You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on Nov 4, 2026. It is now read-only.
Repository navigation
Commit 0e2ddd4
Browse filesBrowse the repository at this point in the historyBrowse files
feat(types): NamedListView declares the 17 protocol members, each with a read point (objectui#8980) (#9534)
Fixes#8980
`NamedListView` (`@object-ui/types`) declared **none** of the 17 live
members the
protocol declares on the same surface. objectui was NARROWER than the
protocol on
all 17 — the direction the maintainer's standing principle forbids. This
declares
all 17 with their types taken from the protocol, and gives every one of
them a
read point in the same delivery.
Governing ruling: director-seat class-one adjudication, objectui#8980
comment
`5651818304`, 2026-09-13. Its four items are executed as written; none
is
re-litigated here.
## Premise, re-derived on this branch — the card still holds
Re-measured at `origin/main` `bbc9dc34e3`, non-shallow, against the
resolved
`@objectstack/spec@17.4.0`, with the instrument self-tested on a firing
positive
control (two real interfaces, 9 and 2 members, both verified against
source) and
a non-firing negative control (a nonexistent interface name, exit 3):
| reading | value |
| --- | --- |
| `ObjectListViewSchema` shape keys | **50** |
| `NamedListView` members, before | **47** |
| in protocol, absent from objectui | **22** |
| of those, protocol tombstones | **5** — `bordered` `performance`
`responsive` `striped` `virtualScroll` |
| ⇒ live protocol members objectui did not declare | **17** ✅ |
**The counting rule, published with the number** — the seat deliberately
certified no total, so this is mine and is stated so it can be
re-derived:
*top-level property signatures at brace-depth 1 of the `export interface
NamedListView` body, line and block comments stripped, `readonly` and
`?`
allowed; members of nested object-literal types are NOT counted.* Under
that rule
the interface reads 47 before and **64** after. The interface's own
census pin
(`object-view-unmirrored-keys-7779.test.ts`) walks the same population
with the
TypeScript parser and agrees.
**The tombstones are measured, not assumed.** Each refuses a plausible
value with
a message naming its 17.0.0 removal. ⚠️ Six other keys also refused my
first
probe values (`name` `pageName` on a 2-character minimum, `data`
`kanban` `gantt`
`timeline` `chart` `grouping` on missing required inner fields) — that
is the
PROBE's fault and not six more tombstones, the same self-correction the
earlier
seat recorded. The discriminator is the refusal MESSAGE, not the
refusal.
⚠️ The card's line citation had drifted as the seat warned:
`NamedListView` opens
at `objectql.ts:2088`, not `:2073`. Everything here binds to the SYMBOL.
## Ruled item 3 — `tabs`, measured before it was declared
The protocol spends the word `tabs` on two different keys, and this
establishes
which one the list shape carries:
- `ObjectListViewSchema` is `ListViewShapeSchema.omit({ userFilters
}).extend({ userFilters: ObjectUserFiltersSchema })`, so the **top-level
`tabs: z.array(ViewTabSchema)`** passes through untouched. That is the
key declared here.
- `userFilters.tabs` is the one `ObjectUserFiltersSchema` OMITS as
page-only, in its own words: *"an object view's tab bar is its
saved-view switcher (ViewTabBar), and a second one would collide."*
objectui#9517's landing on this same file does not change that reading:
it moved
the user-filter FIELD shape and declared the container a dialect, not
the
container's `tabs`. objectui's `userFilters` dialect keeps carrying its
own
legacy `tabs` separately, so the two keys now sit on one interface and
the
declaration says so.
## What each of the 17 reaches
`ObjectView` has three exits and the members do not all leave by the
same one.
