Repository navigation
[Deferred by ruling] Unify flow field-expression assignment onto the CEL engine — the B half of the #11060 ruling, awaiting maintainer appetite #11182
Description
Ruled: 5805777944 · letter D · 2026-09-24T01:26Z
Blocked-by: #19938
Filed by the triage seat as the ruled follow-up of #11060 (maintainer 2026-08-23, live PM chat, verbatim: 「其他同意你的意见」— adopting "A now + B follow-up on hold pending appetite").
What this card is
#11060's ruled option A ships the six-function table (round/floor/ceil/abs/min/max, semantics mirrored from CEL) to unblock the quote flow. That is deliberately a bridge: it leaves the platform with TWO expression vocabularies — condition slots speak CEL (35 functions), assignment slots speak the template dialect (6). This card holds the end-state question: migrate assignment expressions onto the CEL engine so there is one dialect, one function set, declared = enforced.
Why it is on hold rather than queued
The migration cost is real and unmeasured — the {var} template dialect's compatibility surface across existing flows was the recorded confidence gap in #11060's analysis. Scheduling it is an appetite call the maintainer explicitly deferred.
(retired line, batch #218 item 1 D: Restart-when closed #11060 — fired on 2026-09-23)
(When A lands and #11060 closes, the unlock scan returns this card to the decision inbox for an appetite ruling — with A's landed shape and the by-then-measurable {var} dependency surface as inputs. ⛔ Do not implement from this card without that second ruling.)
Refs: #11060 (ruling + analysis) · hotcrm#1206 (the measured business pull).
Activity
objectstack-fleet commentedon Sep 23, 2026
Claim:
domain:servicesPM seat · 2026-09-23T13:19ZSeat: domain:services#1
Session: session_01AhQASwqJr2Z7XfGWUdvnbF
Branch: claude/issue-11182-assignment-expressions-cel
Thread-read: 5792259887
Clause-②: no
Declared file face: the flow assignment-expression evaluation in
packages/services/service-automation(the{var}template dialect and #11060's six-function table), plus the measurement across this repo's flows,examples/, and HotCRM.Clause-② reading, with its citation:⚠️ If the work turns out to need a
SKILL.md:524negative boundary. Which evaluator a runtime slot uses is runtime behaviour, not a published contract face.packages/specchange, it stops and the change goes to its own sub-card. The same applies to any change on the CEL engine side (packages/formula), as the ruling says. ⛔ This seat does not cross lanes.Take basis:
priority:p1, from the maintainer ruling on 2026-09-23 (5792259887), 「11182 当前版本就优先处理,统一为cel」. It is neither a tooling card nor a contract-hygiene card, so North Star clause 3 does not reach it.Round shape, following the ruling's three points:
{var}spelling today, with sites and counts.NOW()/TODAY()— every other identifier is rewritten tonull, so a computed money value can never be rounded to its field's declaredscale#11060 A's shape. It never falls into a lenient default.⛔ Zero
packages/specin this lane.Generated by Claude Code
objectstack-fleet commentedon Sep 23, 2026
os-dev-report
{
"issue": 11182,
"status": "needs_decision",
"branch": "claude/issue-11182-assignment-expressions-cel",
"pr": null,
"session": "session_01AhQASwqJr2Z7XfGWUdvnbF",
"premise_still_valid": true,
"summary": "MEASURE-FIRST round, stopped before any behaviour change per the ruling's point 2 because the measurement found spellings that break or change meaning. Measured surface (TS-AST walk of every inline flow node of type assignment / create_record / update_record, slots assignments.* + legacy shapes + fields.* incl. same-file spreads; objectstack @ 48c91e9, HotCRM @ 087b7c5dc4, read via anonymous clone): HotCRM src has 92 template-bearing assignment-slot sites, objectstack examples+src 21, and 106 of those 113 authored sites sit in create_record/update_record fields.. That slot has no CEL path today: the spec ledger declares only assignment.assignments. as a value-role envelope slot. Every measured spelling was evaluated through the shipped template dialect AND its best CEL conversion through the shipped CEL engine over the same variables (13 SAME / 12 DIFF). The silent-meaning break is the #11060 business pull itself. HotCRM quote-generation's 2 money sites
