Skip to content

Commit 8cf7e51

Browse files
committed
Merge remote-tracking branch 'origin/main' into claude/project-thread-xoyhoi
2 parents cfe7f4e + 208d3e2 commit 8cf7e51

169 files changed

Lines changed: 7404 additions & 759 deletions

File tree

Some content is hidden

Large Commits have some content hidden by default. Use the searchbox below for content that may be hidden.
Lines changed: 10 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,10 @@
1+
---
2+
'@objectstack/spec': patch
3+
---
4+
5+
fix(spec): a `currencyConfig.homeAmount` home or rate field that lacks `readonly: true` and declares an `expression` is told both faults in one refusal
6+
7+
Clause-②: no
8+
Metadata-protocol: none
9+
10+
An object whose `homeAmount.field` or `homeAmount.rateField` names a field that lacks `readonly: true` AND declares an `expression` was refused with only the `readonly` fault, so the author added `readonly: true` and met the `expression` fault on the next parse. The one issue at that pointer now names both faults and both remedies: add `readonly: true` to the field and delete its `expression`. The accept set is unchanged; a field with only one of the two faults is told exactly what it was told before.

‎.changeset/28893-flow-region-shape-at-validate-build.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -9,6 +9,6 @@ Metadata-protocol: none
99

1010
**BREAKING** accept-set narrowing at `os validate` and `os build` (`@objectstack/cli`): a stack carrying a flow whose `loop` body, `parallel` branches or `try_catch` region is not single-entry/single-exit or is cyclic, at any depth, or whose `parallel` has fewer than two branches, now fails validation and the build instead of passing and emitting an artifact.
1111

12-
Both commands now call `findFlowRegionRefusals`, which hands each flow (each `packages[]` entry's flows, or the stack's own) to the automation engine's own registration verdict: `FlowSchema`, then the spec's `validateControlFlow`. A refused flow prints `Flow '<name>' would be refused at registration: <where>: <the engine's message>` (`VALIDATION_ERROR`, exit 1), and the build writes nothing to `dist/`. Before, both passed it, and the engine refused it at boot with an `[Automation] failed to register flow` warning while the server ran without the flow. What the engine registers is unchanged. The fix is the engine's own sentence, e.g. add one node every exit of the region leads to.
12+
Both commands now hand each flow to the automation engine's own registration verdict: `FlowSchema`, then the spec's `validateControlFlow`. A refused flow is reported by the author-time rule `flow-region-shape-invalid` as `Flow '<name>' would be refused at registration: <where>: <the engine's message>` (exit 1), and the build writes nothing to `dist/`; `os lint` reports it too. Before, both passed it, and the engine refused it at boot with an `[Automation] failed to register flow` warning while the server ran without the flow. What the engine registers is unchanged. The fix is the engine's own sentence, e.g. add one node every exit of the region leads to.
1313