| member | read point |
| --- | --- |
| `kanban` `calendar` `gallery` `timeline` `gantt` `map` |
`generateViewSchema` — the AUTHORED path — now resolves the protocol's
top-level block and merges the legacy `options.KIND` nesting under it,
so each reaches `ObjectKanban` / `ObjectCalendar` / `ObjectGallery` /
`ObjectTimeline` / `ObjectGantt` / `ObjectMap` |
| `chart` `tree` | the same branch, on the one route objectui#5321
leaves open to a named view (see the report below) |
| `grouping` `rowColor` | `ObjectGrid` on the authored grid path (it
already reads both), and `ListView` through the host delegation |
| `fieldOrder` `appearance` `userActions` | `ListView` through the host
delegation. `fieldOrder` is the live third key of the protocol's
`columns` times `hiddenFields` times `fieldOrder` composition —
objectstack#15184 **ruling B**, read on the card rather than assumed:
that card is OPEN, and its ruling B KEPT the key |
| `data` | `ListView` — and the `as any` cast the renderer reached this
key through is gone, which is objectui#7928's open half answered |
| `name` | the named-view tab strip, between `label` and the record key
|
| `tabs` `pageName` | ⚠️ no renderer behaviour — REPORTED, see below |
**The merge is per-KIND and per-KEY, not wholesale**, and identity is
preserved
when no canonical block is declared — which is the whole existing
population,
since these keys could not be authored before this change. So the
renderer
produces byte-identical nodes for every stored document, and there is an
absence-control case beside every fix arm asserting exactly that.
## Ruled item 2's report — three members measured, none invented, none
dropped
*"A member for which the renderer has no behaviour to attach is REPORTED
on this
card with the measurement."* Posted in full on objectui#8980.
- **`tabs`** — no reader on this surface. objectui's tab bar for an
object is the HOST-owned saved-view switcher (ADR-0053); the only `tabs`
read under `packages/plugin-list` is `UserFilters`' `config.tabs`, which
is the page-only key, not this one.
- **`pageName`** — `page` is not a member of `NamedListView['type']`
(seven values against the protocol's ten), so no authored named view can
select the branch this key configures.
- **`chart` / `tree`** — READ, and the pin proves it, but objectui#5321
ruled both view KINDS host-composition-only. A named view reaches them
only by declaring no `type` of its own while a host `views` entry
selects the kind. The read is real; the authored route to it is not.
All three stay declared. ⛔ No read point was invented to make the count
come out
even, and ⛔ no member was dropped to avoid the report.
## Types are taken from the protocol
Sixteen index this package's own spec-derived `list-view` node type
(`ListViewSchema`), whose members arrive from `SpecListViewSchema.shape`
by
reference — the derivation `NamedListView.userFilters` already used one
member
up, and the one that keeps a named view and the `list-view` node it is
relayed
INTO the same type for every key that crosses. `name` indexes the
protocol's own
published authored type (`ListView`, `@objectstack/spec/ui`) instead: on
the node
that spelling resolves through `BaseSchema.name`, the component-name
slot — a
different contract wearing the same word.
⚠️ **A first cut put that derivation in the zod mirror and was correctly
refused**
by `imported-defaults-8317.test.ts`: importing a spec schema into a
mirror is a
crossing of the objectui#8317 import boundary and must appear in that
file's
differential. The type-only import in `objectql.ts` crosses no boundary
and the
mirror is untouched.
## Two narrowings the protocol owns, found by tsc while writing the pins
Reported rather than worked around, because the declared face is now the
protocol's:
- **`gantt` requires `titleField`** alongside `startDateField` /
`endDateField`. Unlike the four blocks this package keeps a local
`.partial()` dialect for, `gantt` flows into the mirror straight from
`SpecListViewSchema.shape`, so the protocol's required bindings are
required here.
- **`chart` is the ADR-0021 dataset-bound shape only** (`dataset` +
`values` required). The renderer still carries a legacy inline-aggregate
branch below it (`xAxisField` / `valueField` / `aggregation`) that the
declared face cannot express. ⛔ Not widened locally — that legacy shape
still reaches the branch through the untyped `options.chart` bag, and
narrowing an authoring surface toward the protocol is the direction the
principle asks for.
## The census ledger is re-taken, not adjusted
`object-view-unmirrored-keys-7779.test.ts` pins `NamedListView`'s
population by
name and by count, and its own failure message names the four files that
carry
the figures together. All four are updated in this change:
**64 declared = 21 read + 43 unread** (was 47 = 6 + 41, plus one
undeclared cast
read). `tabs` and `pageName` are two of the 43, and the ledger's
docblock says
that is the ruled outcome rather than an oversight. The instrument
agreement is
re-pinned too: parser 64 = strict regex 64, loose regex 76 — still a gap
of
exactly 12, because every new member is a single-line type reference
that adds no
nested object literal.
