refactor(react)!: autosizeColumns becomes setAllColumnsAutoWidth, and one default column width - #542
Merged
Merged
Conversation
… one default column width `autosizeColumns()` never sized anything to content: it put every column into the auto-width SET — the mode bit meaning "the grid manages this column's width". Nothing in the column-width path measures a cell, so the name promised a feature the library does not have. It is now `setAllColumnsAutoWidth(auto)`, symmetric with the per-column `setColumnAutoWidth(columnId, auto)` and, unlike the old name, able to express the other direction — `false` freezes every column at the engine's stored width. The surface prop `autosize?: boolean | AutosizeOptions` becomes `allColumnsAutoWidth?: boolean`, and `AutosizeOptions` is deleted from @pretable/core and @pretable/react: every field on it tuned a measurement pass that does not exist. The second half is the constant. An undeclared-width column was DRAWN by renderer-dom at 140px and STORED by grid-core at 160px, so turning auto width off on a never-resized column jumped it 140 -> 160. Both now read `DEFAULT_COLUMN_WIDTH_PX` (140, and 220 wrapped) from the new layout-core/src/column-defaults.ts, and @pretable/react seeds the engine's stored width through renderer-dom's own `resolveColumnWidth` — so the stored width and the drawn width of a never-resized column are the same number by construction, and the freeze moves no pixel. 140 won because it is what such columns have always painted at; moving the renderer to 160 would have repainted every example, bench scenario, and docs fixture that leaves widths undeclared. grid-core already depended on layout-core, so the new edge is one the manifest already allows. Also: double-clicking a column's resize handle was wired to an empty function. It now calls `setColumnAutoWidth(columnId, true)` — the pointer shortcut for handing that column's width back to the grid. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…er existed grid/column-layout.mdx's "Autosize" section taught `grid.autosizeColumn(columnId, options?)` and `grid.autosizeColumns(options?)` as fits to measured content width, with an `AutosizeOptions` paragraph about sample size and padding. No handle has ever carried either method, and the width path measures nothing — a reader who typed the line got a compile error out of the page that taught it. Its "Reset" section documented `grid.resetColumnLayout()`, which has also never existed: the only reset the library ships is the tool panel's Reset columns button. The section is now "Auto width" and describes the mode bit in the register tool-panel.mdx already uses — on means the grid manages the width (the renderer's default, or a flex share), off means manual at the engine's stored width, and nothing measures cell content anywhere. With the width defaults unified, turning auto off on a never-resized column no longer jumps, and the page says so. Two more phantoms the new guard found on other pages: grid/editing.mdx read `grid.getSnapshot().editing` (the handle's reader is `getState`) and grid/keyboard.mdx called `grid.selectAll()` (it is `selectAllVisibleRows`). Both fixed. The grid index's "not yet documented" list advertised a content-resizing `autosize` option on `usePretable`; that entry is gone rather than deferred, because the capability it promises is not deferred, it is imaginary. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Three phantom `grid.` methods survived every existing check because the
guards' vocabulary stops short of them: the import sweep reads
`import { X } from "@pretable/*"`, and the prose sweep matches
`Pretable[A-Z]...`, i.e. TYPE names. A lowercase method call in a sentence
or a fence is neither.
The new sweep matches `grid.<name>(` across every docs page and requires
the name to be declared as a member at the top level of some exported
interface or object type alias, pooled across every reported package. It
deliberately does not ask WHICH handle the surrounding prose holds — a
page may be writing about `PretableReactGrid`, `PretableSurfaceGrid`, or
the headless `PretableGridUiCore`, and deciding that from prose is a
reader's job, not a regex's. Erring wide is the conservative direction:
the guard is an authority on whether a method is REAL, not on where it
lives. Scoping to a `grid.` receiver plus a following `(` is what holds
the false-positive rate at zero on the current corpus — `pretable/
grid.css` (31 hits) and `gridRef.current` are not calls.
