Question: Why is dcm2niix saving to /tmp

[mirestrepo]

Summary

In our logs I can see that dcm2niix is saving to a random folder in /tmp. For instance I get messages like this one:

INFO: [Node] Running "convert" ("nipype.interfaces.dcm2nii.Dcm2niix"), a CommandLine Interface with command:
dcm2niix -b y -z y -x n -t n -m n -f func -o . -s n -v n /tmp/dcm2niix0odmo4o6/convert

If we isolate our containers from host, i.e, --contain flag in singularity, then the container /tmp may run out of memory to store all the nifti files. While we have ways to work around this, it took a while to figure out what was going on. Our nii.gz where just showing missing blocks. I think it is unexpected that heudiconv would be using a hardcoded /tmp to temporarily store conversion. Is there a way to change the path or make that assumption more clear to the user?

Platform details:

Choose one:

• Local environment

• Container

This is a custom container
https://hub.docker.com/repository/docker/brownbnc/xnat-tools
All versions contain the same version of heudiconv and dcm2nii
While we create a docker container we use it mostly as a singularity container from our HPC

• Heudiconv version:

• heudiconv —version 0.6.0
• dcm2niiX version v1.0.20190902

[yarikoptic]

On one hand I am not sure if heudiconv should use any other directory as "temporary" unless explicitly instructed to do so. But on the other -- large converted files might then end up on a different partition from the target one, so not only exhaust /tmp but also result in less efficient transfer to the target location (yet to check if we cp or mv here ;)). So we might want to reconsider.

In either of those cases a workaround would be to pass/set environment variable TMPDIR for the run of the container to point to some location with sufficient storage/on the same file system (-e TMPDIR=PATH for docker? didn't try)
Note: Due to still open #462 care should be taken to ensure that the TMPDIR is not within target output directory, since it might pollute target output location.

[yarikoptic]

Oh, and on "why" -- we "group" dicoms to be converted from a single series (source directory could have many series dumped into a single directory) and place into that temporary directory to pass to dcm2niix.

edit: I guess heudiconv could be smarter, and "if to be converted files are the only files in the directory, do not bother copying to /tmp first", but it would need a check if it is not something nipype does on our behalf (didn't check the code) and would be harder to avoid.

[mirestrepo]

I was just running fmriprep and was reminded that they have an input -w, --work-dir, to set their working directory, which I think it serves the same purpose. Would it make sense for heudiconv to do the same?