SoftwareRequirement: environment modules changes to env not propagating to tool

Expected Behavior

If I have a CWL file with a SoftwareRequirement that is resolved by an environment module, then I would expect that the environment variables set propagate to the tool execution.

Actual Behavior

They do not: only PATH is modified by the environment module

Workflow Code

Note this depends on having environment modules set up. Modify the cwl to load a module you have available:

env.cwl:

#!/usr/bin/env cwl-runner
# -*- mode: yaml; -*-

cwlVersion: v1.1
class: CommandLineTool
$namespaces:
  cwltool: http://commonwl.org/cwltool#

hints:
  SoftwareRequirement:
    packages:
      ncl:
        version: ["6.6.2"]

inputs: []

baseCommand: env

outputs:
  output:
    type: stdout

modules.yml:

- type: modules
  modulecmd: /usr/bin/modulecmd

cwltool runs OK:

 cwltool --beta-dependency-resolvers-configuration modules.yml env.cwl 
INFO /lustre/home/shared/dc118/sw/cwl/3.0.20201117141248/bin/cwltool 3.0.20201117141248
INFO Resolved 'env.cwl' to 'file:///lustre/home/dc118/nashvest/env.cwl'
INFO [job env.cwl] /tmp/7u0vetk_$ env > /tmp/7u0vetk_/d5514f0d054e63af38037aba76940ead71fd7324
Autoloading gcc/6.3.0
Autoloading gsl/2.6-gcc6
INFO [job env.cwl] completed success
{
    "output": {
        "location": "file:///lustre/home/dc118/nashvest/d5514f0d054e63af38037aba76940ead71fd7324",
        "basename": "d5514f0d054e63af38037aba76940ead71fd7324",
        "class": "File",
        "checksum": "sha1$488689425380b273e1e59a0dd3fdc884a57a842c",
        "size": 471,
        "path": "/lustre/home/dc118/nashvest/d5514f0d054e63af38037aba76940ead71fd7324"
    }
}
INFO Final process status is success

Module that should have been loaded:

$ module show ncl
-------------------------------------------------------------------
/lustre/sw/modulefiles/ncl/6.6.2:

conflict        ncl
Autoloading gcc/6.3.0
module load     gcc/6.3.0
Autoloading gsl/2.6-gcc6
module load     gsl/2.6-gcc6
setenv          NCARG_ROOT      /lustre/sw/ncl/ncl-6.6.2
prepend-path    PATH /lustre/sw/ncl/ncl-6.6.2/bin
prepend-path    LD_LIBRARY_PATH /lustre/sw/ncl/ncl-6.6.2/lib
prepend-path    LD_LIBRARY_PATH /lustre/sw/ncl/ncl-6.6.2/lib64
prepend-path    MANPATH /lustre/sw/ncl/ncl-6.6.2/man
-------------------------------------------------------------------

Actual environment

$ cat d5514f0d054e63af38037aba76940ead71fd7324 
PATH=/lustre/sw/ncl/ncl-6.6.2/bin:/lustre/sw/gsl/2.6/bin:/lustre/sw/gcc/6.3.0/bin:/lustre/home/shared/dc118/sw/cwl/3.0.20201117141248/bin:/lustre/sw/spack-cirrus/opt/spack/linux-centos7-x86_64/gcc-8.2.0/git-2.21.0-rcchd4zgfdherdlklrr2y3amq7p73svi/bin:/lustre/sw/epcc-utils/bin:/opt/clmgr/sbin:/opt/clmgr/bin:/opt/sgi/sbin:/opt/sgi/bin:/usr/share/Modules/bin:/usr/local/bin:/usr/bin:/usr/local/sbin:/usr/sbin:/opt/c3/bin:/sbin:/bin
HOME=/tmp/7u0vetk_
TMPDIR=/tmp/9af437kp

If I change the command to say cwltool --beta-dependency-resolvers-configuration modules.yml --preserve-environment NCARG_ROOT env.cwl then the output of env is unchanged

Your Environment

• cwltool version: 3.0.20201117141248

@mr-c and I have discussed briefly on gitter: https://gitter.im/common-workflow-language/common-workflow-language?at=5fbfb71303ddd74df2f90e59 Where Michael suggests pinging @jmchilton and @hmenager for comment?

[rupertnash]

Maybe instead we should:

• construct an environment as cwltool does now
• run the shell commands produced by DependencyManager
• capture that environment
• use the captured environment to run the tool

Thoughts?

[rupertnash]

Galaxy docs say:

A DependencyManager attempts to resolve named and versioned dependencies by searching for them under a list of directories. Directories should be of the form:

    $BASE/name/version/...

and should each contain a file ‘env.sh’ which can be sourced to make the dependency available in the current shell environment.

A brief review of the resolvers suggests that they all act in the bold text manner. This make me think that the way cwltool is trying to use this feature (getting the PATH from the modified shell environment and then taking total control of the environment) is a bit wrong