⭐ One genuine ledger change beyond the numbers: the AST derivation and
the
`currentNamedViewConfig?.KEY` regex it replaced now disagree by one
NAME, not
just by sites — `name` is read only at the tab strip. The difference is
asserted
by name rather than smoothed over.
## Verification
Run from the repo root in a dedicated worktree, every heavy run through
the
shared verify lock.
- `pnpm exec vitest run packages/types/ packages/plugin-view/` — **229
files / 4707 tests passed**, lock VERDICT command-exit 0.
- `pnpm --filter @object-ui/types --filter @object-ui/plugin-view
type-check` — exit 0. It covers the new test file: its first run
reported the two fixture errors that produced the `gantt` / `chart`
findings above, so the coverage is a reading and not an assumption.
- `pnpm --filter '@object-ui/plugin-view^...' build` — exit 0, 16
projects. The `@object-ui/types` dependency closure is EMPTY (it depends
only on `@objectstack/spec` and `zod`), so that package's own closure
build is a no-op and is reported as such rather than as a pass.
- Gate families derived by hand from this repo's own `package.json` and
`.github/workflows/`, ⛔ not from the sibling repo's
`dispatch-gates.mjs`, which answers only about the tree it lives in:
`check:control-bytes` `check:new-line-citations` `check:spec-symbols`
`check:vi-mock-specifiers` `check:vi-mock-inherit`
`check:vi-mock-override-shape` `check:test-path-roots`
`check:changeset-claims` `check:comment-mask-corpus`
`check:unreferenced-sources` `check:lint-rule-coverage`
`check:handler-key-reads` `check:changeset-presence` — **all exit 0**.
- `check:eager-closure`, `check:sdui-registration-pins` and
`check:readme-exports` are **NOT MEASURED** here, ⛔ not red: each
refuses for a missing console/workspace build in its own words ("this is
a broken gauge, not a passing budget"; "the population COLLAPSED — this
run proves nothing", citing 21 unbuilt packages). CI builds and owns
them.
- `eslint` on the six changed files: **0 errors**, 115 pre-existing
warnings, exit 0. Package-scoped `lint` for both affected packages: **0
errors** (286 + 348 warnings, none new-rule). The narrowing is a
measurement, with its three pieces: the covered population is `eslint .`
minus this config's own `ignores`; the file count is 6, read from
`--format json`; and the invariance holds because the config configures
NO type-aware linting (zero `project` / `projectService` entries), so
this diff cannot move the verdict on any file it does not touch.
Measured on final head `0e69156`.
- Control bytes: `grep -naP` over every changed file finds none, with a
firing positive control on a synthetic file.
## Reverse verification — direction predicted before the run
The renderer alone was restored to the merge-base on disk (blob
`b6c0cc566b` to `008aa55149`, mutation proven by hash comparison AND by
five
marker counts going 4/1/1/1/1 to 0/0/0/0/0), the new pin re-run, then
restored
under a `trap ... EXIT INT TERM` and proven back by blob hash and an
empty
`git diff HEAD` on that path.
Predicted: every FIX arm RED, every CONTROL arm GREEN. Measured: **13
failed, 7
passed** — the 13 are exactly the fix arms (six view kinds, the
chart/tree route,
the canonical-wins and per-key-merge arms, the grid pair, the delegation
six, the
`userActions` merge, the `name` fallback) and the 7 that stayed green
are the
controls (the legacy-nesting control, both absence controls, the `label`
precedence and record-key controls, the `tabs`/`pageName` absence pin,
and the
`data` behavioural arm — which passes on both trees because the old cast
relayed
the same value, exactly as that case's own comment says, its source half
being
pinned in the types census instead).
⚠️ No `dist` preflight: this repo's root vitest config aliases every
`@object-ui/*` specifier to that package's `src`, so the suite runs from
source
and an on-disk proof is the right evidence here.
## Scope
- **`Clause-②: no`**, measured: the runtime accept set does not move.
`ObjectViewSchema.listViews` is UNMIRRORED on the zod side and rides
`BaseSchema`'s passthrough, so no parse changes; the TS face gains
optional members, which widens rather than narrows. The `--pair` exit
code is reported to the PM seat.