1414
<!-- adr-0087: not-required (no-migration-prescription) No key of any schema is added, removed, renamed or re-shaped, and no export or type changes: `os validate` and `os build` now run the check `registerFlow` already runs, with the same function (`validateControlFlow` from `@objectstack/spec/automation`) and its message. Every flow they now refuse was already refused at registration and never ran, so there is nothing for `objectstack migrate meta` to rewrite and no tombstone; the remedy is the engine's own sentence (close the region with one exit node, or give the parallel a second branch). -->
Lines changed: 20 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,20 @@
1+
---
2+
'@objectstack/lint': patch
3+
'@objectstack/cli': patch
4+
---
5+
6+
feat(lint): `os lint` refuses a flow whose structured region the automation engine refuses to register (`flow-region-shape-invalid`, ADR-0031)
7+
8+
Clause-②: no (narrowing)
9+
Metadata-protocol: none
10+
11+
**BREAKING** accept-set narrowing at `os lint` (`@objectstack/cli`, through the `@objectstack/lint` rule table): a stack carrying a flow whose `loop` body, `parallel` branches or `try_catch` region is not single-entry/single-exit or is cyclic, at any depth, or whose `parallel` has fewer than two branches, now fails `os lint` (exit 1) instead of passing. `os validate` and `os build` already refused it; their verdicts are unchanged.
12+
13+
- **One rule for all three commands.** The check is now the author-time rule `flow-region-shape-invalid` (`validateFlowRegionShape` and `FLOW_REGION_SHAPE_INVALID`, exported from `@objectstack/lint`). It calls the automation engine's own registration verdict, `FlowSchema` then the spec's `validateControlFlow`, and its message is the engine's refusal behind the flow's name. `os explain flow-region-shape-invalid` prints the full reasoning.
14+
- **What changes at `os validate` and `os build`.** They refuse the same flows as before, now from the rule table, so the refusal prints under `Author-time rules failed` and appears once. Under `--json` it is a rule finding (`rule`, `where`, `path`, `message`, `hint`), in `errors` for `os validate` and in `issues` for `os build`. A flow in a multi-package artifact is located at its place in the artifact's flows (`flows[N]`), with the flow's name in `where`.
15+
- **A flow the flow schema refuses** is reported by the same rule as one that could not be judged, with the schema's issues, never passed. `os validate` and `os build` refuse such a flow at the schema step already; `os lint`, which does not parse, now reports it here.
16+
- **Not judged:** a cycle in a flow's top-level graph, outside any region.
17+
18+
**The remedy.** Fix the flow the way the engine's sentence says: add one node every exit of the region leads to, remove the edge that loops back inside the region, or give the `parallel` a second branch.
19+
20+
<!-- adr-0087: not-required (no-migration-prescription) No key of any schema is added, removed, renamed or re-shaped: `os lint` now runs the check `registerFlow`, `os validate` and `os build` already run, with the same function (`validateControlFlow` from `@objectstack/spec/automation`) and its message. Every flow it now refuses was already refused by `os validate`, `os build` and at registration, and never ran, so there is nothing for `objectstack migrate meta` to rewrite and no tombstone. -->

‎.changeset/29059-flow-region-gate-hardening.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -7,5 +7,5 @@ fix(cli): the os validate / os build flow-region check never passes a flow it co
77
Clause-②: no
88
Metadata-protocol: none
99