It fails closed twice: an empty member vocabulary would pass everything
and an empty corpus would check nothing, so both are asserted non-trivial.
Mutation-proven: renaming a documented call to
`grid.fitColumnToContent(columnId)` fails the guard by name and page, and
reverting turns it green.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
`AutosizeOptions` leaves both public reports, `autosizeColumns` becomes `setAllColumnsAutoWidth`, and the surface's `autosize` prop becomes `allColumnsAutoWidth`. `pnpm api:check` is green against the regenerated reports. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
The latest updates on your projects. Learn more about Vercel for GitHub. |
Contributor
Vercel preview readyPreview: https://pretable-g9i0jynvn-cacheplane.vercel.app Updated automatically by the |
`setColumnWidth` cleared a column's auto bit unconditionally, and a consumer with a controlled `state.columnWidths` replays its whole widths map through `setColumnWidth` on every write-back pass. So ANY re-render of such a consumer silently took every column in the map back out of the auto set. Both entry points to the mode bit were defeated in practice: the tool panel's Auto width toggle and the resize handle's double-click each set the bit and had it un-set before paint. Reproduced deterministically against the shape of the column-layout docs example — the very example the new prose points a reader at. The bit is now cleared only when the write actually MOVES the stored width. The comparison reads grid-core's store on both sides rather than comparing to the argument, because grid-core clamps against the column's min/max: a request that clamps back onto the current width is not a move either. The double-click no longer fires `onColumnWidthsChange`. That callback reports the ENGINE's stored widths and this gesture moves none of them — it announced a change that had not happened, and under a controlled consumer the announcement came straight back as the `setColumnWidth` replay above. The controlled round trip is pinned beside the double-click test, and mutation-checked: reverting the guard fails it with `expected 120 to be 140`, the declared width reasserted over the renderer's default. The unconditional clear predates this branch — it shipped with the tool panel's auto-width toggle — so the changeset discloses it as a user-visible fix rather than a consequence of the rename. Also softens the new docs guard's failure message: an author whose fence legitimately binds a local `grid` is now told to rename the variable rather than sent to regenerate a report. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Merged
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.
The rename
The grid handle's
autosizeColumns()never sized anything to content. It put every column into the auto-width set — the mode bit meaning "the grid manages this column's width". Nothing in the column-width path measures a cell, anywhere (verified in a browser during the SP5 arc). The name promised a feature the library does not have.It is now
grid.setAllColumnsAutoWidth(auto: boolean), symmetric with the per-columnsetColumnAutoWidth(columnId, auto)that shipped with the tool panel, and — unlike the old name — able to express the other direction:falsefreezes every column at the engine's stored width. The surface propautosize?: boolean | AutosizeOptionsbecomesallColumnsAutoWidth?: boolean, andAutosizeOptionsis deleted from@pretable/coreand@pretable/react: every field on it (averageCharWidth,cellPaddingPx, min/max) tuned a measurement pass that does not exist.Also fixed in passing: double-clicking a column's resize handle was wired to an empty function body. It now calls
setColumnAutoWidth(columnId, true)— the pointer shortcut for handing that column's width back to the grid, which is what the docs claimed all along.One default column width
An undeclared-width column was drawn by renderer-dom at 140px but stored by grid-core at 160px. Turning auto width off on a never-resized column therefore jumped it 140 → 160 for no reason a user could see — a jump SP5 had to document as expected behavior.
Both numbers now come from one home,
packages/layout-core/src/column-defaults.ts(DEFAULT_COLUMN_WIDTH_PX = 140,DEFAULT_WRAPPED_COLUMN_WIDTH_PX = 220), and@pretable/reactseeds the engine's stored width through renderer-dom's ownresolveColumnWidth. The stored width and the drawn width of a never-resized column are now the same number by construction, so the freeze moves no pixel.Why 140 and not 160. 140 is what every undeclared-width column has always actually painted at. Moving the renderer to 160 instead would have repainted every example, bench scenario, and docs fixture that leaves widths undeclared; moving the engine to 140 repaints nothing.
