EPBDS-16211 make the plan tell the truth: 121 corrections - #51
Merged
Merged
Conversation
A verification sweep read all 1329 lines of docs/UPGRADE-PLAN.md and checked
every trackable claim against the repository. The document was wrong in both
directions, and the plan is what the next person reads before touching this
code, so both directions matter.
104 STALE MARKERS — work presented as open that had in fact shipped. Most of
it because a PR fixed something and never came back to the plan. Phases 1, 2
and 3 are all complete and all three read as pending; the status row still
described a "React 17 hosted checkpoint" months after the React 18 bump.
16 FALSE CLAIMS — assertions the repository contradicts. The ones that could
mislead someone into a decision:
- "react-final-form is the ONLY remaining React ≤18 cap" — it is not.
react-final-form-arrays 3.1.4 is a direct dependency carrying the
identical cap. Stated in three places, including the comment in ci.yml
that justifies --legacy-peer-deps. Corrected there too: the §9.7-F4 bump
has to move both packages together (arrays 5.0.0 requires rff ^7.0.1,
verified against the registry).
- Appendix C ticks "React 19 flip: peers widened to include ^19". They are
not widened; package.json still reads ^16.14.0 || ^17.0.0 || ^18.0.0.
A reader could publish that claim to a consumer.
- The pack measurements (295 files / 7.25 MB / 2.53 MB) were stale in five
places. Re-measured with the repo's own gate: 299 / 6.44 MB / 2.39 MB —
more files, less weight, after the semantic-ui exit.
- The plan's own correction saying FIELD.METHODS "does not exist" is wrong.
It exists at rules.js:1109 and :1127, and it is precisely the hazard
class finding 5 catalogues: a per-instance write onto a module-global.
- Every module size in Appendix A was understated. Render.js is 223 lines,
not the recorded 133. Phase 6 is scoped against those numbers.
- "No legacy React APIs anywhere in src" is ticked while Upload.js:34 calls
event.persist(); §2.4 records this correctly, Appendix C never did. The
global-listener inventory missed the two document listeners the SUIR exit
added (Listbox.js:206, TooltipPop.js:369).
Nothing is deleted. Every correction supersedes the old text and keeps it
visible, in the style the document already uses, so a reader can see what
changed and why. 13 candidate corrections were dropped — two because the
re-check refuted them and the plan was right as written, the rest as
duplicates already covered by an adjacent edit.
The version bump is NOT part of this: the release stays the maintainers' call.
AlexSamBY
added a commit
that referenced
this pull request
Sep 17, 2026
Replaces #52, which conflicted after #51 was squash-merged: this is the same single H1 commit, rebased onto `master`. Identical diff — 32 files, +61 −3150. ## What went | | | |---|---| | **12 orphan components** | `Avatar`, `Badge`, `Carousel`, `Collapse`, `ErrorContent`, `ErrorTable`, `FloatNumber`, `ImageSwatch`, `MenuButton`, `Square`, `TabList`, `Tags` | | **10 LESS files** | the whole `src/style/unused/` tree | | **4 icomoon artifacts** | `demo.html`, `demo-files/`, `Read Me.txt` | | **2 functions** | `formatTime`, `toHours` in `time.js`, with the imports that existed only for them | ## What deliberately stayed **The engine `TabList`** (`pages/main/components/TabList.js`) — `mapper.js:33` resolves it, and it is *not* interchangeable with the pack copy: both `import './Tabs'`, and those two `Tabs` differ by ~100 lines. Swapping them would have silently regressed tab behaviour. **The pack `Tabs.js` also stays**, so H6's duplicate pair is only half resolved. **icomoon's `selection.json` and `style.css`** — `selection.json` is the project file you re-upload to regenerate the icon set, the only round-trip path back to it. Dead to every build, but its loss is not recoverable the way a demo page's is. **`FIELD.TYPE.DATE`**, which H1 also lists — delete it, or give it a `view: 'Date'` branch? That is an owners' call, so it was left alone. ## One decision this forced `ErrorTable` could not go alone (`ErrorContent` imported it), and it was also the last in-repo caller passing `inverted`/`striped` to `Table`. `SUPPORTED-PROPS.md` recorded that step 1 deferred their fate to exactly this deletion. **They are kept.