mirror of
https://github.com/supabase/supabase.git
synced 2026-09-06 09:59:03 +08:00
create-pull-request/patch
10 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
Charis
|
fa5eb17277 |
feat(studio): discriminated snippet union + source-aware writes (#48313)
Stacked on #48305. ## What PR 3 of the stacked SQL-editor query-source series (Database vs Logs). Stacked on the PR 2 branch `charislam/log-sql-content-shape`. Turns `SnippetWithContent` into a discriminated union on `type` and makes all snippet writes source-aware: - `data/content/sql-folders-query.ts`: `SnippetWithContent` is now `{ type: 'sql'; content?: SqlSnippets.Content } | { type: 'log_sql'; content?: LogSqlSnippets.Content } | { type: 'report'; content?: never }`. `report` is kept (the content endpoints' wire type carries it) but has no SQL content — its body is `Dashboards.Content`, loaded through the separate `Content` union. - `setSql` brands per type (`untrustedLogSql` vs `untrustedSql`). - `buildUpsertPayload` persists `snippet.type` (no longer hardcoded `'sql'`). - `createSqlSnippetSkeletonV2({ source })` emits the matching type + content shape with the `as any` cast removed. - New `components/interfaces/SQLEditor/querySource.ts`: `SqlSnippetSource` + `getSnippetSource`. - `seedSnippet` test helper gains a `source` arg. - New `remapWireSnippet` boundary helper in `content-remap.ts` concentrates the single wire->domain assertion, so `content-id-query` / `content-upsert-mutation` call sites are cast-free (no `as unknown as`). - Collateral: query result types aligned to the union; `updateSnippet` no longer accepts `type` (source is immutable); db-only editor read paths narrow away `log_sql`. ## Why Impossible-states-impossible typing: a snippet's brand follows its content type, so logs SQL and database SQL can never cross execution paths. No behavior change for existing database snippets. ## Testing - \`pnpm typecheck\` — clean - \`pnpm --filter studio run lint:ratchet\` — no new warnings - \`pnpm test:studio\` (data/content, SQLEditor, state/sql-editor) — passing, including new tests for \`getSnippetSource\`, source-aware \`setSql\`, type-aware \`buildUpsertPayload\`, and both skeleton shapes. <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added source-aware creation for SQL editor snippets, including log-based SQL snippets. * Introduced backend source mapping so log snippets are treated as log_sql. * **Bug Fixes** * Improved SQL retrieval/prettification so log snippets no longer use the wrong fallback content. * Ensured log snippets are sanitized and preserve correct type, content, identifiers, and statuses during save/upsert flows. * **Tests** * Expanded unit and integration coverage for log snippet creation, source mapping, editing, prettification, and upsert payloads. <!-- end of auto-generated comment: release notes by coderabbit.ai --> |
||
|
Ayaan Gazali
|
5db1137c56 |
fix(sql-editor): guard removeFavorite against missing snippet like addFavorite (#48111)
## I have read the
[CONTRIBUTING.md](https://github.com/supabase/supabase/blob/master/CONTRIBUTING.md)
file.
YES
## What kind of change does this PR introduce?
Bug fix
## What is the current behavior?
Fixes #48110
In the SQL editor Valtio store, `removeFavorite` guards against a
missing snippet with `if (storeSnippet.snippet)`, which reads `.snippet`
off `undefined` and throws `TypeError: Cannot read properties of
undefined (reading 'snippet')` whenever the id is not loaded in
`sqlEditorState.snippets`. Its counterpart `addFavorite` guards
correctly with `if (storeSnippet)` and no-ops on the same input.
## What is the new behavior?
`removeFavorite` now uses the same `if (storeSnippet)` guard as
`addFavorite`, so un-favoriting an id that is not in the store is a safe
no-op instead of a crash. Behavior for loaded snippets is unchanged.
Since `StateSnippet.snippet` is a required field, the old check was
always true whenever `storeSnippet` existed, so the only real world
difference between the two guards was the crash on the missing case.
