Issues with tool that accepts a paired input, inside or out of a subworkflow

[hexylena]

Describe the bug

I have a list of paired datasets. I need to run shovill (and some other tools) over that list of datasets. When I run shovill at the top level, it'll produce a collection as an output. Beautiful! However when it's run in a subworkflow, either one that accepts list:paired or paired, it outputs a single dataset (according to the workflow editor.) When this is actually run, it outputs a collection. The workflow editor's got it wrong (afaict)

This becomes a problem because I want to take a list of pair, run shovill over each of those pairs (in a subworkflow that uses a tool that does a 'reduce' step, so it cannot easily be run at the top level), and then merge that output collection with some others. The fact that it's recognised as a 'dataset' of course prevents its use in the merge collections tool.

Galaxy Version and/or server at which you observed the bug
Galaxy Version: 24.1.4.dev0

Browser and Operating System
Linux/chrome

[hexylena]

Here's the sample workflow if it helps. Galaxy-Workflow-shovill-comp.ga.txt

[hexylena]

Running this workflow, every single output is a collection. 5/6 of which can't be used in a merge collection tool.

My Expectations

WF Input	Subworkflow Input	Current Output	Expected Output
paired	none	dataset	dataset
paired	paired	dataset	dataset
paired	list:paired	dataset	❌ collection? It is odd to provide a paired input to a list:paired
list:paired	none	collection	collection
list:paired	paired	dataset	❌ collection
list:paired	list:paired	dataset	❌ collection

[hexylena]

another variant of the issue: i need to have a way to force collection processing rather than multiple=true processing. I have a step that wants to be a reduce step, it'll reduce whatever it's given. But I don't want that. I want my collection identifiers so I can properly re-merge the collections later. I'm stuck accepting the reduce and using 'split on column' which doesn't ensure we have the same length of collection.

Having a way to force mapping would definitely help here.