feat: Backstage 1.55.0 remove ignored github:repo:create inputs

[schultzp2020]

Summary

Closes #153. For Backstage 1.55.0 Template YAML, remove the 20 now-ignored direct input keys only on spec.steps[] with exact action: github:repo:create. Preserve unrelated steps/keys, comments, YAML spelling, and multi-document boundaries. Quoted target actions under tags/anchors, including aliases to quoted anchors, receive an idempotent review TODO rather than silently skipping affected input. Folded github:repo:create-extra stays untouched. An unreferenced anchored input mapping is edited directly; shared anchored input, merges, ambiguous aliases/expressions, and folded target actions retain original input and receive a review TODO. No AI fixup.

Fixture before → after (the non-target github:repo:push step retains its protectDefaultBranch):

     - id: create

       action: github:repo:create

       input:

         repoUrl: github.com?repo=demo&owner=team

-        protectDefaultBranch: true

-        sourcePath: ./skeleton

-        requiredLinearHistory: true

         description: keep me

Package name: @backstage/remove-ignored-repo-create-inputs (44 chars), rather than the 51-character ID proposed in #153; approved to satisfy the registry limit. Minor changeset: .changeset/little-emus-invite.md. Root README.md is byte-identical to current PR base (594ea8f5735e99fa549d3efe9228c145f2cf3d67): manual README index sync remains separate per CONTRIBUTING.md.

Evidence

• Follow-up gate commit 6fe89df5b1b664860df7090767e6afee83306dd2, fast-forwarded from 36cbcf2dcf71ad35d8ee2790ccd6444c3271e827 (prior edit fix); merge commit 522798c04b49b3287975a539c35f5e2b71d64b7e integrates main 594ea8f5735e99fa549d3efe9228c145f2cf3d67. No force push; independent exact-head reconciliation review found no blockers.

• Dense gate now runs from package yarn test → root yarn test and changed-package CI. Same 1,200-step workflow runs three no-op and three real transforms, taking the median of each; the no-op changes one action character per step but keeps byte length/parser shape. Real median must stay below 5× no-op median; no absolute timeout. Every result is byte-compared, plus a repeat run verifies idempotence. Swapping only scripts/codemod.ts to old 5781917 caused the new gate to fail (no-op=1,526ms, real=33,676ms; exit 1); current source passed (1,524ms/1,528ms; exit 0). Repeated samples reduce scheduling noise; timings remain synthetic and machine-dependent, not a production latency guarantee.

• Full shipped-workflow diagnostic, 3,005 files (2,700 eligible inert + 300 node_modules excluded + five positives): no-op 1.046s → real 1.047s → repeat 3.060s. Five positive files exactly match fixture SHA-256; no-op and repeat full-corpus hashes unchanged relative to preceding runs; excluded files unchanged. Dense shipped-workflow 600 steps: no-op 1.042s → real 1.039s → repeat 1.039s; 1,200 steps: 1.044s → 1.051s → 1.037s. Both dense outputs byte-match generated expected text, retain every description and repoUrl, remove 1,200/2,400 deprecated keys, and remain byte-identical on repeat. Baseline independent evidence: 600 real 9.927s, 1,200 real 35.979s (3.62× doubling); new observed real times are near no-op. Hash snapshots, argv, return codes, and CLI logs: C:/Users/pschu/AppData/Local/Temp/jssg-pr170-optimized-q3tbn6q4. First no-op benchmark run completed but log writing failed on Windows cp1252; rerun succeeded and is the reported no-op timing. Callback counts were not instrumented.

• 55/55 JSSG fixtures and workflow validation passed, covering CRLF without EOF newline, Unicode, comments, aliases/TODO and idempotence. yarn format, yarn format:check, yarn lint, package yarn test, root yarn test, yarn check:changed, and git diff --check passed locally; root test median=1,525ms no-op/1,517ms real. Husky pre-commit/pre-push hooks active; neither bypassed. No --allow-fs, YAML reserialization, hand-edited versions, or root README churn. Existing minor changeset .changeset/little-emus-invite.md remains.

• Full root yarn test creates the known feat(v1.55.0): add remove-relations-compatibility-config codemod #162 untracked 145-byte codemods/v1.55.0/remove-relations-compatibility-config/tests/flow-inline-comment-preserved-when-catalog-emptied/metrics.json (Git blob 42847c1041d21d4278619381aa2d51455bc1b654). The exact path/size/blob were verified, the file backed up outside the worktree at /tmp/pr170-metrics-a44Lh0.json, then only that artifact was removed; root tests do not leave the worktree clean without this cleanup.

• Exact-head GitHub CI Test and Changeset Check succeeded at 522798c04b49b3287975a539c35f5e2b71d64b7e: https://github.com/backstage/codemods/actions/runs/38457958066 . Test passed 55 fixtures and the integrated dense gate (no-op=1,270ms, real=1,271ms; synthetic medians of three).

Merge Danger

Door: two-way; additive codemod package + fixtures + changeset, reversible by revert. Blast radius: only this YAML migration package; existing templates change only when users run it. Root README unchanged versus PR base. Dense-file timing is synthetic, repeated same-size relative evidence, not a production throughput guarantee; fresh independent exact-head review found no blockers. Only #170 is in scope for merge; Version Packages #177 and draft recipes #158–#160 remain unmerged; publishing stays separate.

[schultzp2020]

Release blocker (awaiting issue owner decision): the issue-required package ID @backstage/remove-ignored-github-repo-create-inputs is 51 characters; CI Package name length check enforces max 50 for Codemod Registry. This PR and dependent draft recipe #158 must not merge/publish until an approved coordinated #153/#156 package-ID change is made. No CI bypass or speculative rename.

[schultzp2020]

Independent code review gate remains blocked despite name-length CI fix: flow input mapping reconstruction may discard unrelated comments; spec/steps helper descendant search may select nested path. Sole writer adding commented-flow and nested-path negative fixtures with surgical or conservative AST fix. Keep draft until fresh corrected review and exact-head CI.

[schultzp2020]

Fresh actual-diff final review at ed2b941 found P2 false TODO: anchored direct input: &shared { repoUrl, description } with NO removed keys still gets manual-review TODO because shouldAddTodo initializes from inputHasAnchor unconditionally (scripts/codemod.ts:387). Add red anchored-no-removed-noop fixture and gate TODO on affected/ambiguous removed keys; keep draft. Reviewer’s earlier hung disposable CLI probe was terminated with supervisor approval/unverified; separate 15/15 fixtures, check:changed, idempotence and synthetic perf checks passed.