Skip to content

Database governance built for humans and agents — controlling changes and access across every major database.

docs: corroborate nested releases too, format sheet names from the owner

Danny6f100f271846
diff --git a/docs/design/sheet-blob-scoping.md b/docs/design/sheet-blob-scoping.md
index e5b7463879..785d0caf15 100644
--- a/docs/design/sheet-blob-scoping.md
+++ b/docs/design/sheet-blob-scoping.md
@@ -219,6 +219,13 @@ Three constraints together close it:
single release has not identified anything, and `EXISTS` would accept whichever duplicate happens
to predate the revision and carry the hash. This is the composite-PK identification rule in
`AGENTS.md` applied to an alternate key that turns out not to be one.
+
+ **A release is resolved in two places, and both need the same test.** Directly, from the
+ revision's own `release`; and indirectly, when a corroborating task's `payload.release` names one.
+ The indirect path is easy to leave on a bare `EXISTS` — it sits nested inside the task branch and
+ reads like a lookup rather than a corroboration — but a duplicate release carrying the hash would
+ corroborate a task whose intended release did not, which is the same over-grant by a longer route.
+ Exactly-one, age, and hash apply wherever `(project, release_id)` is resolved.
- **Temporal.** Require the corroborating row's `created_at` to be no later than the revision's.
A revision is written after the task run it records completes, so a genuine reference always
predates it, whereas a row that reused the ID after a purge is necessarily newer — the purge, the
@@ -518,12 +525,19 @@ WHERE r.payload->>'sheetSha256' IS NOT NULL
AND (
-- task.payload.source is a oneof: a sheet hash, or a release naming one
t.payload->>'sheetSha256' = r.payload->>'sheetSha256'
- OR EXISTS (
- SELECT 1 FROM release rel2
- CROSS JOIN LATERAL jsonb_array_elements(COALESCE(rel2.payload->'files', '[]'::jsonb)) AS f2
- WHERE rel2.project = (regexp_match(t.payload->>'release', 'projects/([^/]+)/'))[1]
- AND rel2.release_id = (regexp_match(t.payload->>'release', 'releases/([^/]+)'))[1]
- AND f2->>'sheetSha256' = r.payload->>'sheetSha256'
+ OR (
+ -- same exactly-one / age / hash test as the direct release branch
+ (SELECT count(*) FROM release rel2
+ WHERE rel2.project = (regexp_match(t.payload->>'release', 'projects/([^/]+)/'))[1]
+ AND rel2.release_id = (regexp_match(t.payload->>'release', 'releases/([^/]+)'))[1]) = 1
+ AND EXISTS (
+ SELECT 1 FROM release rel2
+ CROSS JOIN LATERAL jsonb_array_elements(COALESCE(rel2.payload->'files', '[]'::jsonb)) AS f2
+ WHERE rel2.project = (regexp_match(t.payload->>'release', 'projects/([^/]+)/'))[1]
+ AND rel2.release_id = (regexp_match(t.payload->>'release', 'releases/([^/]+)'))[1]
+ AND rel2.created_at <= r.created_at
+ AND f2->>'sheetSha256' = r.payload->>'sheetSha256'
+ )
)
)
);
diff --git a/docs/design/sheet-history-on-database-transfer.md b/docs/design/sheet-history-on-database-transfer.md
index ca84f21c41..e515cf2fa6 100644
--- a/docs/design/sheet-history-on-database-transfer.md
+++ b/docs/design/sheet-history-on-database-transfer.md
@@ -93,6 +93,8 @@ Change history follows the database; statement content does not.
- When a revision's sheet is not readable by the current project, the statement is withheld and the
response says so explicitly — naming the owning project where that can be verified — rather than
returning NotFound for the sheet.
+- The sheet *name* is formatted from that same resolved owner, not from the database's current
+ project. A name is emitted only when it is true.
- No sheet references are carried between projects. A database transfer grants no new read access,
and the four reassignment paths above need no changes.
@@ -223,8 +225,13 @@ corruption outweighs the modelling improvement.
## Implementation
- A field on the revision response carrying the withheld-content reason and the owning project,
- populated by the resolution order above — provenance first, `sheet_blob_ref` second, unknown last
- — and distinguishing a live owning project from a purged one.
+ populated by the resolution order above — provenance first, `sheet_blob_ref` second, unknown last.
+- `convertToRevision` formats `Revision.sheet` from the resolved owner rather than from
+ `database.ProjectID` (`backend/api/v1/revision_service.go:304`), and emits no sheet name when the
+ owner does not resolve. Leaving it built from the current project would keep exactly the false
+ resource name R7 exists to eliminate — a response asserting that project A owns the statement while
+ handing the client `projects/B/sheets/{sha}` to follow, which then 404s. The owner is already
+ computed for the withheld-content field, so this costs nothing extra.
- A UI affordance for the withheld state that names the owning project rather than rendering an
error, and says the project no longer exists when that is the case.
- No *carry* logic in `UpdateDatabase`, `BatchUpdateDatabases`, `updateInstanceLifecycle`, or
@@ -254,15 +261,16 @@ Tests:
is a regression test rather than a contrived one.
- Ambiguity: two projects in the caller's own workspace holding refs for one hash resolves to
unknown, not to whichever row comes back first.
+- Sheet name: after a transfer, `Revision.sheet` names the resolved owner and resolves to that
+ project's readable sheet for a caller who holds rights there; when the owner is unknown, no sheet
+ name is emitted rather than one pointing at the current project.
## Open items
-**Sheet name attribution.** `convertToRevision` formats the sheet name with the database's current
-project, which is incorrect after a transfer: it names the destination as owner of a sheet the source
-authored. The withheld-content response resolves this through provenance, but the `name` field itself
-is still built from `db.project` and stays wrong. Fixing it means formatting the name from the same
-provenance chain — which works for revisions carrying `release` or `taskRun` and leaves the rest
-without a correct name, so it is a partial fix rather than a clean one.
+**Revisions with no resolvable owner have no sheet name.** Where the owner resolves, the name is
+correct and unchanged for every database that never moved. Where it does not, the response carries
+no sheet name at all — a behavior change for clients that assume the field is always populated, and
+the residual cost of refusing to emit a name that is not true.
**Workspace-level roles have no escape hatch.** The gate checks the ref table for the named project,
so workspace admins and DBAs are refused as well, despite holding `bb.sheets.get` broadly. Whether