round(... * 100) / 100become CEL int/int division: 123.46 turns into 123 and 1111.1 into 1111, with no error. The 14 NOW()/TODAY()-family sites change type from ISO string to Timestamp, and CEL has no string(timestamp) overload to recover the string. Every path spelling flips from undefined to a fault when its key is absent. Two structural blockers sit outside this lane: the ADR-0087 conversion registry is packages/spec/src/conversions/registry.ts, and fields.* would need a spec ledger slot. Nothing was implemented, no PR was opened, and the pushed branch equals origin/main.","tests": "No code changed, so no gate is owed; the deliverable is the measurement. (1) Dependency closure built under the lock:
pnpm --workspace-concurrency=2 --filter '@objectstack/service-automation^...' buildprintedos-verify-lock: VERDICT command-exit 0 · held the lock 302s. (2) Site census is a TypeScript-AST walk overgit ls-filesin both repos. Per-spelling counts (authored src/examples, tests separate): HotCRM src has bare {x} 31, dotted {a.b} 35, {$User.Id} 5, {NOW()} 7, {TODAY()} 3, {TODAY() + N} 3, {TODAY() + var} 1, {round(... * 100) / 100} 2 (quote-generation.flow.ts:107-108), {list.0} 1, embedded text 4, plus plain literals 38 and number/bool/null/const 22. HotCRM test has bare 1, dotted 3, NOW() 1, list.0 1, embedded 1, and 4 refusal probes (LOWER / TRIM / toLowerCase x2 in account-name-normalized-match.test.ts). objectstack examples+src has bare 9, dotted 10, {$User.Id} 1 (app-todo task.flow.ts:457), {$error.message} 1 (app-showcase flows/index.ts:1225), plus plain literals 12, array 1, bool 1. objectstack tests have bare 7, dotted 16, embedded 4, round-scale-2 1 (flow-field-expression-scale.integration.test.ts:146), list.0 1, and one CEL envelope. Zero sites anywhere: {$User.Email}, TODAY() - N in assignment slots, floor/ceil/abs/min/max, and ternary/compare/logical/string-literal expressions. (3) Template-vs-CEL probe (tsx over service-automation src interpolate() and @objectstack/formula ExpressionEngine, same scope shape as engine.ts celScope; scratch files deleted, not committed). SAME: resolved bare/dotted/list.0-to-[0]/$error-to-vars["$error"]/embedded-string/embedded-number/null value/plain literal as a quoted CEL string/round with an integer-valued amount (126000, which is why the #11060 oracle passes). DIFF:round(oppRecord.amount * (discount / 100) * 100) / 100with amount 1234.56 and discount 10 gives template 123.46 against CEL 123, and total_price gives 1111.1 against 1111;10 / 4is 2 in CEL;{NOW()}is a string in the template dialect butnow()is a Date;{TODAY()}is the string '2026-09-23' against today() returning a Date at midnight UTC;string(now())has no overload for google.protobuf.Timestamp;{missing}resolves to undefined againstUnknown variable: missing;{oppRecord.owner}with the key absent resolves to undefined againstNo such key: owner;{userList.0}on an empty list resolves to undefined against index out of bounds; embedded{o.name}null renders '' againstno matching overload for string(null);current_user.idin the flow CEL scope givesUnknown variable: current_userbecause celScope binds no user; plainconvertedas CEL source givesUnknown variable: converted.","mcp_calls": "0 — no MCP GitHub tools used (reads went through REST curl; HotCRM through an anonymous git clone after add_repo returned read_available)",
"api_writes": "1 — POST /repos//issues/11182/comments (this os-dev-report, via scripts/pm/post-stamped.mjs); plus git push of the empty branch (not a REST write)",
"open_questions": [
{
"question": "Q1 (the gate question): measured spellings do NOT all convert losslessly, so how is compatibility for the existing template-dialect assignment values decided? The two HotCRM quote money sites change value silently under a verbatim CEL conversion, the 14 date-macro sites change type, and all path spellings flip from silent-undefined to a fault on absent keys.",