10-
- **No silent pass.** The flow-region check `os validate` and `os build` run (the automation engine's registration verdict, ADR-0031) parses each flow with `FlowSchema` before it judges the regions. When that parse failed it used to skip the flow with no row, so the flow passed unjudged. It now refuses it with a `VALIDATION_ERROR` row at the flow's `path`, naming the flow and the schema's issues. No authored stack reaches this branch today: the stack schema has already parsed every flow the check reads, and `FlowSchema` accepts its own output. Nothing refused or accepted today changes.
10+
- **No silent pass.** The flow-region check `os validate` and `os build` run (the automation engine's registration verdict, ADR-0031) parses each flow with `FlowSchema` before it judges the regions. When that parse failed it used to skip the flow with no row, so the flow passed unjudged. It now refuses it with a `flow-region-shape-invalid` finding at the flow's `path`, naming the flow and the schema's issues. No authored stack reaches this branch today: the stack schema has already parsed every flow the check reads, and `FlowSchema` accepts its own output. Nothing refused or accepted today changes.
1111
- **Refused flows in the boot banner.** When the automation engine refuses a declared flow at registration, boot still continues and the per-flow reason is still under Boot diagnostics. The `os serve` / `os dev` banner now also prints one line beside the declared count, for example `⚠ Flows: 2 flow(s) declared, 1 refused at registration`, read from the engine's own record (`getFlowRegistration`). It prints even when every declared flow was refused and nothing registered. A boot with nothing refused prints no such line.
Lines changed: 16 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,16 @@
1+
---
2+
'@objectstack/spec': minor
3+
---
4+
5+
feat(spec): the push channel's protocol face, declared ahead of its producer: the widened device registration request, the stored device shape, optional `INotificationService` device members, the push delivery data with its same-server `route`, and the APNs / FCM transport configuration by `credentialRef` (ADR-0012 §6)
6+
7+
Clause-②: yes (widening)
8+
Metadata-protocol: added
9+
10+
Nothing here is served or read yet. Every new or touched schema carries a `[PLANNED — not yet served]` or `[PLANNED — not yet read]` marker in its description, and no route row is added: the device routes land in `DEFAULT_NOTIFICATION_ROUTES` together with their handlers, so the catalog never advertises a route that answers 404. Nothing a client or an operator writes today is refused.
11+
12+
- **`RegisterDeviceRequestSchema`** gains three optional members: `vendor` (`apns` | `fcm`; absent, the server derives it from `platform`, `ios` → `apns` and `android` → `fcm`, and a `web` registration names `apns` or `fcm`), `appId` and `locale` (a BCP 47 tag, the house `LocaleSchema`). `token`, `platform`, `deviceId` and `name` are unchanged, so every body a client sent before still parses as it did. The describes now say that re-registering a `deviceId` refreshes that device, and that a request without one gets a server-minted id the client keeps. A `vendor` outside `apns` / `fcm` is refused; since the member is new, no body sent before carries one. `RegisterDeviceResponseSchema` and `UnregisterDevice*` keep their members and gain only their description.
13+
- **`UserDeviceSchema`** (new, `system/push-device.zod.ts`) is the stored device: `deviceId`, `userId`, `vendor`, `token`, `platform`, `enabled`, and optional `appId`, `locale`, `name`, `invalidatedAt` and `lastSeenAt`. A device belongs to its user alone (no organization member), a token has one live row (the same user's registration without a `deviceId` refreshes the row that holds it, and a second user's registration takes it over), and a token the vendor reports dead keeps its row with `enabled: false` and `invalidatedAt`.
14+
- **`INotificationService`** gains four optional members: `registerDevice(userId, request)`, `unregisterDevice(userId, deviceId)`, `listDevices(userId)` and `invalidateDeviceToken(token)`. Their docs state the semantics an implementation owes. Absence means the capability is unavailable, so every existing implementation stays valid.
15+
- **`PushDeliveryDataSchema`** (new, `api/push-delivery.zod.ts`) is the data a push carries to the shell: `notificationId` and an optional `route`. `PushRouteSchema` admits only a same-server relative path (one leading `/`; no scheme, host, backslash, whitespace or control character; no dot segment and no percent-encoded `/`, `\`, `.` or `%` in its path). The route is derived, never authored: it is the notification's existing `actionUrl` (app-relative `/OBJECT/ID` or an `/apps/…` route) when that passes the schema, and absent otherwise. The shell resolves it on tap as the inbox does, never verbatim. FCM carries this data in `message.data`; APNs at the top level of the payload, beside `aps`. No `notify` key is added.
16+
- **`PushServiceConfigSchema`** (new, `system/push-config.zod.ts`, beside the email transport configuration): an `apns` block (token auth: `teamId`, `keyId`, `topic`, `environment: sandbox | production`, the `.p8` key by `credentialRef`) and an `fcm` block (HTTP v1: `projectId`, the service-account JSON by `credentialRef`). Secrets are references only (ADR-0143 D3). An inline secret is refused by name under the keys an operator would reach for, and refused as the content of a reference (a PEM block or a JSON document). `PushVendorSchema` names the two vendors. It is not a stack-definition key until its reader ships.