I also added a small vitest file covering both methods (favorite set
plus needsSaving queued for loaded snippets, no-op for missing ids). The
missing-id test for `removeFavorite` fails with the exact TypeError
above when run against the old guard, and passes with this fix.
## Additional context
Root cause: `apps/studio/state/sql-editor/sql-editor-state.ts` line 260
(compare `removeFavorite` at lines 258 to 264 with `addFavorite` at
lines 250 to 256).
Gates run locally on top of current master (
|
||
|
Charis
|
1987f19d0a |
feat(sql-editor): add manual save feature preview (#47745)
## What Adds an opt-in **SQL Editor manual save** feature preview that switches the SQL Editor from autosaving every edit to saving only on demand, and hardens the tab-close flow so unsaved edits are handled correctly. ## Changes **Feature preview** - New `sqlEditorManualSave` flag + `UI_PREVIEW_SQL_EDITOR_MANUAL_SAVE` local-storage toggle, wired into the Feature Preview modal with an explanatory panel. - `useIsSqlEditorManualSaveEnabled` gates behavior on both the flag and the user's preview opt-in. **Editor toolbar** - Save button (with `Cmd+S`) next to Run, plus an autosave status indicator showing dirty/saving/saved state and a shortcut to disable autosave (emits a `sql_editor_autosave_disable_clicked` telemetry event). **Discard on close** - Closing a snippet tab with unsaved edits prompts for confirmation and, on confirm, actually discards the local edits and evicts the cached server copy so the snippet reopens clean. **Decouple tab layout from SQL specifics** - Tabs store gains a generic per-type close-handler registry (`registerTabCloseHandler` / `getCloseConfirmation` / `closeTabs`). The SQL editor registers its discard + confirmation behavior from the save coordinator. - Low-level `removeTab`/`removeTabs` (rename/move re-keying, stale cleanup) intentionally do **not** trigger discard. - Adds `statusOnDiscard` lifecycle transition and `clearSnippetContent` store action. ## Testing - `pnpm --filter=studio typecheck` — clean. - Added unit tests for the close-handler registry (fires on single/multi close, skips re-keying/cleanup removals, respects tab type, selects confirmation copy, unregisters cleanly). <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * Added a SQL editor manual-save preview with a “Save” button and `Cmd+S`, plus a modal option to disable manual-save/preview. * Added “unsaved changes” tab status indication when manual-save is enabled. * Introduced tab-type-specific close confirmations (shown only when needed). * **Bug Fixes** * In manual-save mode, closing a SQL tab with unsaved edits now clears local snippet content and refreshes it on reopen. * **Tests** * Added coverage for tab close handlers and confirmation behavior. * **Chores** * Added a persisted setting allowlist entry and tracked autosave-disable clicks via telemetry. <!-- end of auto-generated comment: release notes by coderabbit.ai --> |
||
|
Charis
|
7c1ea30e43 |
refactor(studio): extract useSnippetEditor from MonacoEditor (#47500)
## What
PR 7 of the SQL editor state re-layering stack. Extracts the snippet
editing lifecycle out of `MonacoEditor` into a co-located
`useSnippetEditor` hook, and consolidates the edit debounce.
`useSnippetEditor` owns:
- creating the snippet in the store on first edit and routing to its URL
(replace vs push for a `?content=` deep link)
- writing changes back to the store via `setSql` (with
`wasNeverPersisted` → `shouldInvalidate`)
- seeding the editor from the `content` param
- the read-only determination (`canEditSnippet`)
`MonacoEditor` now consumes `{ snippet, disableEdit, handleEditorChange
}` and keeps only the editor shell + Monaco action wiring. It sheds the
`router`/`profile`/`project`/`params`/store/tabs hooks.
`handleEditorChange` was also flattened with an early return.
## Debounce consolidation
Previously there were **two 1s debounces in series**: `useSnippetEditor`
debounced editor changes before writing to the store, and the save
mechanism (`createSaveMechanism`) already debounces persistence. That
added latency (up to ~2s to save) and split the "when to persist" timing
policy across two layers — at odds with PR 5's design where the
scheduler/mechanism owns *when* and dirty state is meant to be
immediate.
