Plan: API Tests for Roundtrip Representation Artifacts

Status: IMPLEMENTED — commit 0406f008e1

Goal

Verify that Galaxy correctly executes workflows whose native tool_state contains representation artifacts from schema-unaware format2 round-trip conversion.

The Artifacts

Four representation changes from schema-unaware gxformat2 conversion:

1. Multiple select: comma-delimited string → JSON list

• Native: "comment_char": "35" (comma-delimited string)
• After conversion: "comment_char": ["35"] (JSON list)
• Galaxy’s multiple_select_value_split() accepts both str and List[str]

2. Sections with all-None values dropped

• Native: "global_trimming_options": {"trim_front1": null, "trim_tail1": null}
• After conversion: key absent entirely
• gxformat2 omits state keys whose values are None/null

3. Empty repeat lists dropped

• Native: "queries": [], "sql_stmts": []
• After conversion: key absent entirely
• gxformat2 omits empty lists

4. Boolean string case normalization

• Native: "shorten_values": "False" (capitalized string)
• After conversion: "shorten_values": false (JSON boolean)
• IWC workflows (cutandrun, atacseq, VGP) have capitalized string booleans

Implementation

Test Data

lib/galaxy_test/base/data/wf_conversion/
├── README.md
├── format2/                        # Source of truth — human-authored
│   ├── multiple_select.gxwf.yml    # gx_select_multiple tool, single + multi value
│   ├── allnone_section.gxwf.yml    # gx_section_boolean tool, no state set
│   ├── empty_repeat.gxwf.yml       # simple_constructs tool, no repeat instances
│   └── boolean_case.gxwf.yml       # gx_boolean tool, true + false steps
├── multiple_select.ga              # ["--ex1"] list-form in tool_state
├── allnone_section.ga              # section key absent entirely
├── empty_repeat.ga                 # repeat key absent entirely
└── boolean_case.ga                 # true/false JSON booleans

All format2 workflows use test tools from test/functional/tools/parameters/ — no ToolShed dependencies.

API Tests

lib/galaxy_test/api/test_wf_conversion_artifacts.py — 6 workflow tests:

lib/galaxy_test/api/test_tool_execute.py — 5 direct API tests (x3 input formats = 15):

Findings

Key Finding: Absent Repeats Are Mostly Safe

Initial testing with simple_constructs showed absent repeats causing job errors. Follow-up with gx_repeat_optional revealed this was a tool template bug (dangling && when repeat is empty), not a fundamental absent-repeat problem.

visit_input_values (line 236: input_values.get(input.name, [])) silently defaults missing repeats to [] in the in-memory state. This is sufficient for tools that use len() and #for (which handle empty lists). It fails only for tools that:

• Directly index into the repeat ($files[0]) when empty
• Have template syntax errors exposed by empty repeats (dangling &&)

Two different state initialization paths exist but both handle absent repeats:

The workflow path’s backfill is sufficient for execution when the template handles empty repeats gracefully. Schema-aware conversion preserving [] would still be cleaner but is not strictly required for correctness.

Absent Section Requires UI Follow-up

The allow_tool_state_corrections flag is needed because Galaxy’s workflow upgrade checker flags the missing section as needing correction. TODO: add a UI/selenium test to verify these workflows can be opened in the editor and the upgrade banner is handled correctly.

Resolved Questions

• Which stock tools exercise each artifact type? gx_select_multiple (multiple select), gx_section_boolean (section), simple_constructs + gx_repeat_optional (repeat), gx_boolean (boolean) — all in test/functional/tools/parameters/
• Absent section vs section-with-all-None? Galaxy treats them equivalently with allow_tool_state_corrections, but flags the absent section as an upgrade message.
• Absent repeat vs empty repeat? visit_input_values defaults absent repeats to [] in memory. Tools with good templates work fine. The simple_constructs failure was a template bug (dangling &&), not a parameter handling issue.
• Does allow_tool_state_corrections matter for absent repeats? No — Galaxy doesn’t generate an upgrade message for absent repeats. visit_input_values silently backfills [] regardless of the flag.
• Is this a workflow-specific problem? The code paths differ (populate_state vs params_from_strings), but both ultimately handle absent repeats. Direct API execution is cleaner (initializes defaults upfront); workflow execution relies on the backfill in visit_input_values. Both work.
• Boolean case? Galaxy handles all forms (true, "true", "True", True) equivalently.
• Multi-value multiple selects? Tested — ["--ex1", "ex2"] works correctly.