"options": [
"A: Additive unification. CEL envelopes become accepted in create_record/update_record fields.* too (the same #15137 envelope shape assignments.* already has), and the template dialect stays as-is for 17.x. A lint finding nudges authors to envelopes, and nothing existing changes meaning. Cost: spec sub-card (ledger slot + AssignmentValueSchema reach) plus an executor change in service-automation (M). Two dialects keep coexisting in value slots until v18 retires the template form. Anti-AI-error: CEL is available everywhere, but the old spelling is still accepted.",
"B: Converting unification now, in the ADR-0087 shape. A spec-side conversion rewrites every {…} value to a CEL envelope with semantic-preserving rewrites: integer literals become doubles (100 to 100.0), paths are wrapped for absent-key tolerance (has()/coalesce), list.0 becomes [0], and $User.Id becomes current_user.id with the flow CEL scope binding the user. Date macros need a packages/formula sub-card first (a string rendering of Timestamp, or date fields accepting Timestamp), otherwise 14 HotCRM sites change type. Cost: L, and three lanes (spec conversion + ledger, formula stdlib, service-automation executor/scope). The rewrites are semantic transforms, not renames, so they stretch ADR-0087 D2's 'losslessly mappable only' scope. The converted sources are also uglier than what an author would write.",
"C: Loud-refusal unification. The template dialect in value slots is refused at registration with a per-spelling prescription (the #11060 refusal shape), and authors migrate by hand. Cost: breaks all 92 HotCRM and 21 objectstack authored sites at once on a 17.x minor, which is a breaking change on a shipped slot mid-major and contradicts the ADR-0087 intent.",
"D: Hybrid. Do A now. Also fix the false 'round(x * 100) / 100 is the CEL idiom' guidance (see finding), and file the formula date-string sub-card. Put B's conversion plus C's refusal of the template form on the v18 train, where ADR-0087 majors are the conversion window."
],
"recommendation": "D. Four axes: (1) business pull: the quote flow works today on the template dialect, and A gives it CEL in fields.* with nothing broken; B's verbatim path would silently corrupt exactly that flow's money. (2) Long-term: one dialect is still the end state, and D reaches it at the only point where ADR-0087 sanctions a shape break (a major). The semantic rewrites B needs (int-to-double literals, absent-key wrapping) go beyond D2's rename-class scope. (3) Anti-AI-error: with A, AI authors can write CEL in every value slot today, and the template form can gain a lint 'prefer envelope' finding. C's 17.x mass refusal would red every existing flow. (4) Focus: A is one spec ledger sub-card plus one executor change. B needs three lanes, including packages/formula. Confidence gap: HotCRM is the only out-of-tree consumer measured, and third-party stacks are unmeasurable from here."
},
{
"question": "Q2 (boundary, independent of Q1): every option except C needs packages/spec changes (the conversion registry lives at packages/spec/src/conversions/registry.ts, and fields.* needs a value-role slot in the expression ledger resolveFlowNodeExpressions / builtin-node-config.zod.ts). Options B and D also need a packages/formula change for Timestamp-to-string. Both are out of this lane by the claim. Should the seat file the spec sub-card (and the formula one) as Blocked-by children of #11182?",
"options": [
"A: File 'spec: declare create_record/update_record fields.* as a value-role CEL envelope slot' (plus the ADR-0087 conversion entry if Q1=B) and 'formula: render a Timestamp to an ISO / date string in CEL' as sub-issues of #11182 in the objectstack repo, with #11182 Blocked-by both",
"B: Treat Q1's answer first and file only what it needs"
],
"recommendation": "B. File after Q1 is ruled. Under D only the spec ledger sub-card is needed now, and the formula one rides v18."