This PR removes the editor-side debounce: edits write to the store
synchronously on every change, and the save mechanism's 1s debounce is
the sole throttle. Net effects:
- the store — and the snippet's dirty status — reflects the latest edit
immediately (correct for the future manual-save mode's Save button / nav
guard)
- save fires ~1s after the *last* keystroke instead of up to ~2s
- only the active snippet's own reactive consumers (a lightweight
sidebar item) re-render per keystroke; Monaco is uncontrolled
(`defaultValue`) so it is unaffected
Note: the double-debounce was legacy (the pre-refactor god store had the
same `useDebounce(value, 1000)` in MonacoEditor plus a debounced
module-load subscribe).
## Notes
- Behavior-preserving in outcome — autosave still lands ~1s after typing
stops, just with lower latency and immediate store consistency.
## Validation
- `pnpm --filter studio typecheck` ✅
- `pnpm exec vitest --run state/sql-editor/` ✅ (110 passed)
- lint ✅ (0 errors; no ratcheted-rule regressions)
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* SQL editor changes now apply immediately, with unsaved status
reflected as soon as you edit.
* The editor now keeps the latest snippet details available for saving,
improving reliability when using “Save Query.”
* **Bug Fixes**
* Improved handling for creating and opening snippets from shared links
or prefilled content.
* Fixed status updates so saved snippets correctly switch to unsaved
after edits.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
|
||
|
Charis
|
d153bab849 |
refactor(studio): extract SQL editor session store from god store (#47349)
## What PR 6 of the SQL editor state re-layering stack. Moves ephemeral, never-persisted SQL editor state out of the snippet/folder "god store". **Session store** — `state/sql-editor/sql-editor-session-state.ts` holds per-snippet, read-by-many session state: - query `results` - `explainResults` - the row `limit` …with their mutators (`addResult`/`addResultError`/`resetResult`, `addExplainResult`/`addExplainResultError`/`resetExplainResult`, `resetResults`, `setLimit`). `removeSnippet` drops a snippet's session entries via `clearForSnippet(id)`. **Diff-request slice** — `state/sql-editor/sql-editor-diff-request.ts`. The Assistant's "Insert code" / "Replace code" diff is *not* per-snippet session state: it's a transient, fire-and-forget command produced outside the editor (e.g. query blocks / assistant) and consumed exactly once by whichever editor is active. It's modeled as a consume-once request (`requestDiff` / `consumeDiffRequest`) rather than durable state — the editor drains it on apply, so a stale diff can't leak into a later editor or session. (Previously this was `diffContent` in the god store: never cleared and triggered by object-reference identity.) Consumers read session state from `useSqlEditorSessionSnapshot` and the diff channel from `useSqlEditorDiffRequestSnapshot`, keeping `useSqlEditorV2StateSnapshot` only for snippets/folders. ### Why not the TanStack Query cache for results/explain? Editor execution is a **mutation**, not a keyed query — `mutation.data` is per-hook-instance and not keyed by snippet id, and there's no caching value to capture (re-running SQL must return *fresh* data, never a cached result). `EXPLAIN ANALYZE` actually executes the statement, so a declarative/auto-refetching `useQuery` is semantically wrong. Results/explain are imperative mutation outputs, scoped to the session, read by several decoupled consumers keyed by snippet id — exactly what a small in-memory keyed store models honestly. ## Consumers migrated - `SQLEditor.tsx` — results/explain/limit reads + `addResult`/`addResultError`/`addExplainResult`/`addExplainResultError`/`setLimit`; diff-apply effect now drains a consume-once request - `UtilityPanel.tsx`, `UtilityTabResults.tsx`, `UtilityTabExplain.tsx`, `UtilityActions.tsx` - `QueryBlock/EditQueryButton.tsx` — produces via `requestDiff` ## Notes - Result/explain types are kept verbatim from the god store (pre-existing `any` row/error types come along unchanged; tightening them is