test(plugin-view,plugin-list,plugin-calendar): drop the retired target key from three tree fixtures (objectui#8031) (#9818)

Fixes #8031

Test-only. Three `type: 'tree'` fixture fields drop the retired
snake_case target key —
in `plugin-view`'s expand-gate suite, `plugin-list`'s relational-sort
suite and
`plugin-calendar`'s expand-gate suite — plus one empty-frontmatter
changeset declaring
that nothing releases. No renderer, no `detectParentField`, no sibling
field touched.

## Every suite's verdict is unchanged, measured both ways

Same command before and after the edit, from the repo root, through the
shared
serialisation lock:

```
pnpm exec vitest run --reporter=verbose \
  packages/plugin-view/src/__tests__/ObjectView.expandGate.test.tsx \
  packages/plugin-list/src/__tests__/ListView.relationalSort.test.tsx \
  packages/plugin-calendar/src/__tests__/ObjectCalendar.expandGate-6453.test.tsx
```

| reading | before | after |
|---|---|---|
| summary | `Test Files 3 passed (3)` · `Tests 23 passed (23)` |
identical |
| per file | plugin-view 8 · plugin-list 6 · plugin-calendar 9 |
identical |
| test-name set | — | `diff` of the two sorted verbose logs is EMPTY |

`turbo run type-check` for the three packages (each runs `tsc --noEmit`
and then
`tsc -p tsconfig.test.json`, so the edited test programs are in it): 19
tasks
successful, 19 total. `check-changeset-presence`, `check-control-bytes`,
`check-new-cross-file-line-citations`, `check-test-path-roots` and
`check-governed-queue-guard --test` all exit 0; the last prints
`NOT GOVERNED — 4 path(s) checked`.

## Why DROP and not rename — re-measured on the installed copy

`@objectstack/spec@17.4.0`, `FieldSchema.safeParse`, with a nonsense key
as the
negative control:

| input on a `tree` field | verdict |
|---|---|
| no target key (what this PR leaves) | ACCEPT |
| retired snake_case target key | REFUSE — `unrecognized_keys` **plus**
a rename hint |
| nonsense key `zz_nonsense_key` | REFUSE — `unrecognized_keys`, **no**
hint |
| declared `reference`, this object's own name | ACCEPT |
| declared `reference`, another object's name | ACCEPT |

The hint is attached to a refusal, not to an acceptance — the card's
two-directional
measurement reproduces at 17.4.0. On the read side `detectParentField`
has read only
the declared spelling since objectui#6837 half 2, and the list sort
picker's
object-def branch reads only that spelling too. So the key was refused
by the schema
and read by nobody, and a refused key annotates nothing: renaming it
would turn a
refusal into an ACCEPTED self-annotation these fixtures never made.

## Three premises corrected — read this before reusing the card's
framing

1. ⚠️ **These three are NOT foreign-target.** Each container declares an
object
identity and the retired key already named exactly it: `TASK_SCHEMA` is
`task` and
the field named `task`; the relational-sort suite's object def is
`contacts` and
the field named `contacts`; `VISIT_SCHEMA` is `visit` and the field
named `visit`.
What made these three refused is the SPELLING, not the target. The
card's title,
its "whose target names another object" framing, and its disposition
clause "none
of these three containers declares an object identity a self-reference
could name"
are all false of these three. The disposition survives the correction;
the
   classification does not.
2. ⚠️ **The seven instances are therefore not indistinguishable.**
objectui#7839's
four were target-foreign or identity-less (three of them declare no
object at all;
the fourth declares `deal` and named `deals`). These three are
target-self. What
unites all seven is an inert, parse-refused target key on a `tree`
field, which is
   what the shared disposition rests on.
3. ⚠️ **"A shape the spec refuses at parse" is true upstream, not of the
installed
pin.** Row five above: a declared `reference` naming another object is
ACCEPTED at
17.4.0. `refuseForeignTreeReference` landed on objectstack `main` after
this pin
   was cut, exactly as objectui#7839's own suite docblock records.

A fourth correction is the card's line number for the `plugin-view`
fixture, which the
dispatching seat had already caught: every address was re-derived on
today's `main`.

## The shape scan was re-run, and it found a fourth emit — reported, not
touched

Re-run as a shape scan rather than a key grep: every `type: 'tree'` emit
in the tree
(single- and double-quoted, template literals, JSON spellings, every
file extension,
constants holding the string, and field factories that take the type as
a parameter).
31 literal emits. Of those, three are these fixtures; two carry a
declared target that
names the DECLARING object (legal self-references in `plugin-tree`); the
deliberate
foreign arms in objectui#7839's own suite are that suite's subject and
must stay; the
rest carry no target or are view entries rather than field defs.

