Repository navigation
New datasource: the file path field holds a real :memory: value, and Test connection passes on the broken path it produces #30331
Description
epic:#26676#1 · zhuangjianguo · session_cdb7a717def74e719988f089fbaa18ff
Filing gate: ③ the maintainer's direct order, verbatim: 「你发现的问题都立卡，安全问题总专用的安全密卡通道，需要进一步分析的进一步测试，还是没有把握的用决策卡。」 (2026-11-05). Found in a real-user browser walkthrough of Setup on main 04f36b63f8, with a test app and a local test deployment.
Business reason. An administrator connecting a company database relies on Test connection. A pass here means the connection is right, so a false pass sends them on to a connection with none of their data.
Reader. The UI lane dev for ui/packages/app-shell/src/views/metadata-admin/datasource/DatasourceResourcePage.tsx, and the server lane for the test in packages/services/service-datasource.
Duplicates. Searched datasource filename memory prefilled (2 hits) and datasource test connection memory false (134 hits). They are QA run anchors and seat cards; none has this subject.
Measured
- Open 新建数据源 and leave the SQLite driver selected. The Filename box looks like a hint, and it already holds
:memory:as its value. - Click in and type a path. The box then reads
:memory:/private/tmp/…/parts.db. POST /api/v1/datasources/testwith that value answersok: true.
A correct path that does not exist is refused properly, so the check works in that case.
I did not save a datasource with the broken value, so what it would then read is not measured.
Expected
- The default is a placeholder, not a value, so typing replaces nothing by accident.
- A filename that is neither exactly
:memory:nor a usable path is refused by the test and by save, with a message that says why.
Activity
objectstack-fleet commentedon Nov 5, 2026
triage#1 · os-zhuang · session_01AavokzJ5DndAwitDXvKy4U
Path: none
Triage: first-touch audit · p2 ·
domain:server·pm:queue· 2026-11-05T07:11Z⛔ Not a claim.
:memory:value, and Test connection passes on the broken path it produces. A false success, so p2. Platform first:service-datasource's test-connection must check the real target. The dialog default is thedomain:uihalf.Generated by Claude Code
objectstack-fleet commentedon Nov 5, 2026
domain:server#4 · os-bill · session_01Utp6CZMkHvuDeVYgTD73qF
Claim: PM loop round 1
Session:
session_01Utp6CZMkHvuDeVYgTD73qF(seat) · dev card threadsession_01Utp6Cb83ajcseWprEF9C9gAccount:
os-billBranch:
claude/project-thread-541rod(the card thread's designated push branch, the recorded platform exception toclaude/issue-<n>-<slug>)Worktree:
objectstack-issue-30331Domain:
domain:serverSeat:
domain:server#4File surface:
packages/services/service-datasource/**(the test-connection door and the datasource save, at the one place both pass through), their tests, one changeset. ⛔packages/spec/**,docs/adr/**,.claude/**, repo-rootAGENTS.md,ui/**(the dialog's:memory:default is the ui residue, recorded at landing),skills/**(stop on breach; explain in the report)Container & model:
S,mode:thread,model: default judgment tier (opus) — dispatch-gates --tier: no path-derived mandate,effort: defaultClause-②: yes (narrowing) — a SQLite filename that is neither exactly
:memory:nor a usable path, accepted today by the test and save doors, is refused (the dev confirms the arm)Metadata-protocol: none
Responsibility:
the datasource test door answers ok: true for a SQLite filename such as ':memory:/private/tmp/…/parts.db' | none: the administrator is told the connection works (card, measured on main 04f36b63f8) | any administrator connecting a file database through Setup, whose field is prefilled with ':memory:'Thread-read: 6652062778
Premise: holds
Serial constraints cleared: no card in flight touches
service-datasource(seats 3 and 5 confirmed).Dispatch order:
/mnt/project-files/server-seat4/dispatch-30331.md(project shared folder). Ruling section: the card's Expected ("refused by the test and by save, with a message that says why") and triage audit 6652062778 ("Platform first"). PRRefs #30331 (server half), no closing keyword. A contract review atCONTRACT_REVIEW_TIERis owed before queue (narrowing).Posted 2026-11-05T07:50Z.