out of scope for this move). - `ref()` on result rows is preserved to avoid Valtio proxying large row sets. ## Tests - `sql-editor-session-state.test.ts` — result/explain mutators, `resetResults`, `clearForSnippet`, `limit` - `sql-editor-diff-request.test.ts` — `requestDiff`, `consumeDiffRequest` (drain + queue-of-one) Validation: - `pnpm --filter studio typecheck` ✅ - `pnpm exec vitest --run state/sql-editor/` ✅ (110 passed) - lint ✅ (no new errors) <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **New Features** * SQL editor query results, EXPLAIN output, and the “Limit results to” setting now persist more reliably across a session. * AI-assisted SQL insert/replace actions now use a pending diff workflow to apply updates more consistently. * **Bug Fixes** * Results/EXPLAIN rendering and downloads stay in sync with the latest executed data. * Switching databases/snippets now clears the correct temporary results. * Diff application is more resilient when an editor is still loading, including empty-vs-non-empty editor cases. * **Tests** * Added coverage for the session and diff-request state logic. <!-- end of auto-generated comment: release notes by coderabbit.ai --> --------- Co-authored-by: Joshen Lim <joshenlimek@gmail.com> |
||
|
Charis
|
5cb81123ae |
refactor(studio): move SQL editor save trigger into a scheduler + provider (5/9) (#47316)
## What
PR 5 of a stacked refactor. Moves *when to save* out of a module-load
`subscribe` and into an injectable **scheduler** armed by a headless
**provider**, splits the save queue, and adds an unsaved-close warning.
### Scheduler (`sql-editor-save-scheduler.ts`)
`createSaveScheduler({ state, saveMechanism, notify, getSaveMode })`
owns the save *policy*:
- **auto** mode drains the dirty snippet queue as edits land; **manual**
mode (the seam for a future opt-in; defaults to `auto`) leaves snippets
queued until `requestSave`. Folder saves always drain.
- `start()` returns an unsubscribe; `requestSave(id)` is the
explicit-save entry.
### Provider (`sql-editor-save-coordinator.tsx`)
Headless `SqlEditorSaveCoordinatorProvider` instantiates the mechanism
(invalidation via the **React Query client from context**, not the
global `getQueryClient`) + scheduler, `start()`s it in an effect
(start/stop with the provider), and exposes `requestSave` via
`useSqlEditorSaveCoordinator()`. Mounted in `ProjectContext` (under the
app's QueryClientProvider). Cmd+S and the SavingIndicator Retry now go
through `requestSave`.
### Queue split
`needsSaving` (snippets) and `pendingFolderSaves` (folders) are separate
queues, drained independently — the old snippet-vs-folder `if/else` is
gone.
### Unsaved-close warning
A `beforeunload` guard triggers the browser's native "Leave site?"
prompt while any snippet's `status !== 'saved'` (failed / in-flight /
never-saved).
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Improved SQL editor saving with a centralized save flow, including
automatic/manual save handling and immediate “Save Query” requests.
* Added unsaved-change detection so the app can warn before closing or
reloading when edits are still pending.
* **Bug Fixes**
* Retry actions now use the updated save flow for more reliable
re-saving.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
|
||
|
Charis
|
16526bd6bf |
refactor(studio): extract SQL editor save mechanism + model folder lifecycle (4/9) (#47276)
## What
PR 4 of a stacked refactor of the SQL editor snippet/folder state. It
pulls the persistence logic out of the store into an injectable
mechanism, and replaces the folder `'new-folder'` id sentinel with an
explicit lifecycle — plus a concurrency bug fix that surfaced along the
way.
### Save mechanism (`sql-editor-save.ts`)
`createSaveMechanism({ state, upsertContent, createSQLSnippetFolder,
updateSQLSnippetFolder, invalidate, notify, debounceMs })` → `{
saveSnippet, createFolder, updateFolder }`. The store's subscribe now
dispatches to it; *when* to save still lives in the subscribe (the
scheduler/provider move is PR 5). Per-id debounce cache lives in the
factory closure (no module-global leak).