⭐ One emit is invisible to BOTH sweeps and is a live instance of the
card's own
lesson:
`packages/app-shell/src/utils/expandableFamily.identity-5874.test.ts`
builds
its fields through a factory that takes the type as an argument and
attaches
`reference: 'accounts'` to every one of them, including the `tree`
member
`parent_node` on an object named `deal`. A grep for the key sees a
helper with no
tree in sight; a grep for the emit sees no literal type at all. It is
NOT in this
PR's scope and NOT the same disposition — the target there is
load-bearing
(`expect(out.referenceTo).toBe('accounts')` reads it), so dropping it
would go red.
It is handed back for triage in the report, with dedupe words.

## The pin — a truthful "not soundly pinnable here"

The population-shaped gate the card asks for ("no `type: 'tree'` emit
carries a target
naming another object") is writable but not soundly: I have a
counterexample in hand.
A gate keyed on literal emits would go green on the app-shell factory
above, and that
green would read as "the population is closed" — the exact false closure
the card's
own lesson names, repeated one level up. Judging "names another object"
also needs the
declaring object's identity, which in a hand-built fixture is an
arbitrary neighbouring
constant a scanner would have to guess at. The sound route is a
type-level one — type
these fixtures against the spec's own field type, so the compiler
refuses the retired
key and the tree arm — and that is a separate change, not something to
smuggle in
under a test-only card. ⛔ A pin that re-asserted these suites' existing
assertions
would measure nothing, so none was added.

## Acceptance notes

- **To file (handed back, not filed here — this seat does not open
cards):** the
app-shell factory emit above. Dedupe words:
`expandableFamily.identity-5874`,
`field('tree')` factory, foreign tree target, `resolveActionParams`
`referenceTo`,
  tree-emit enumeration blind spot.
- **Noted, not filed:** at the parse door the retired spelling is
refused on a
`lookup` too (measured), so "dead on a `tree` field only" is about the
READERS, not
about `FieldSchema` — several app-shell readers still read the snake
spelling for
the lookup family, which is why the sibling lines are legitimately
load-bearing and
untouched. Carrier: the next card that moves the lookup-family spelling
reads this;
  no PR is pending on it today.
- `Clause-②: no` — three hand-built test fixtures. No symbol exported,
removed,
renamed or retyped, no published `.d.ts` member and no accept set moves.
Measured,
not inferred from a diff grep: the three files sit under
`src/__tests__/`, which
every published tsconfig excludes from the emitting program (enforced by
`pnpm check:published-tsconfig-exclude`); each container is a
module-local `const`;
all three packages ship `dist` only; the diff adds zero `export` lines
and deletes
one property from an object literal whose own annotation already
declares that key
  optional.

---
_Generated by [Claude
Code](https://claude.ai/code/session_012EpHzwH4wTy5sd7ibkD2yq)_

Co-authored-by: Claude <noreply@anthropic.com>

Author: os-justin

.changeset/8031-tree-fixture-retired-target-key.md

@@ -0,0 +1,45 @@
+---
+---
+
+Test-only. Drops the retired snake_case target key from the three `type: 'tree'`
+fixture fields that objectui#7839's `reference:`-keyed sweep could not see — in
+`plugin-view`'s expand-gate suite, `plugin-list`'s relational-sort suite and
+`plugin-calendar`'s expand-gate suite. Nothing published moves, so this declares
+no release: the three files sit under `src/__tests__/`, which every published
+tsconfig excludes from the emitting program (`pnpm check:published-tsconfig-exclude`),
+each container is a module-local `const`, and all three packages ship `dist` only.
+
+`FieldSchema` refuses that spelling BY NAME. Re-measured on the installed
+`@objectstack/spec@17.4.0` with a nonsense key as the negative control: the
+retired spelling draws `unrecognized_keys` **plus** a rename hint, the nonsense
+key draws the same refusal with **no** hint — so the hint is attached to a
+refusal, not to an acceptance. On a `tree` field the renderer side has read only
+the declared `reference` since objectui#6837 half 2, and the list sort picker's
+object-def branch reads only that spelling too. So the key was refused by the
+schema and read by nobody; every rule under test reads the declared TYPE. All
+three suites' verdicts are unchanged — 3 files, 23 tests, identical test-name
+sets before and after.
+
+⚠️ One correction to the framing these three were filed under, measured on the
+fixtures themselves rather than inherited: they are NOT foreign-target. Each
+container declares an object identity and the retired key already named exactly
+it — `TASK_SCHEMA` is `task` and the field named `task`, the `objectDef` under
+test in the relational-sort suite is `contacts` and the field named `contacts`,
+`VISIT_SCHEMA` is `visit` and the field named `visit`. What made these three
+refused is the SPELLING, not the target; the foreign-target reading belongs to
+objectui#7839's four, whose containers either declare no identity at all or
+named a different one.
+
+The key is still DROPPED rather than renamed, and the correction is why that
+matters: a refused key annotates nothing, so renaming it would turn a refusal
+into an ACCEPTED self-annotation these fixtures never made. Measured on the same
+installed copy, a `tree` field with no target and one that self-annotates are
+both accepted, and a FOREIGN target is accepted too — `refuseForeignTreeReference`
+landed upstream after this pin, so the canonical spelling would be a live claim
+none of these suites needs.
+
+`Clause-②: no` — three hand-built test fixtures. No symbol exported, removed,
+renamed or retyped; no published `.d.ts` member and no accept set moves. The
+diff deletes one property from a module-local object literal and adds comments;
+the containers' own type annotation already declares that key optional and is
+untouched, as is every sibling field that legitimately carries it.

