[Proposal] Deprecating CommandInputEnumSchema#inputBinding

[tom-tan]

It is available since CWL v1.0 but its usecase can be covered with CommandInputParameter#inputBinding and CommandInputRecordField#inputBinding, I guess.

In addition to that, the current behavior of CommandInputEnumSchema#inputBinding with CommandInputParameter#inputBinding in cwltool seems to be unintended.

Here is an example.

enum-order-root-leaf.cwl:

class: CommandLineTool
cwlVersion: v1.2

baseCommand: python
inputs:
  - id: args.py
    type: File
    default:
      class: File
      location: args.py
    inputBinding:
      position: -1
  - id: inp1
    type:
      type: enum
      symbols: [a, b, c]
      inputBinding:
        position: 4
        prefix: -x
    inputBinding:
      position: 1
      prefix: -p
  - id: inp2
    type:
      type: enum
      symbols: [d, e, f]
      inputBinding:
        position: 3
        prefix: -z
    inputBinding:
      position: 2
      prefix: -q
outputs:
  - id: args
    type:
      type: array
      items: string

hints:
  - class: DockerRequirement
    dockerPull: docker.io/python:3-slim

enum-order-job.json:

{
    "inp1": "a",
    "inp2": "d"
}

I expect that it returns the following in which the value a and d occur only once:

{
    "args": [
        "-p",
        "-x",
        "a",
        "-q",
        "-z",
        "d"
    ]
}

However, here is the actual output of cwltool:

$ cwltool enum-order-root-leaf.cwl enum-order-job.json 
...
{
    "args": [
        "-p",
        "a",
        "-x",
        "a",
        "-q",
        "d",
        "-z",
        "d"
    ]
}

There are three options to fix it:

• Deprecate CommandInputEnumSchema#inputBinding not to care about such complicated cases
  • I prefer this option because I do not know practical usecases for it
  • It also makes the spec simpler
• Fix cwltool for nested inputBinding
  • It is possible. However, deprecating this field is easier for us
• Clarify the spec to support current behavior of cwltool
  • It needs much effort because the current behavior is hard to explain

What do you think about it?

Related: common-workflow-language/common-workflow-language#783

[tom-tan]

Note: Once we have a consensus about the deprecation, we can leave some of the cases for common-workflow-language/common-workflow-language#783 as undocumented in CWL v1.2.1.

[mr-c]

@tom-tan I agree with your expected results; @tetron and I think that this is a cwltool bug; we should add conformance tests for this.

[tom-tan]

we should add conformance tests for this.

Does it mean that we take the second option? (i.e., "Fix cwltool for nested inputBinding" in the above description)

[mr-c]

we should add conformance tests for this.

Does it mean that we take the second option? (i.e., "Fix cwltool for nested inputBinding" in the above description)

Yes, the second option