- ⛔ No label added or removed, `needs:contract-review` included — labels
are the PM seat's alone.
- The 19 legacy spellings `NamedListView` declares beyond the protocol
are objectui#7924's remedy and are untouched (ruled item 4). That card
shares this file and is serial behind this one.
- ⛔ `packages/app-shell` is NOT touched, though it is where `name` is
written today (`mergeViewsIntoObjects` stamps `name: key` on every
composed entry and primary-view promotion matches on it — recorded at
the declaration as the evidence that this key has been travelling on
this surface undeclared). It is outside the claim's declared file
surface.
- ⛔ Opened as a draft. Not undrafted, approved, enqueued or merged.
---
_Generated by [Claude
Code](https://claude.ai/code/session_01L5xpA5q533BgTTNADibEFt)_
---------
Co-authored-by: Claude <noreply@anthropic.com>
`userActions` — all seventeen were declared by `ObjectListViewSchema`
15
+
(`@objectstack/spec/ui`), the declared value type of both `ViewSchema.listViews`
16
+
and `ObjectSchema.listViews`, and by none of them by `@object-ui/types`. An
17
+
author writing the shape the protocol teaches got a view that rendered as though
18
+
nothing had been configured, with no diagnostic anywhere.
19
+
20
+
**Types are taken from the protocol, not restated.** Sixteen index this
21
+
package's own spec-derived `list-view` node type (`ListViewSchema`), whose
22
+
members arrive from `SpecListViewSchema.shape` by reference — the derivation
23
+
`NamedListView.userFilters` already used, and the one that keeps a named view
24
+
and the `list-view` node it is relayed into the same type for every key that
25
+
crosses. `name` indexes the protocol's own published authored type (`ListView`,
26
+
`@objectstack/spec/ui`) directly, because on the node that spelling resolves
27
+
through `BaseSchema.name` — the component-name slot, a different contract
28
+
wearing the same word.
29
+
30
+
**What an author can now do that silently did nothing before:**
31
+
32
+
| key | reaches |
33
+
| --- | --- |
34
+
|`kanban``calendar``gallery``timeline``gantt``map`| the renderer `ObjectView` dispatches to for that `type` — `ObjectKanban`, `ObjectCalendar`, `ObjectGallery`, `ObjectTimeline`, `ObjectGantt`, `ObjectMap` — at the protocol's own top level. The legacy `options.<kind>` nesting keeps working; a canonical block wins key-by-key over it, so a partially declared block does not blank its legacy neighbour |
35
+
|`chart``tree`| the same dispatch, on the one route objectui#5321 leaves open to a named view (it declares no `type`, a host `views` entry selects the kind) |
36
+
|`grouping``rowColor`|`ObjectGrid` on the authored grid path, and `ListView` through the host delegation |
37
+
|`fieldOrder``appearance``userActions`|`ListView` through the host delegation. `fieldOrder` is the live third key of the protocol's `columns` × `hiddenFields` × `fieldOrder` composition (objectstack#15184 ruling B) |
38
+
|`data`|`ListView` — and the `as any` cast the renderer used to reach this key through is gone, which answers objectui#7928's open half |
39
+
|`name`| the named-view tab strip, between `label` and the record key |
40
+
41
+
**Two members are declared and NOT read, and that is the ruled outcome rather
42
+
than an oversight** — the ruling requires a member with no renderer behaviour to
43
+
be reported with its measurement, never silently dropped from the type.
44
+
`tabs` on the list shape is the `ViewTabSchema[]` multi-tab definition list (not
45
+
`userFilters.tabs`, which the protocol omits from an object view as page-only),
46
+
and objectui's tab bar for an object is the host-owned saved-view switcher
47
+
(ADR-0053). `pageName` configures the protocol's `type: 'page'` branch, and
48
+
`page` is not a member of `NamedListView['type']`. Both measurements are reported
49
+
on objectui#8980.
50
+
51
+
**Compatibility.** Every addition is an optional member, and every read is a new
52
+
rung whose source could not be authored before this change — so on existing
53
+
documents the renderer produces the same nodes it produced before, proven by an
54
+
absence-control case beside each fix. The 19 legacy spellings `NamedListView`
55
+
declares beyond the protocol are objectui#7924's remedy and are untouched.
0 commit comments