Argo workflow template execution failing when I use when condition with withParam loop along with the artifact input - conditional-operator

Question

I've the following workflow template which has when condition when: "'{{item}}' =~ '^tests/'", artifacts input as file path in AWS S3 bucket and withParam loop.
Here is my workflow template
apiVersion: argoproj.io/v1alpha1
kind: WorkflowTemplate
metadata:
name: process-wft
spec:
entrypoint: main
templates:
- name: main
inputs:
parameters:
- name: dir-process
default: true
- name: dir
artifacts:
- name: Code
dag:
tasks:
- name: process-wft-tests
when: "'{{item}}' =~ '^tests/'"
templateRef:
name: tf-wf-rn
template: main
arguments:
parameters:
- name: dir-process
value: "{{inputs.parameters.dir-process}}"
artifacts:
- name: Code
from: "{{inputs.artifacts.Code}}"
withParam: "{{inputs.parameters.dir}}"
here is my input artifact Code extracted result which is being passed from my workflow
inputs:
artifacts:
- archive:
tar:
compressionLevel: 9
archiveLogs: true
globalName: GitSource
name: Code
path: /mnt/out/code
s3:
key: process-kfxqf/process-kfxqf-1938174407/GitSource.tgz
It is giving the below Error when I run my workflow
message: failed to resolve {{inputs.artifacts.Code}}
What's the mistake am doing here? if it doesn't work what's the alternate way to get this worked?
Note: I tried workflow execution by removing when condition, it is working fine. It is giving issue only when I add when condition.

Answer 1

This is a bug.
It seems that when the condition evaluates to false, some code skips populating the artifact (makes sense, save some time), but some other code doesn't respect the when condition and still expects the artifact to be populated.
Potential workarounds:
Move the conditional logic into the container.
remove the when condition
pass the dir parameter to the main template in your tf-wf-rn WorkflowTemplate
change the main template to run the regex against the dir parameter - if it doesn't match, just exit 0
This could make the workflow much slower, because you'll have to spin up a pod for each iteration of the loop to determine if there's actually any work to be done.
If you can calculate all the information about the artifact, pass that information as parameters to the main template in your tf-wf-rn WorkflowTemplate. Then actually load the artifact in that non-conditioned, non-looped template. (Basically hopscotch over the problematic code.)
Try an older version.
If you find a working older version, please 1) comment on the bug report and 2) make sure the older version doesn't have any relevant security vulnerabilities before running it on a production system.