- **`saveSnippet`** reads the live store snippet, guards
`isLoadedSnippet` so a content-less snippet can **never PUT an empty
body** (directly unit-tested), then builds the payload + drives status
transitions + gated invalidation.
- **`toast` is injected** as a `Notifier` (new generic DI contract in
`lib/notifier.ts`) — the mechanism no longer imports sonner.
- **create vs rename are two named-arg functions**, not an `isNew`
branch; rollback is deterministic per operation instead of matching on
`error.message` text.
- **caught errors are `unknown`**, narrowed via the existing
`getErrorMessage` util with a generic fallback — no `any`.
### Folder lifecycle (replaces the `NEW_FOLDER_ID` sentinel)
- **`FolderStatus`** enum (`new_editing | new_saving | editing | saving
| idle`) collapses the persistence and progress axes into one enum —
same pattern as `SnippetStatus` — with `isNewFolder` / `isFolderEditing`
/ `isFolderSaving` predicates. Tagging a folder as new/persisted is now
an explicit field, not an id convention.
- New placeholders get a **unique local id** (`crypto.randomUUID`);
`NEW_FOLDER_ID` is deleted, which also lifts the accidental
one-unsaved-folder-at-a-time limit.
### Bug fix: folder-rename rollback race
The shared `lastUpdatedFolderName` field let two in-flight renames
clobber each other's rollback target (and a shared `finally` could wipe
it). Replaced by a **per-folder `previousName`** on
`StateSnippetFolder`, so concurrent renames of different folders are
isolated. A new test runs two failing renames concurrently and asserts
each restores its own previous name.
## Tests
`sql-editor-save.test.ts` (mechanism — fakes + fake timers, incl.
content-less no-PUT and concurrent-rename isolation) and
folder-lifecycle predicate tests. `pnpm --filter studio typecheck`
clean; 82 state/sql-editor unit tests pass.
<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## Summary by CodeRabbit
* **New Features**
* Improved SQL editor folder handling with clearer create, rename, and
save states.
* Added a more consistent notification flow for successful and failed
save actions.
* **Bug Fixes**
* Improved rollback handling when folder renames fail, helping restore
the previous name reliably.
* Updated save behavior to better protect against duplicate or
out-of-order updates.
<!-- end of auto-generated comment: release notes by coderabbit.ai -->
|
||
|
Charis
|
d5653f1f92 |
refactor(studio): unify snippet save + persistence into SnippetStatus (3/9) (#47251)
## What PR 3 of a stacked refactor of the SQL editor snippet state. Replaces the two overlapping pieces of snippet lifecycle state — the `savingStates` map (`IDLE|UPDATING|UPDATING_FAILED`) and the `isNotSavedInDatabaseYet` boolean — with a single `SnippetStatus` enum. ## Status is attached at the data layer (never absent) - `SnippetStatus` + `SnippetWithContent` now live in `data/content`. The snippet queries attach `status: 'saved'` via a typed `withSavedStatus()` helper, and `upsertContent` returns `SnippetWithContent` so move/rename responses carry status too. - A SQL-typed `getSqlSnippetById`/`useSqlSnippetByIdQuery` returns `SnippetWithContent` (the generic `useContentIdQuery` stays for Reports, which use it). `[id].tsx` loads content with **no casting**. - `'new'` is attached on local creation (`createSqlSnippetSkeletonV2`). ## Behavior Behavior-preserving for the existing auto-save flow (faithful mapping of both old fields, including the replication-lag swallow). One incidental fix: the read-only/saving indicator now also covers a brand-new snippet's first save (previously only re-saves of persisted snippets had distinct saving/failed states in some paths). ## Tests New `sql-editor-lifecycle.test.ts` (29 tests) covering every predicate and transition; existing rules tests updated. `pnpm --filter studio typecheck` clean; 52 state/sql-editor unit tests pass. <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit ## Release Notes * **Refactor** * Restructured SQL snippet persistence tracking, replacing boolean flags with a comprehensive status system for clearer visibility into save progress. * Enhanced saving indicator UI to reflect accurate snippet save states. * **Tests** * Added test coverage for snippet persistence state transitions and lifecycle scenarios. <!-- end of auto-generated comment: release notes by coderabbit.ai --> |
||