}
],
"out_of_scope_findings": [
"class: a · The shipped guidance tells authors that
round(x * 100) / 100is 'the CEL idiom' / 'identical in CEL' / 'matching the CEL authoring pattern', but under the CEL engine it is int/int division and truncates silently. Probe:round(12345.6) / 100gives 123 andround(12345.6) / 100.0gives 123.46. Copying the example into any CEL slot (assignment envelope, formula field) silently loses the cents. Sites: packages/services/service-automation/src/builtin/template.ts:83 (docblock) and :111 (the arity-refusal message text, shipped runtime), content/docs/automation/flows.mdx:1750, skills/objectstack-automation/SKILL.md:236, service-automation CHANGELOG 8155855 entry. In the template dialect itself the pattern is correct; only the 'same in CEL' claim is false. Seam: runtime:template.ts requireArity message → renderer: none; consumer: authors following the hint into CEL slots · dedupe words: 'round integer division CEL', 'round(x * 100) / 100', 'CEL idiom scale 2 truncation', '100.0 double literal'","carrier: the implementation round of #11182 (whichever Q1 option) will touch template.ts and flows.mdx and is the natural place to fix the wording to
/ 100.0, so the text is not repeated in two places · noted in the open question, not filed separately unless the seat prefers a standalone card"]
}
Generated by Claude Code
objectstack-fleet commentedon Sep 23, 2026
测量完成,兼容性需要维护者裁定 ·
pm:dispatched→needs-user-decision· 2026-09-23T13:35Zdomain:servicesPM 席(席位帖 #6021)。本轮按认领评论 5795584973 的形状跑到第 2 步就停下了:测出有写法会被破坏或改变含义。按裁定第 2 点,这种情况不能在代码里悄悄选。**没有改任何代码,没有开 PR。**测量全文见 os-dev 报告 5795796051。测到了什么(objectstack
48c91e9e46· HotCRM087b7c5dc4)create_record/update_record的fields.*,而这个槽位今天根本没有 CEL 路径:spec 的表达式台账只把assignment.assignments.*声明成 value-role 信封槽。quote-generation.flow.ts:107-108的round(... * 100) / 100,原样转成 CEL 后是整数除法,123.46 → 123,1111.1 → 1111,不报错。这恰好就是 Flow field expressions can call no function butNOW()/TODAY()— every other identifier is rewritten tonull, so a computed money value can never be rounded to its field's declaredscale#11060 当初要解决的那个业务流程。NOW()/TODAY()在模板方言里是 ISO 字符串,在 CEL 里是 Timestamp,而且 CEL 没有string(timestamp)可以转回去。undefined,CEL 直接报错。packages/spec/src/conversions/registry.ts;fields.*要有 CEL 路径,需要在 spec 台账里加一个槽位。两处都不归本车道。选项
fields.*也接受 CEL 信封(与assignments.*今天的形状相同),模板方言在 17.x 保持原样,再加一条 lint 提示作者改用信封。**没有任何现有写法改变含义。**代价:一张 spec 台账子卡,加一次引擎改动(M)。两套方言会并存到 v18。{…},并做语义保持的改写:100→100.0、缺失键包一层has()、$User.Id→current_user.id。日期类要先在packages/formula立子卡。代价:L 级,跨三条车道。这些改写是语义变换,不只是改名,已经超出 ADR-0087 D2「只做可无损映射的改写」的范围。附带发现(class a,不论选哪个选项都要修)
现行的文字指引说
round(x * 100) / 100是「CEL 的惯用写法」「在 CEL 里一样」,这是错的。在 CEL 里这是整数除法,会静默丢掉分;应当写成/ 100.0。出现的位置:service-automation/src/builtin/template.ts:83(docblock)与:111(运行时报错文案,已发布)、content/docs/automation/flows.mdx:1750、skills/objectstack-automation/SKILL.md:236。任何一个选项的实现轮都会碰到template.ts和flows.mdx,顺手修掉就行,所以本席不单独立卡。下一步
维护者裁定后,本席按裁定需要的范围立 spec 子卡(以及可能要的 formula 子卡),都挂
Blocked-by,然后把本卡放回队列。Generated by Claude Code
objectstack-fleet commentedon Sep 24, 2026
Ruling: batch #218 item 1 · letter D · maintainer 「11182 D 其他同意」 2026-09-24T01:23Z
Director seat, summon #28 (续) (
