@Adam Taylor Updating your PR that removes ext.when for the segmenters so it's mergeable. Linting uncovered that you edited the local copy of the cellpose and mesmer modules without creating a

.diff

. I'm looking at the most recent upstream versions of those modules and it looks like cellpose added

process_gpu

so maybe we should just update for that. Can you explain what was fixed by adding the

stageAs

arg to

path(model, stageAs: '?/*')

?

@Jeremy Muhlich can you point me towards the PR... will review as to exactly what past me had run into... it will have been a Seqera Platform issue

it was https://github.com/nf-core/mcmicro/pull/141

stageAs addressed a consequence of the fact that the workflow invokes all three segmenters unconditionally. Which one actually runs is only decided by

ext.when

in

conf/modules.config

. eg

withName: "CELLPOSE" { ext.when = {params.segmentation…contains('cellpose')} }

Here CELLPOSE always feeds a default

params.cellpose_model

of

[]

, so if only MESMER is specified, the CELLPOSE process still gets an empty model input, which failed on

path(model)

when model was empty — on Batch this validation occurs before the process even kicks off, so it kills the workflow. Using

path(model, stageAs: '?/*')

as a way to stage optional paths allows us to bypass this step and enable the workflow to continue as expected.

Does the new

if (params.segmentation?.split(',')?.contains('cellpose'))

fix that without messing with

stageAs

?

I don't think so, because the workflow still gets invoked.