PickValueModule Post Job Actions — Debrief

What was done

Added PJA support to PickValueModule — change datatype, rename, tags, remove tags, column set. Users can configure these in the workflow editor UI and they execute during workflow invocation without a Job object.

Backend (lib/galaxy/workflow/modules.py)

• PJA round-trip: __init__, from_dict, from_workflow_step, save_to_step, get_post_job_actions — mirrors ToolModule pattern
• Execution: _apply_post_job_actions delegates to ActionBox.execute_on_mapped_over — no duplicated action logic, gets rename #{var|basename} template substitution for free
• _to_pja helper converts dict/PostJobAction values for DB persistence

Backend (lib/galaxy/job_execution/actions/post.py)

• Added execute_on_mapped_over to ChangeDatatypeAction — handles HDA and HDCA (collection element iteration)
• Added execute_on_mapped_over to ColumnSetAction — extracted _apply_column_set to deduplicate with job-based execute()
• Neither added to mapped_over_output_actions list (avoids changing existing tool behavior)

Frontend

• FormPickValue.vue — renders FormSection under “Additional Options” heading when datatypes available; accepts datatypes, nodeInputs, postJobActions props
• FormDefault.vue — extracts PJA data from useStepProps, passes to FormPickValue, wires onChangePostJobActions event
• NodeInspector.vue — forwards onChangePostJobActions from FormDefault to parent Index.vue

gxformat2 (~/projects/worktrees/gxformat2/branch/pick_value)

• converter.py — transform_pick_value now processes out section for PJAs (same pattern as tool steps)
• export.py — added _convert_post_job_actions call for pick_value export (round-trip)
• Syntax: out: { output: { change_datatype: txt } } — same as tool steps, no special shorthand

Framework workflow tests (4 tests, all passing)

What works well

• PickValueModule PJA handling mirrors ToolModule — consistent codebase pattern
• gxformat2 round-trip complete (converter + exporter)
• Frontend event chain properly wired end-to-end
• ColumnSetAction._apply_column_set refactor is a DRY improvement
• Reuse of ActionBox.execute_on_mapped_over avoids duplicating action logic

Known issues

1. Bug: PJAs applied to skipped outputs

Severity: High

In first_or_skip mode when all inputs are null, a skipped HDA (expression.json) is created and _apply_post_job_actions runs unconditionally. ChangeDatatypeAction would corrupt the skipped HDA’s datatype away from expression.json, breaking downstream skip detection.

The job-based ChangeDatatypeAction.execute() has a if job.state == SKIPPED: return guard, but execute_on_mapped_over does not.

Fix: Add skip guard at top of _apply_post_job_actions:

if self._is_null_or_skipped(output):
    return

2. _to_pja fallback key parsing

Severity: Low

When value is neither dict nor PostJobAction, action_type defaults to the full concatenated key (e.g., "ChangeDatatypeActionoutput" instead of "ChangeDatatypeAction"). Unlikely to hit — frontend always sends dicts, from_workflow_step produces PostJobAction objects.

3. ColumnSetAction.execute_on_mapped_over on HDCA

Severity: Low

Would fail on collection output (all_non_null mode) since collections lack .metadata. Low probability — column set on a collection is unusual.

Test gaps

• No framework test for when: false (skipped output + PJA) — would catch issue #1
• No framework test for all_non_null mode with PJAs (HDCA output path)
• No framework test for RemoveTagDatasetAction or ColumnSetAction
• No framework test for rename template substitution (#{input_name})
• No frontend unit tests for PJA rendering/events in FormPickValue.test.ts

Files changed