|
Charis
|
e1e2498db0 |
refactor(studio): extract SQL editor domain rules into pure module (2/9) (#47204)
## What PR 2 of a stacked refactor of the SQL editor snippet state. **Stacked on #47203 (PR 1)** — review/merge that first. Extracts scattered business rules + the upsert-payload builder into a new **pure** module `apps/studio/state/sql-editor/sql-editor-rules.ts` (no Valtio, React, toast, or runtime data-layer imports): - `canEditSnippet` — read-only rule (shared snippet you don't own), was inline in `MonacoEditor` `disableEdit` - `isSnippetOwner` — owner check, was inline in `ReadOnlyBadge` / `SavingIndicator` - `validateMoveToFolder` — 'shared snippet cannot be within a folder', was a buried `toast.error` - `buildUpsertPayload` — the PUT /content payload, was an inline object literal (all `??` defaults preserved) - `isLoadedSnippet` — type guard (see below) ## Bug fix: no more empty-content saves (and no non-null assertion) The old payload builder used `{ ...content!, content_id: id }`. Tracing that `!` upstream surfaced a real bug: **favoriting a snippet from the sidebar that had never been opened** enqueued a save with no loaded content, producing a PUT with an empty content body (rejected by API). The requirement that a persisted snippet has loaded content is now enforced **at the type level** rather than by a runtime assertion or comment: - `buildUpsertPayload` accepts only a `LoadedSnippet` (content non-nullable) — the `!` is gone. - the save subscriber crosses that boundary via the `isLoadedSnippet` type guard. - the sidebar favorite toggle loads content first (mirroring `onSelectDuplicate` / the share modals), narrowing the fetched union content to the SQL variant via its discriminant — **no type cast**. <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Bug Fixes** * Improved consistency in read-only behavior and ownership checks across the SQL editor by centralizing permission logic. * Fixed favorite toggle to ensure snippet content is fully loaded before persisting changes. * **Refactor** * Centralized SQL snippet permission rules and validation logic into a dedicated helper module. <!-- end of auto-generated comment: release notes by coderabbit.ai --> |
||
|
Charis
|
2b24065c7b |
refactor(studio): relocate SQL editor store into state/sql-editor/ with facade (1/9) (#47203)
## What PR 1 of a stacked refactor that re-layers the SQL editor snippet state (`apps/studio/state/sql-editor-v2.ts`). This first PR is a **pure structural move with zero behavior change** — no consumer files are touched. - Relocates the Valtio store body into `apps/studio/state/sql-editor/sql-editor-state.ts` - Extracts the type declarations into `apps/studio/state/sql-editor/types.ts` - Adds `apps/studio/state/sql-editor/index.ts` as the public surface - Keeps the old `apps/studio/state/sql-editor-v2.ts` path as a thin re-export **facade**, so all existing importers keep working unchanged ## How to read the diff `sql-editor/sql-editor-state.ts` (~507 lines) is the **verbatim relocation** of the former `sql-editor-v2.ts` body — not new code. Git does not show it as a rename because the old path is intentionally retained as the facade. The only genuinely new lines are `types.ts` (20), `index.ts` (8), and the facade itself (8). ## Why The store has accreted four tangled responsibilities (snippet/folder CRUD, query results, persistence, Assistant diff). The stack incrementally splits these into pure rules, a persistent store, a session store, and an injectable save mechanism whose trigger is a swappable policy (setting up a future auto→manual save migration). Each PR stays ≤300–400 non-test lines and behavior-preserving. ## Verification - `pnpm --filter studio typecheck` passes (only pre-existing unrelated module-resolution errors remain). - Lint passes (no new errors). - No consumer imports changed. <!-- This is an auto-generated comment: release notes by coderabbit.ai --> ## Summary by CodeRabbit * **Refactor** * Restructured SQL editor state management into a modular architecture with improved separation of concerns and enhanced code organization. <!-- end of auto-generated comment: release notes by coderabbit.ai --> |