packages/plugin-calendar/src/__tests__/ObjectCalendar.expandGate-6453.test.tsx

@@ -141,7 +141,13 @@ const VISIT_FIELDS: Record<string, { type: string; label: string; reference_to?:
   amount: { type: 'currency', label: 'Amount' },
   owner: { type: 'user', label: 'Owner' },
   account: { type: 'lookup', label: 'Account', reference_to: 'account' },
-  parent_visit: { type: 'tree', label: 'Parent', reference_to: 'visit' },
+  // No target key. It carried the retired snake_case spelling, which
+  // `FieldSchema` refuses BY NAME, so the line annotated nothing and the
+  // `$expand` expectation is derived from the declared TYPE. On a `tree` the
+  // target is optional; this map's object is `visit` and the value named it,
+  // so renaming would turn a refused key into an accepted self-annotation
+  // this fixture never made (objectui#8031).
+  parent_visit: { type: 'tree', label: 'Parent' },
   line_item: { type: 'master_detail', label: 'Line item', reference_to: 'line_item' },
 };
 

packages/plugin-list/src/__tests__/ListView.relationalSort.test.tsx

@@ -36,7 +36,14 @@ const objectDef = {
     owner: { type: 'lookup', label: 'Owner', reference_to: 'sys_user' },
     account: { type: 'master_detail', label: 'Account', reference_to: 'accounts' },
     assignee: { type: 'user', label: 'Assignee', reference_to: 'sys_user' },
-    parent: { type: 'tree', label: 'Parent', reference_to: 'contacts' },
+    // No target key. It carried the retired snake_case spelling, which
+    // `FieldSchema` refuses BY NAME, and the sort picker's object-def branch
+    // reads only the declared `reference` (objectui#6837 half 2) — so the
+    // line annotated nothing and was read by nothing. On a `tree` the target
+    // is optional; this object is `contacts` and the value named it, so
+    // renaming would turn a refused key into an accepted self-annotation this
+    // fixture never made (objectui#8031).
+    parent: { type: 'tree', label: 'Parent' },
     // The picker's SECOND withholding rule (#4243, and the shared judgement as
     // of objectui#3950): computed on read, no column to order by. Present in
     // this fixture so the exact option lists below pin its absence rather than

packages/plugin-view/src/__tests__/ObjectView.expandGate.test.tsx

@@ -122,7 +122,14 @@ const TASK_FIELDS: Record<string, { type: string; label: string; reference_to?:
   due_date: { type: 'date', label: 'Due' },
   owner: { type: 'user', label: 'Owner' },
   account: { type: 'lookup', label: 'Account', reference_to: 'account' },
-  parent_task: { type: 'tree', label: 'Parent', reference_to: 'task' },
+  // No target key. It carried the retired snake_case spelling, which
+  // `FieldSchema` refuses BY NAME (measured: `unrecognized_keys` plus a
+  // rename hint, where a nonsense key draws the same refusal without one), so
+  // the line annotated nothing and the `$expand` expectation is derived from
+  // the declared TYPE. On a `tree` the target is optional; this map's object
+  // is `task` and the value named it, so renaming would turn a refused key
+  // into an accepted self-annotation this fixture never made (objectui#8031).
+  parent_task: { type: 'tree', label: 'Parent' },
   line_item: { type: 'master_detail', label: 'Line item', reference_to: 'line_item' },
 };