** Both are genuinely styled (`table:not(.as-layout).inverted`, `table.striped tr:nth-child(2n)`), `Table.js` still destructures them, and a consumer meta can still set them. Dropping a supported prop because our own last caller went away would be a breaking change bought for nothing. The generated "Attributes at the call sites" table no longer lists them — which is *correct*, since that table reports what the codebase passes, not what the component accepts — so `wrapper-prop-curation.js` now says so explicitly. ## Verified before deleting One agent per target, each asked to **refute** that its target was safe: mapper registration, the components barrel, the published entry, every import spelling, tracked metas, docs, schema, CSS, the sibling checkouts. `src/style/unused/` was cleared by compiling `index.less` through the repo's own `less-options` and reading less's resolved import list (44 files, none under `unused/`) — not by grepping basenames, which would have been wrong here: `avatar.less` and `carousel.less` exist elsewhere too. ## Measured after | | before | after | |---|---|---| | suites / tests | 166 / 2539 | 166 / **2527** | | statements | 94.83% | **97.04%** | | branches | 90.19% | **94.06%** | | functions | 93.28% | **96.88%** | | lines | 95.36% | **97.76%** | The −12 tests are exactly the two time suites (−9, −3); a per-suite diff confirms no other suite's count moved. Coverage thresholds green. `static/all.css` sha256 **unchanged** at `dcdb0a40a6d8ee79…` — deleting `unused/` and the icomoon page touched no byte a consumer downloads. `lint:js`, `lint:css`, `css:fixture:check`, both docs checks, `build-lib` and `build-css` all green. ## Side effects worth naming - `moment` is now imported **nowhere** in `src/core/utils/` — progress on §9.7-F2. - Workstream A loses 4 class components and 3 `UNSAFE_*` sites (29/14 → **26/12**, re-measured). - The third function-component `defaultProps` site is resolved by deletion. The plan is updated in the same commit — 28 places closed or corrected, including the orphan counts, which are now a historical record rather than a list of files to go looking for. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
23 agents read all 1329 lines of
docs/UPGRADE-PLAN.mdand checked every trackable claim against the repository. The document was wrong in both directions. The plan is what the next person reads before touching this code, so both directions matter.104 stale markers — work presented as open that had shipped
Mostly because a PR fixed something and never came back to the plan. The largest case: phases 1, 2 and 3 are all complete and all three read as pending. The status row still described a "React 17 hosted checkpoint" months after the React 18 bump landed in 713c1e8.
react-17createRootatmain.jsx:12, RTL 12→16, devDeps^18.3.1meta.schema.json, error boundaries, 16/17 matrix16 false claims — assertions the repository contradicts
The ones that could mislead someone into a decision:
react-final-formis "the ONLY remaining React ≤18 cap" — it is not.react-final-form-arrays@3.1.4is a direct dependency carrying the identical cap. Asserted in three places, including the comment in.github/workflows/ci.ymlthat justifies--legacy-peer-deps. Corrected there too: the §9.7-F4 bump has to move both packages together —react-final-form-arrays@5.0.0requiresreact-final-form ^7.0.1, verified against the registry.Appendix C ticks "React 19 flip: peers widened to include ^19". They are not widened —
package.jsonstill reads^16.14.0 || ^17.0.0 || ^18.0.0. Nothing is broken in the repo; the risk is a reader treating React 19 as supported and telling a consumer so.The pack measurements were stale in five places. 295 files / 7.25 MB / 2.53 MB → re-measured with the repo's own gate: 299 / 6.44 MB / 2.39 MB. More files, less weight, after the semantic-ui exit.
The plan's own correction saying
FIELD.METHODS"does not exist" is wrong. It exists atrules.js:1109and:1127— and it is precisely the hazard class finding 5 catalogues: a per-instance write onto a module-global object.Every module size in Appendix A was understated.
Render.jsis 223 lines, not the recorded 133 (+68%). Phase 6 is scoped against those numbers."No legacy React APIs anywhere in src" is ticked while
Upload.js:34callsevent.persist()— §2.4 records this correctly, Appendix C never did. The global-listener inventory also missed the twodocumentlisteners the SUIR exit added (Listbox.js:206,TooltipPop.js:369).Method
Extraction and verification fanned out across the document; every edit was drafted against a re-check of the evidence, then applied serially with a uniqueness assertion on each anchor — all 120 matched verbatim and exactly once. 13 candidate corrections were dropped: two because the re-check refuted them and the plan was right as written, the rest as duplicates already covered by an adjacent edit.
Nothing is deleted. Every correction supersedes the old text and keeps it visible, in the style the document already uses, so a reader sees what changed and why.
Structure verified unchanged: 104 headings, 196 table rows, balanced fences.
ci.ymlre-parsed — all five jobs intact, comment-only diff.lint:js,lint:css,css:fixture:checkand both docs checks green.The version bump is not part of this — the release stays the maintainers' call.
🤖 Generated with Claude Code