objectstack-fleet commentedon Nov 5, 2026
domain:server#4 · os-bill · session_01Utp6CZMkHvuDeVYgTD73qF
os-dev-report
Status: done (server half). Draft PR #30395 carries
Refs #30331 (server half), headcca010a64. The card stays open for the ui half.:memory:/…passed. Every layer reads a name that starts with:as fileless (driver-sqlresolveSqliteAbsentFileTarget/sqliteFilename, the wasm connection'sisEphemeral), so the probe's missing-file refusal skipped it. better-sqlite3 then opens it as a relative path whose directory does not exist:connect()only warns andcheckHealth()answersfalse, which the probe does not read as a failure. sqlite-wasm opens an ephemeral in-memory database.sqlite: every statement fails "Cannot open database because the directory does not exist".sqlite-wasm: writes succeed and are gone at the next connect or restart, with no error.DatasourceAdminService.assertValidConfig, the one judgement the test, create and update doors share, also refuses asqlite/sqlite-wasmfilename that is empty,:-prefixed but not exactly:memory:, or a path that can be neither opened nor created (driver-sql's write-free judgement). Same envelope and code as every invalid config (400 DATASOURCE_ADMIN_ERROR;result.ok: falsefrom the test). No new error code.:memory:, an existing file, and a creatable missing path (test still refuses it per datasources: the draft probe POST /datasources/test creates the SQLite database file on disk although nothing is saved #27512, save accepts it) answer as before.Follow-up owed (seat files): the spec
SqliteConfigSchema.filenamerefinement, so the build door andPUT /api/v1/meta/datasource/:namejudge the same; ui residue: the dialog's:memory:value instead of a placeholder (DatasourceResourcePage.tsx).{ "issue": 30331, "status": "done", "branch": "claude/project-thread-541rod", "pr": "https://github.com/objectstack-ai/objectstack/pull/30395", "session": "session_01Utp6Cb83ajcseWprEF9C9g", "premise_still_valid": true, "summary": "Server half only. Reproduced on main 8589756a through the real POST /api/v1/datasources/test: ':memory:/tmp/x30331/parts.db' answered ok:true for sqlite and sqlite-wasm. Cause: ':'-prefixed names are fileless to every layer, so the probe's absent-file refusal skipped it; native connect only warns and checkHealth false is read as ok; wasm opens an ephemeral database. A saved one fails every native statement, and on wasm loses every write at reconnect. The shared config gate (assertValidConfig, used by test, create and update) now refuses an empty, ':'-prefixed-but-not-':memory:', or unopenable/uncreatable filename with the value and reason, in the existing envelope.", "tests": "New src/__tests__/sqlite-filename-neither-memory-nor-path.test.ts, 14 passed (7 per driver) through the real HonoHttpServer, admin routes, DatasourceAdminServicePlugin and default factory. Ablation (sqliteDatabaseFilenameIssue short-circuited, src imported by relative path, no dist leg): 8 failed / 6 passed, the 8 refusals red and the 6 controls green; restored, 14/14. Full @objectstack/service-datasource: 72 files, 1183 passed. Typecheck exit 0.", "gates": "dispatch-gates --commands derived 61 families; each run at head cca010a64 with exit recorded before any pipe: 61/61 exit 0 (check-plugin-teardown-shape --self-test after fetching 621a4876; check:dual-build-cjs-loads after the full build). --ran: 61 derived, 61 run, 0 NOT-MEASURED, 0 UNRUN, exit 0. pnpm lint exit 0. CI check-runs on #30395: not yet reported.", "dogfood_battery": "cca010a64 · os-verify-lock: VERDICT command-exit 0 · Test Files 419 passed / 3 skipped (422) · Tests 3492 passed / 36 skipped (3528) · no failures", "line_budget": "source +78 -9 (2 files), tests +222 (1 file), changeset +20 (1 file); 4 files, +320 -9", "deviations": "none against the dispatch. Save now also judges an unusable path (the dispatch's pin list asks for it); the empty filename is refused too, as the card's 'neither :memory: nor a usable path' covers it. The harness's model-named commit trailer was not used (repo hook); trailers are model-free.", "files_changed": [ ".changeset/30331-sqlite-filename-neither-memory-nor-path.md", "packages/services/service-datasource/src/__tests__/sqlite-filename-neither-memory-nor-path.test.ts", "packages/services/service-datasource/src/datasource-admin-service.ts", "packages/services/service-datasource/src/sqlite-database-filename.ts" ], "mcp_calls": "2: mcp__github__issue_read x2 (get, get_comments); no GitHub write tool. Besides GitHub MCP: one subscribe_pr_activity on #30395 (session event subscription, writes nothing to GitHub)", "api_writes": "2 through the relay: pr_create #30395 (draft, no assignee, no labels, read-back identical) and this comment. Plus 1 git push to claude/project-thread-541rod (not REST)", "open_questions": [], "security_findings": "none", "out_of_scope_findings": [ "class: b · Seam: spec:SqliteConfigSchema.filename ('Database file path, or :memory:') → runtime: build door and PUT /api/v1/meta/datasource/:name · reach: not measured at those doors (the admin doors are fixed here) · evidence: sqlite.zod.ts types filename as placeholderFree(z.string()) · dedupe words: sqlite filename refinement, :memory: prefix, meta datasource filename", "carrier: ui lane (#30331 stays open) · the New datasource dialog holds ':memory:' as a value instead of a placeholder (ui/packages/app-shell/src/views/metadata-admin/datasource/DatasourceResourcePage.tsx)", "noted, not filed (no public input measured after this fix): DatasourceAdminServicePlugin.probe reports ok:true when the handle's ping (checkHealth) answers false — recorded in the PR's Acceptance notes" ] }Report taken 2026-11-05T09:34Z.
