rPkgsInstall: command not found
FATAL:   While performing build: while running engine: exit status 127
FATAL:   While performing build: build image size <= 0

[Remi-Gau]

It seems that the R install of AFNI is not something we test regularly: https://github.com/ReproNim/neurodocker/actions/runs/6520678281/job/17708466328

We should probably fix that.

That being said AFNI is the most "broken" install that neurodocker has at the moment so any help to improve that is most welcome

[stebo85]

Dear @andreifoldes,

You can have a look at this neurodocker recipe, which works for installing the R-packages in AFNI:
https://github.com/NeuroDesk/neurocontainers/blob/master/recipes/afni/build.sh

In particular, you could try fedora 35 as a base image, which I have found to work best for AFNI:

   --base-image fedora:35 \
   --pkg-manager yum \
   --afni version=latest method=binaries install_r_pkgs='true' install_python3='true'

The error you get might also be related to singularity - it could be that building this with docker would actually work. It seems that it can't find the executable rPkgsInstall on the PATH, which could be a path error that somehow works in Docker?

If you just want an AFNI container with all required dependencies and R-packages installed and tested, you can also use the singularity container from the Neurodesk project (https://www.neurodesk.org/docs/getting-started/neurocontainers/singularity/):

curl -X GET https://objectstorage.us-ashburn-1.oraclecloud.com/n/sd63xuke79z3/b/neurodesk/o/afni_23.0.07_20230302.simg -O

I hope that helps
Kind regards
Steffen

[andreifoldes]

Greetings+

Manually editing the recipe from
from
rPkgsInstall -pkgs ALL
to
/opt/afni-latest/rPkgsInstall -pkgs ALL seems to have done the trick!

The error you get might also be related to singularity - it could be that building this with docker would actually work.
Unfortunately my cluster only support Singularity :(

If you just want an AFNI container with all required dependencies and R-packages installed and tested, you can also use the singularity container from the Neurodesk project (https://www.neurodesk.org/docs/getting-started/neurocontainers/singularity/)

Amazing! Thx I see that it contains python too!

[andreifoldes]

I stand corrected... while adding full path in the recipe does get the build issue fixed, once actually trying it out on real life data produces:

Loading required package: lme4
Loading required package: Matrix
Error: package or namespace load failed for 'lme4':
 package 'lme4' was installed before R 4.0.0: please re-install it
Failed with error:  'package 'lme4' could not be loaded'

This occurs with the neurodesk simg file specifically!

In terms of neurodocker - can yum install specific version of R? I have a feeling that if only I could get R 3.6 all 3dLMEr would work.

[stebo85]

Dear @andreifoldes,

neurodocker cannot install a specific version of R through yum - the best way I found to get a specific version of R is to choose a baseimage version that brings this R version. Another option could be to install R manually in a specific version.

So, 3dLMEr only works with R 3.6? Is that correct?

[andreifoldes]

choose a baseimage version that brings this R version

I see! Can you recommend which baseimage would be it for R 3.6, how could I check?

So, 3dLMEr only works with R 3.6? Is that correct?

I haven't checked this in all Linux distributions, but we have Centos7 on our cluster and there I had trouble getting lme4 working with r-base>=4 and had better success with 3.6. 3dLMEr needs lme4, lmerTest, phia and snow to work. In terms of singularity image I would be happy to have any working solution ^^

Another option could be to install R manually in a specific version.

Right, but then I would need to install it every time I call the function right? Or would you turn it into a sandbox?

[stebo85]

I think we can close this issue here. As discussed in this issue (NeuroDesk/neurocontainers#505), the problems @andreifoldes experienced come from running the singularity container without --cleanenv and R-packages from the host environment interfered with the container environment.