galaxy-tool-codemod

A LibCST-shaped framework for structural refactors of Galaxy tool XML. The structure tier of the Galaxy refactoring architecture:

Status

M1–M3.5 shipped: framework primitives (Module, Cursor, CodemodCommand), the corpus_check.py codemod sweep (retains failures as regression fixtures), and two ordered pipeline contracts run by the tier-4 app (galaxy-tool-refactor-cli):

• canonical_codemods() = FixTypos → NormalizeBooleanValues → RepairHelpRst → TrimAttributeWhitespace → ReplaceOutputElement → DropRedundantParamName → ReorderParamAttributes → ReorderToolAttributes → ReorderToolChildren → WrapCommandCdata → WrapHelpCdata → SingleQuoteCommandVars — the safe canonical/format pipeline (the app's format command), derived from the codemods declaring the "default" ruleset and ordered by meta.order (no hardcoded tuple, §36). Never changes profile=.
• AUTO_UPGRADE_CODEMODS = FixTypos → NormalizeBooleanValues → UpgradeToLatest — the opt-in profile-upgrade pipeline (the app's upgrade command).

The codemods:

• FixTypos — repair near-miss spelling typos so a well-formed-but-globally-invalid tool validates (in both pipelines);
• NormalizeBooleanValues — rewrite Python-style boolean attribute values (True/Yes/…) to canonical xs:boolean so a globally-invalid tool validates; schema-type-aware and behaviour-preserving (in both pipelines);
• UpgradeToLatest — loop UpdateProfile (declare the newest profile the tool validates at, bump-up-only) + single-step upgrade_vN codemods from upgrades.py to bring a tool to the latest profile (UpdateProfile is a building block run inside this loop, not a separate pipeline entry);
• ReorderParamAttributes / ReorderToolAttributes — IUC <param> and documented <tool> attribute order;
• ReorderToolChildren — IUC <tool> child-element order (best-practice #52); validity-safe because the <tool> content model is order-free (xs:all).

Every bundled codemod carries a RuleMeta GTR code; catalog.coded_codemods() is the authoritative enumeration (the prose list above is illustrative — format also runs the planemo-parity fixers GTR035–037 and the partition .1 fixers GTR018.1/019.1/020.1 and GTR089.1 RepairHelpRst). See docs/decisions.md §15–37.

The upgrade registry is grown empirically: the corpus_check codemod sweep reports STICKING POINT versions still needing an upgrade_vN, and each upgrade_vN's advance count. Upgrades shipped: Upgrade24_1 (24.1 → 24.2), Upgrade25_1 (25.1 → 26.0). See docs/decisions.md §11–14.

M4 (matcher language) and M5 (Cheetah reference resolver) are not yet implemented — see PLAN.md.

Public API

from pathlib import Path

from galaxy_tool_codemod.parse import parse_module
from galaxy_tool_codemod.canonical import canonical_codemods

module = parse_module(Path("tool.xml"))
for codemod_cls in canonical_codemods():
    codemod_cls().apply(module)
# module.document.tree now reflects the canonical structural form

Setup

From the workspace root:

uv sync
uv run --package galaxy-tool-codemod pytest galaxy-tool-codemod/tests/

Relationship to fmt and the app

Neither galaxy-tool-fmt's library nor its CLI depends on this package — fmt is cosmetic-only. Orchestration lives one tier up: the galaxy-tool-refactor app CLI (galaxy-tool-refactor-cli) hard-depends on this package and on fmt, runs canonical_codemods() (its format command) or AUTO_UPGRADE_CODEMODS (its upgrade command), and writes the result through fmt's serializer.

Coding standards

Hand-written code follows dignified-python (vendored at .claude/skills/dignified-python/). See CLAUDE.md.