‎.changeset/29176-optional-unique-blank-null.md‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -10,6 +10,6 @@ Metadata-protocol: none
1010
A cleared text input is sent as `''`, and the write path stored it as sent. Every unique constraint the drivers build is NULL-distinct, so `null` never collides, but `''` is a value: the first record whose SKU (tax id, employee number, …) was cleared saved, and the second was refused "A record with this value already exists" although neither held a value. Measured on SQLite and the memory driver: two `''` rows collide, two `null` rows do not.
1111

1212
- **Now.** The engine's write door, which `insert`, `insertMany`, `update` (by id and by predicate) and the dry run (`validate`) apply before anything else reads the payload, stores exactly `''` as `null` on a column that is text-like (`text`, `textarea`, `email`, `url`, `phone`, `markdown`, `html`, `richtext`, `code`, `color`, `signature`, `qrcode`), not `required`, and covered by a unique constraint: the field's own `unique`, or membership in a declared `indexes[]` entry carrying `unique`. Every REST, batch, upsert and import door reaches the engine through those methods.
13-
- **Unchanged.** A `required` field given `''` is refused `required` exactly as before, also when it declares a `defaultValue`. A real duplicate is still `409 UNIQUE_VIOLATION`. Whitespace is a value and is stored as sent. `password` and `secret` keep their `''` refusal. A column no unique constraint covers, and `select` / `lookup` columns, keep storing `''`.
13+
- **Unchanged.** A `required` field given `''` is refused `required` exactly as before, also when it declares a `defaultValue`. A real duplicate is still `409 UNIQUE_VIOLATION`. Whitespace is a value and is stored as sent. `password` and `secret` keep their `''` refusal. A text column no unique constraint covers keeps storing `''`. (Optional single-value `select` / `lookup` columns store a cleared `''` as `null` too, unique or not: a separate change in this release.)
1414
- **Defaults.** An optional unique field that declares a `defaultValue` now takes it on insert when sent `''`, exactly as when sent `null`.
1515
- **Rows already stored.** Existing `''` values are not migrated. They keep matching a `field = ''` filter; a newly cleared row matches `$null: true` instead. The `is_empty` filter (`$empty`) matches both, as before. A newly cleared row no longer collides with a stored `''`.
Lines changed: 20 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,20 @@
1+
---
2+
'@objectstack/spec': minor
3+
'@object-ui/types': minor
4+
---
5+
6+
fix(spec)!: a dashboard global filter keyed `dateRange` beside a declared `dateRange` is refused at every door that parses it
7+
8+
Clause-②: no (narrowing)
9+
Metadata-protocol: narrowed
10+
Skills-docs: none — `skills/objectstack-ui/rules/dashboards.md` teaches `dateRange` as the reserved name of the built-in date range and no file under `skills/**` keys a `globalFilters` entry `dateRange`; every taught use stays true. `content/docs/ui/dashboards.mdx` gains one sentence naming the refusal.
11+
12+
**BREAKING** for a dashboard that declares `dateRange` and also keys a `globalFilters` entry `dateRange`.
13+
14+
- **Before.** The built-in date range is a filter the dashboard declares under the reserved key `dateRange`. A `globalFilters` entry keyed `dateRange` (its `name`, or its `field` when it has no `name`) beside it parsed, and the dashboard save door (`PUT /api/v1/meta/dashboard/X`) answered 200 in draft and in publish. The console builds the date range first and lets the later definition win, so the board drew the entry's control in place of the declared date range and every widget binding `dateRange` was filtered by the entry's field. The only trace was a browser console warning.
15+
- **Now.** `DashboardSchema` refuses the entry, so every door that parses a dashboard refuses it: `defineStack`, `os validate`, `os build` and the dashboard save door (422 `INVALID_METADATA`). One issue per such entry, at its `name` (or its `field`, when it has no `name`): ``globalFilters[i] is keyed 'dateRange', the reserved name of the built-in date range this dashboard declares in `dateRange`. …``. The rule reads the sibling `dateRange`, so it sits on the dashboard object and is exported as `checkDashboardReservedDateRangeGlobalFilters`: a mirror that rebuilds the dashboard from `DashboardSchema.shape` drops it unless it chains the export, as `@object-ui/types`' `DashboardComponentSchema` now does, so the console's authoring door refuses the same dashboard. On `DashboardSchema` itself, zod 4 now throws on `.omit()` / `.pick()` / `.partial()` (an object carrying a refinement); no caller in this repository derives it that way.