session_01GLdRPcbaCBQCTvVmU6YEUY). Presented in this seat's chat with recommendation E (A + C in the current line); the maintainer chose D — A now, the wrong guidance fixed now, B's conversion and C's refusal on the v18 train. E is ⛔ not taken; recorded as the maintainer's letter.Governing text: the maintainer's 2026-09-23 word 「11182 当前版本就优先处理,统一为cel」 (5792259887), re-scoped by this letter — in 17.x 「统一」 means CEL usable in every value slot, one dialect arrives with v18; ADR-0087 D2 (only lossless mappings ride an automatic conversion — the 12 DIFF spellings the round measured are not lossless, which keeps B off 17.x); #11060 ruling A (the loud refusal shape). Readings: the seat's measurement 5795796051 / 5795832575 (113 sites, 106 in
fields.*, 13 SAME / 12 DIFF); the three wrong-guidance sites re-read onorigin/mainfdeeea0—packages/services/service-automation/src/builtin/template.ts:83and:111,content/docs/automation/flows.mdx:1757,skills/objectstack-automation/SKILL.md:236.Ruled — D
create_record/update_recordfields.*accept the CEL expression envelope, the same shapeassignment.assignments.*accepts today. Contract first: the spec expression ledger declaresfields.*as a value-role envelope slot — filed as spec: declarecreate_record/update_recordfields.*as a value-role expression slot in the expression ledger, so the CEL envelope is accepted there (the contract half of #11182 ruling D) #19938 (domain:spec· p1 ·pm:queue,Clause-②: yes); the engine half (this card) lands after it. The{var}template dialect keeps its 17.x meaning unchanged; an author-time lint hint (warning) nudges toward the envelope. ⛔ No spelling changes meaning in 17.x.round(x * 100) / 100is integer division in CEL — the text says so and prescribes/ 100.0— attemplate.ts:83/:111andflows.mdx:1757;skills/objectstack-automation/SKILL.md:236is Tier H and rides its own draft for the maintainer's hand if the same PR cannot carry it./ 100.0,has()guards, a string form forNOW()/TODAY()) and whatever of B is lossless ride the v18 train — filed as [v18] retire the{var}template dialect in flow assignment slots: refuse at registration with per-spelling remedies (the C half of #11182 ruling D, on the v18 train) #19939 (domain:services·pm:on-hold·Restart-when: closed objectstack-ai/objectstack#11182), so the timing question returns to the box when A lands, ⛔ never dropped silently. HotCRM's 92 sites migrate against that card, behind its pin.Execution
needs-user-decision→pm:blockedin this stroke,Blocked-by: objectstack-ai/objectstack#19938on the body;domain:services· p1 ·Clause-②: no(runtime behaviour; the spec half carries its own declaration). The unlock scan returns this card when #19938 closes.objectstack-fleet commentedon Sep 27, 2026
Claim: PM loop round 1 — member of the #19938 joint dispatch (ruling B)
Session:
session_01CiCTczDo7tGhafXjf61dUJAccount:
os-sales(the seat's linked user asGET /useranswers it; the card's assignee)Branch:
claude/issue-19938-fields-value-slotWorktree:
objectstack-issue-19938Domain:
domain:services(claimed bydomain:specseat 4 on the maintainer's ruling below)Seat:
domain:spec#4(seat post #18917)Ruling-ref: 5816929495
File surface: this card's engine half only, on the #19938 branch:
packages/services/service-automation/src/builtin/crud-nodes.ts(evaluate a CEL value envelope increate_record/update_recordfields.*, as the assignment executor does forassignments.*), the value-envelope refusal wording inservice-automation/src/engine.ts,builtin/logic-nodes.tscomments,builtin/config-expression-ledger.test.ts,builtin/template.ts(ruling D point 2:round(x * 100) / 100is integer division in CEL — say so and prescribe/ 100.0), and the author-time lint hint inpackages/lint/src/validate-expressions.ts. The full joint surface is theClaim:on #19938 (5852841056). ⛔ Notskills/objectstack-automation/SKILL.md(Tier H; its own draft per ruling D point 2). (stop on breach; explain in the report)Container & model:
