[BUG] Argparse `boolean_optional` action handled sub-optimally

[michellewang]

With this argdump:

  {
    "$schema": "https://niwrap.dev/argdump/schema-v1.json",
    "$env": {
      "python_version": "3.14.7",
      "python_implementation": "CPython",
      "platform_system": "Linux",
      "platform_release": "6.17.0-1022-azure",
      "platform_machine": "x86_64",
      "argdump_version": "0.1.3"
    },
    "prog": "fmriprep",
    "description": "fMRIPrep: fMRI PREProcessing workflows v26.0.0.dev1+g4092cba59",
    "formatter_class": "ArgumentDefaultsHelpFormatter",
    "actions": [
      {
        "option_strings": [
          "--submm-recon",
          "--no-submm-recon"
        ],
        "dest": "hires",
        "action_type": "boolean_optional",
        "nargs": 0,
        "default": true,
        "help": "Enable or disable sub-millimeter (hi-res) reconstruction."
      }
    ]
  }

I get this descriptor:

{
  "schema-version": "0.5+styx",
  "name": "fmriprep",
  "description": "fMRIPrep: fMRI PREProcessing workflows v26.0.0.dev1+g4092cba59",
  "command-line": "fmriprep [SUBMM_RECON]",
  "inputs": [
    {
      "id": "submm_recon",
      "type": "String",
      "value-key": "[SUBMM_RECON]",
      "description": "Enable or disable sub-millimeter (hi-res) reconstruction.",
      "optional": true,
      "value-choices": [
        "--submm-recon",
        "--no-submm-recon"
      ],
      "name": "submm_recon"
    }
  ]
}

Which is annoying because then the invocation has to be something like this:

{
  "submm_recon": "--submm-recon"
}

Better to have a descriptor like this (mistakes possible):

{
  "schema-version": "0.5+styx",
  "name": "fmriprep",
  "description": "fMRIPrep: fMRI PREProcessing workflows v26.0.0.dev1+g4092cba59",
  "command-line": "fmriprep [SUBMM_RECON_OR_NO_SUBMM_RECON]",
  "inputs": [
    {
      "id": "submm_recon_or_no_submm_recon",
      "description": "Enable or disable sub-millimeter (hi-res) reconstruction.",
      "type": [
        {
          "command-line": "[SUBMM_RECON]",
          "inputs": [
            {
              "id": "submm_recon",
              "type": "Flag",
              "value-key": "[SUBMM_RECON]",
              "command-line-flag": "--submm-recon",
              "name": "submm_recon"
            }
          ],
          "name": "submm_recon",
          "id": "submm_recon"
        },
        {
          "name": "no_submm_recon",
          "id": "no_submm_recon",
          "command-line": "[NO_SUBMM_RECON]",
          "inputs": [
            {
              "id": "no_submm_recon",
              "type": "Flag",
              "value-key": "[NO_SUBMM_RECON]",
              "command-line-flag": "--no-submm-recon",
              "name": "no_submm_recon"
            }
          ]
        }
      ],
      "value-key": "[SUBMM_RECON_OR_NO_SUBMM_RECON]",
      "optional": true,
      "name": "submm_recon_or_no_submm_recon"
    }
  ]

which would allow an invocation like this, which I feel is better:

{
  "submm_recon_or_no_submm_recon": {
    "submm-recon": true
  }
}

Though in general the <A>_OR_<B> pattern feels a bit convoluted. But that seems to be the easiest way to encode mutually exclusive groups without relying on the Boutiques-specific groups field

[nx10]

Fixed for the Python and TypeScript wrappers in #90; the Boutiques descriptor is unchanged. You now get submm_recon: bool | None = True, emitting --submm-recon when true and --no-submm-recon when false. boolean_optional is the one case where argparse tells us outright that two spellings are one boolean (one action, one dest).

On the wider design question:

• The natural type for this pattern is an enum, which we emit just with literals not named members
• e.g. fslcomplex has a nine-way exclusive flag group; as sub-descriptors that would become {"output_type": {"realabs": true}} across nine descriptors instead of output_type="realabs"

What those cases actually want is the enum keeping its member names (realabs) while emitting the token (-realabs). Boutiques is missing an elegant way to express that. It's maybe worth doing for the generated wrappers at some point.