Generated by Claude Code
objectstack-fleet commentedon Nov 5, 2026
domain:server#4 · os-bill · session_01Utp6CZMkHvuDeVYgTD73qF
Landed · PR #30395 · server half · card stays open for the ui half
22f0d656a;Refs #30331, so this card stays open. Seat record on headcca010a6(BREAKING, not governed, ≤ 3000 lines; landed per pm-dispatch on the maintainer's 13:47Z word).mainsourcepackages/services/service-datasource/src/datasource-admin-service.ts:1382const filenameIssue = await sqliteDatabaseFilenameIssue(driver, config);:memory:/…) is refused at create and at a config-touching edit, with a message naming the field.:memory:as the Filename field's value instead of a placeholder (ui/packages/app-shell/src/views/metadata-admin/datasource/DatasourceResourcePage.tsx). With this PR onmainthe mistake it invites is refused with a message instead of passing.falsehealth check is #30407 (seat 3).Re-routed:
domain:server→domain:ui,pm:dispatched→pm:queue, theos-billassignee comes off with this comment.Posted 2026-11-05T15:22Z.
objectstack-fleet commentedon Nov 5, 2026
domain:ui#5 · os-sales · session_01L6bhgVwQZbHRRM9CaXeKif
Claim: PM loop round 9 · 2026-11-05T15:24Z
Session:
session_01L6bhgVwQZbHRRM9CaXeKifAccount:
os-salesBranch:
claude/project-thread-5qt4jg(card thread of this project; a thread can push only its designated branch — measured platform exception toclaude/issue-<n>-<slug>)Worktree:
objectstack-issue-30327Domain:
domain:uiSeat:
domain:ui#5File surface:
ui/packages/app-shell/src/views/metadata-admin/datasource/DatasourceResourcePage.tsx(the New datasource dialog: schema-mode default and wording, the managed datasource's no-sync note, the SQLite Filename placeholder) + newstudionamespace keys (en/zh) + their tests + one root.changeset/. ⛔packages/**: if the driver form labels (Filename,Auto-migrate schema) come from the server's driver schema, the dev reports it and the seat files adomain:servercard.Container & model:
M,mode:thread,model: default,effort: highClause-②: yes (expected: new studio keys; the dev declares the final arm)
Metadata-protocol: none
Responsibility: the New datasource dialog | the SQLite Filename box holds
:memory:as a value, so typing a path appends to it | every administrator connecting a SQLite fileThread-read: 6652062778
Premise: holds (on
origin/main8e5fabe5the dialog still defaults to managed and holds:memory:as a value)Serial constraints cleared: PR #30493 (this seat, the same file) merged 2026-11-05T15:17Z; the branch starts from it. The server half landed in PR #30395 (merged). One PR carries this card's ui half and #30327, because both change the same dialog.
Dispatch order lives in the card thread's opening message.
objectstack-fleet commentedon Nov 5, 2026
domain:ui#5 · os-sales · session_01L6bhgVwQZbHRRM9CaXeKif
Landed at 2026-11-05T16:43Z: PR #30561 merged as
0d54279e73(seat ACCEPT 6663075162 on headfd64ec80, TIER_DEFAULT contract review PASS 6663047260),Fixes #30327andFixes #30331. The merge read 2026-11-05T16:41Z on the PR.git grep -c "NEW_DATASOURCE_SCHEMA_MODE|withBoxDefaults" origin/main -- ui/packages/app-shell/src/views/metadata-admin/datasource/DatasourceResourcePage.tsxState change:
pm:dispatchedremoved. Server follow-ups from the card thread: #30562 (a managed SQL datasource drops and recreates an existing_id-without-idtable, rows lost) and #30563 (driver form labels are English in every console language).