M/L,mode:subagent,model: default judgment tier, one dev for both halvesClause-②: no
Thread-read: 5805777944
Serial constraints cleared: read at 2026-09-27T05:59Z. The joint census is on #19938's dispatch. No open PR touches
packages/services/service-automation/**orvalidate-expressions.ts; draft PR #20170 touches only the front matter ofcontent/docs/automation/flows.mdx.Why the spec seat holds a
domain:servicescard. The maintainer answered Q1 on #19938 with B (triage record5816929495, maintainer verbatim 「19938 同意」): this card's engine commit and #19938's spec commit land on ONE branch in ONE PR, spec first, closing both cards together. That record states the ruling is the authorisation for the services-lane step.Blocked-by: objectstack-ai/objectstack#19938in the body is dissolved by the fold, because both halves ship in the same merge. State:pm:blocked→pm:dispatchedin this stroke. Ruling D (5805777944) still governs this card's content: A now;{var}keeps its 17.x meaning; a lint hint (warning) nudges toward the envelope; C and the lossless part of B ride the v18 carrier #19939.objectstack-fleet commentedon Sep 27, 2026
domain:specseat 4 (session_01CiCTczDo7tGhafXjf61dUJ), 2026-09-27T13:39Z: this card's engine half lands with #19938's spec half in PR #20205, per the maintainer's Q1 = B (5816929495on #19938). The review of record is the ACCEPT5856348996on #19938. At-tier records:5854753486(PASS on the whole change) and5856341116(delta PASS at the landing head56f77479e). The PR is ready with auto-merge armed, and this card closes on the merge through itsFixes #11182line.objectstack-fleet commentedon Sep 27, 2026
Landed — PR #20205 →
e46218674b(2026-09-27T15:52Z)domain:specseat 4 (session_01CiCTczDo7tGhafXjf61dUJ). PR #20205 merged through the merge queue on 2026-09-27T15:47:38Z, and both #19938 and #11182 closedcompletedthrough itsFixeslines. The readings that confirm the landing:e46218674bhas one parent (2bcd5cfeb7), is an ancestor oforigin/main, and the queue branch for the PR is gone.git patch-id --stableequals the reviewed diff's (9f9b9f52b952on both; delta record5857140844at68123f428, the third at-tier record on this change after5854753486and5856341116).FlowValueSlotSchemaappears in 5 files underpackages/spec/src/automation/at the merge, and in 0 on the parent.The path to it: the PR was dequeued once on a
dropped-refinements.baseline.jsonconflict, and two patch rounds re-mergedmainand recounted the ledger. Every non-generated file stayed byte-identical to the reviewed change.pm:dispatchedand the assignee come off both cards in the same act.Downstream: #19961 (the next card on
flow-node-expression-paths.ts) and #20168 (onschemaless-node-config.zod.ts) were serial behind this landing and are now takeable.skills/objectstack-automation/SKILL.md:236(/ 100) stays with its own Tier H draft under ruling D point 2.feat(spec,automation): create_record / update_record fields.* accept …
feat(spec)!: flow value slots refuse the `{…}` template dialect, nami…
docs(skills): objectstack-automation teaches the CEL value envelope i…
feat(formula): isoDate(t) / isoDatetime(t), the string form of a CEL …