16+
- **Not changed.** On a dashboard that does not declare `dateRange`, an entry keyed `dateRange` is an ordinary filter and stays legal. A named entry whose `field` is `dateRange` is keyed by its name and stays legal.
17+
- **The remedy.** Give the entry another `name` (and rename every widget `filterBindings.dateRange` key that means it), or delete it and keep the built-in `dateRange`.
18+
- **Stored rows.** A stored dashboard carrying the pair still loads, with `_diagnostics` reading `valid: false` and carrying the refusal, and its next save (draft or publish) is refused until it is fixed. Nothing rewrites it: whether the entry was meant as the date range or as a filter of its own only the author knows. The upgrade guide entry is `ui-dashboard-global-filter-reserved-date-range-refused`.
19+
20+
<!-- adr-0087: registered ui-dashboard-global-filter-reserved-date-range-refused -->
Lines changed: 20 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,20 @@
1+
---
2+
'@objectstack/objectql': patch
3+
'@objectstack/core': patch
4+
'@objectstack/metadata-protocol': patch
5+
---
6+
7+
fix(objectql): a cleared optional single-value `select` or `lookup` stores `null`, so a second record cleared on a unique one is no longer refused `409 UNIQUE_VIOLATION`
8+
9+
Clause-②: yes (widening)
10+
Metadata-protocol: none
11+
12+
A cleared `select`, `lookup` or single file field sent as `''` was stored as sent. `''` is neither an option code, a record id nor a file id, so it held no value, yet it was one to the store and to the `is_empty` filter: on a unique column the second record cleared was refused "A record with this value already exists", and on every column the cleared record did not match `is_empty` (`$empty`), whose ruled table reads only `null` as empty on these types.
13+
14+
- **Now.** The engine's write door (`insert`, `insertMany`, `update` by id and by predicate, and the `validate` dry run; every REST, batch, upsert and import door reaches it) stores exactly `''` as `null` on every optional column the `$empty` table puts on its `null_only` row whose value is one string: the option types (`select`, `radio`), the reference types (`lookup`, `master_detail`, `user`, `tree`) and the file references (`image`, `file`, `avatar`, `video`, `audio`). Unique or not. Object- and array-valued types (`json`, `composite`, `location`, …) and the engine-owned `formula` / `autonumber` keep what they store.
15+
- **Read shape.** Such a field cleared from now on reads back `null`, not `''`. It matches `is_empty` / `$empty: true`, `$null: true` and `field = null`, and no longer matches `field = ''`.
16+
- **Unchanged.** The `$empty` table itself. A `required` field given `''` is refused `required` as before, also when it declares a `defaultValue`. Whitespace is a value. A `multiple: true` option or reference column keeps what it stores. A real duplicate is still `409 UNIQUE_VIOLATION`.
17+
- **Defaults.** An optional field that declares a `defaultValue` now takes it on insert when sent `''`, exactly as when sent `null`.
18+
- **Rows already stored.** Existing `''` values are not migrated by this change; they keep answering as a value (`$empty: false`, `field = ''`) until a data step cleans them.
19+
- **Seed replay.** A seed that writes `''` on such a column, or on any column whose blank the door already stored as `null` (an optional unique text column, a number / boolean / date column), is compared in that stored form, so replaying it is no longer a change on every boot. Before, the replay compared the seed's `''` with the stored `null` and rewrote the row each time (bumping `updated_at`).
20+
- **New export.** `@objectstack/core` exports the rule itself, `blankAsNullColumns` and `blankStorageForm` (with `ENGINE_LIFECYCLE_COLUMNS`), which the write door and the seed loader both read.

0 commit comments

Comments
 (0)