GTR015 — format=”input” → format_source (runtime-gated)

Contract: behaviour-preserving (provable subset).

Claim

Rewriting format="input" to format_source="<name>" preserves behaviour when the tool has exactly one data input, top-level or conditional/section-nested.

Proof

• Galaxy keys input_datasets by prefixed (qualified) name (tools/actions/__init__.py:222; prefixed keying back to release_16.04); determine_output_format looks format_source up in exactly that map (:1283-1286). Conditional/section ancestors contribute name|; <when> contributes nothing (visit_input_values). Qualified references are upstream-tested (test/functional/tools/format_source_in_conditional.xml).

• Absent-at-runtime corner (unselected branch / empty optional): pre-16.04 format="input" resolved to "data" (no datasets walked); a missing format_source key falls through to the parsed format default — also "data" (parser/xml.py:1453). Galaxy’s own conditional test tool exercises this fallthrough.

Scope boundary (construction-unprovable)

Multi-input (Galaxy resolves bare format="input" to a random input’s ext — its own TODO-marked nondeterminism: no deterministic behaviour exists); zero-input; repeat-nested (instance-indexed prefix r_0| — no static address); outputs already carrying format_source (inert format="input"; §24 guard).

History

Top-level-only → qualified-nested widening, codemod §40 (0 corpus tools — the 1 nested corpus tool is repeat-nested; novel-tool insurance).