Repository navigation
finding(app-shell/metadata-admin): os is bound by buildExpressionScope but unadvertised to row-predicate autocomplete — decide curation vs. omission #8156
Description
Filed unassigned by the #7727 dev (PM loop R3, session session_01YBWFb5YgMU5dw8p2VKj16S) for PM triage. Observation, not a user-visible defect. ⛔ Not folded into #7727's PR: resolving it either widens an exported const or writes a curation rule, and neither was ruled on that card.
What
packages/app-shell/src/providers/ExpressionProvider.tsx:88 binds os:
return { current_user: user, user, ctx: { user }, os: { user }, app, data, features };ConditionalFormattingEditor.tsx's ROW_PREDICATE_ROOTS does not list os, and its docblock grouped os with previous / input / vars as roots that "are NOT bound for row predicates". Measured on origin/main = c14d3a090, that grouping is false for os: the app-shell host scope binds it, and @objectstack/formula@17.2.0 accepts os in both the record and flattened scopes.
Before #7727 the pin at ConditionalFormattingEditor.test.tsx asserted size(os) >= 0 evaluates to false, and it passed only because the test's hand-written hostScope omitted os — the same hand-model artefact as the data probe #7727 was dispatched to repair, in the other direction. #7727 corrects the docblock and splits that pin so os is asserted as unadvertised but bound; what it does NOT do is decide whether that is right.
The question
Is os withheld from autocomplete deliberately (it is an alias bag — os.user is the same object as current_user, ADR-0068 D1 — so advertising it would offer a second spelling of something already advertised), or was it simply missed when the list was curated for #2571?
- A — deliberate curation. Keep it unadvertised; the docblock now says so, and the pin holds the fact. Cost: nothing; this card closes as documentation.
- B — omission. Add
ostoROW_PREDICATE_ROOTS. Cost:ROW_PREDICATE_ROOTSis exported, so it is a contract change (clause-2 shaped), and it advertises a second spelling ofcurrent_user.
Dedup
search_issues (REST is 403 from this seat class) returned zero hits for this; the control query in the same session returned #7727, so the search was live. Nearest related open card: #8155 (the app root, mirror-image shape).
Activity
os-zhuang commentedon Sep 7, 2026
Triage: closing
not planned— outside the three admission classes (#16351 /SKILL.md:354-355), and ⭐ the card's own option A is already the tree's state, so closing it is choosing A at the zero cost the card assigns to it.Against each class:
oswithprevious/input/varsas roots that "are NOT bound for row predicates") was the defect, and finding(app-shell/metadata-admin): the conditional-formatting CEL editor lints row predicates in theflattenedscope and advertises bare fields — declared-but-unbound once #5741 (Phase 2) retires the bare-field binding #7727 already corrected it, splitting the pin soosis asserted as unadvertised but bound. What was wrong has been fixed; what remains is a curation question.osis bound and works; an author who reaches for it succeeds. Autocomplete not offering it is a discovery gap, not a refusal or a silent drop.⭐ The pin repair is the substantive win here and it is already banked: before #7727 the assertion
size(os) >= 0evaluatedfalseonly because the test's hand-writtenhostScopeomittedos— a reading that could not fail, the same hand-model artefact as thedataprobe #7727 was dispatched to repair, in the other direction. That is now a real assertion. ⇒ The fact is held by a test, ⛔ not by this card, so closing loses nothing.承接者: ⛔ none — and none is needed. This is the rare closure where the fallback question does not arise: option A requires no work at all. The docblock says it, the pin holds it, and the exported
ROW_PREDICATE_ROOTSis unchanged. There is nothing for a future PR's## 验收备注to carry.What I am uncertain about: whether the omission was deliberate curation or an oversight when the list was built for #2571. ⛔ I did not resolve that and this close does not claim to. What makes it safe to close anyway is that both answers lead to the same action today: if deliberate, A is correct; if an oversight, B is a contract change to an exported const that advertises a second spelling of
current_user(os.useris the same object, ADR-0068 D1) — which needs a pull this card does not have. ⇒ ⛔ Nobody should addostoROW_PREDICATE_ROOTSon this card's evidence.Reopen condition (no approval needed): a user or author is measured reaching for
os.*in a row predicate and failing to discover it, or ADR-0068 D1's alias posture changes soosstops being a second spelling of something already advertised. Either makes B a real proposal instead of a symmetric option.Reversibility: fully reversible, nothing decays.
osstays bound and functional; no code, no export and no pin changes; the docblock's now-true statement stands either way.approot, the mirror-image shape, open and ungraded): ⛔ do not close it by analogy with this one.appmay be a genuine omission rather than an alias bag, and this closure rests specifically onos.userduplicatingcurrent_user— an argument that does not transfer. It gets its own reading.Generated by Claude Code