Dependency direction. grid-core already declares
@pretable-internal/layout-core: workspace:*, so the new import edge is one the manifest already allowed — no duplicated constants, noas anybridge.pnpm build,pnpm consumer:checkandpnpm react:compatall accept it.The phantom docs
grid/column-layout.mdxhad an "Autosize" section teachinggrid.autosizeColumn(columnId, options?)andgrid.autosizeColumns(options?)as fits to measured content width, plus anAutosizeOptionsparagraph about sample size and padding. No handle has ever carried either method. Its "Reset" section documentedgrid.resetColumnLayout(), which has also never existed — the only reset the library ships is the tool panel's Reset columns button.The section is now "Auto width", written in the mode-bit register
tool-panel.mdxalready uses, and the Reset section points at what actually exists. The SP5 jump paragraph intool-panel.mdxnow records the jump's absence.The guard finding
Three phantom
grid.methods survived every existing docs guard, because the guards' vocabulary stops short of them: the import sweep readsimport { X } from "@pretable/*", and the prose sweep matchesPretable[A-Z]…, i.e. type names. A lowercase method call in a sentence or a fence is neither.Extended honestly, in
docs-api-surface.test.ts: a new sweep matchesgrid.<name>(across every docs page and requires the name to be declared as a member of some exported interface or object type alias, pooled across every reported package. It deliberately does not ask which handle the surrounding prose holds — a page may be writing aboutPretableReactGrid,PretableSurfaceGrid, or the headlessPretableGridUiCore, and deciding that from prose is a reader's job, not a regex's. Erring wide is the conservative direction: the guard is an authority on whether a method is real, not on where it lives. Scoping to agrid.receiver plus a following(holds the false-positive rate at zero on the current corpus (pretable/grid.css, 31 hits, andgridRef.currentare not calls). It fails closed twice — an empty member vocabulary would pass everything, an empty corpus would check nothing.It found two more real defects on its first run, beyond the ones this PR set out to fix:
grid/editing.mdxreadgrid.getSnapshot().editing(the handle's reader isgetState), andgrid/keyboard.mdxcalledgrid.selectAll()(it isselectAllVisibleRows). Both fixed.Mutation-proven: renaming a documented call to
grid.fitColumnToContent(columnId)fails the guard by name and page; reverting turns it green.Test plan
pnpm build→pnpm api→pnpm api:checkgreen. Report moves are exactly the expected ones:AutosizeOptionsremoved from both public reports,autosizeColumns→setAllColumnsAutoWidth,autosize→allColumnsAutoWidth.pnpm test(root, turbo) green. Package counts: react 116 files / 1644 tests, website guards 104 files / 607 tests, core 15 / 181, grid-core 6 / 169, renderer-dom 9 / 168, layout-core 8 / 157, ui 4 / 99.pnpm lint,pnpm typecheck,pnpm format,pnpm consumer:check,pnpm react:compat— all green.next start,tool-panel.spec.tsandgrid-tab-wrap-rows.spec.tswith--workers=1— 52 passed, chromium and webkit.origin/main.Pins that moved
Three, all of them the 160 that no longer exists, each rewritten to state the absence of the jump rather than deleted:
packages/core/.../create-grid.test.ts— the engine's stored default for an undeclared column, 160 → 140.packages/grid-core/.../grid-ui-core.test.ts— same, now readingDEFAULT_COLUMN_WIDTH_PXfrom layout-core rather than re-typing the number, with the constant's value asserted separately.packages/react/.../column-auto-width.test.tsxandtool-panel.test.tsx— these used 140-vs-160 as the observable for auto-set membership, which the unification takes away. Membership is now made observable by writing a width that differs from any default (200), and the old jump's site asserts the freeze is a no-op pixel. A new test pins the double-click gesture.No e2e pin moved: the SP5 specs measure a declared-width 110 column against the 140 renderer default, and neither number changed.
🤖 Generated with Claude Code