* feat(storage): add events/bd_events_journal/bd_events_seq to cursorRealityFloor sentinel set (port of #6547)
Port of the release/1.3.x commit aa3c43b6bd (PR #6547) to main, where it
never landed: v1.3.1 is cut from a release branch that is not an ancestor
of main.
events (ignored/0019) and bd_events_journal/bd_events_seq (ignored/0022)
are dolt_ignored, so they are never part of committed history and a fresh
clone can arrive with the ignored-plane cursor at-latest but without them.
cursorRealityFloor only knew about wisps/wisp_dependencies (floor 0) and
leases.granted_node (floor 11), so such a clone believed its cursor and
never re-created the missing tables.
A new sentinelFlooredTables list clamps the believable cursor to each
table's own replayFloor (18 for events, 21 for the journal pair) instead
of 0, so the replay re-runs the creating migration without re-running
ignored/0007's unguarded `UPDATE wisps SET is_blocked = 0`. Floored
tables are probed after the floor-0 tables and together with the column
sentinels, taking the minimum across everything found absent.
Tests: TestCursorRealityFloor and TestCurrentVersionClampsToSentinelFloor
gain the three single-absence cases, a two-floored-table case pinning the
intra-loop minimum, and a table+column case pinning the cross-loop
minimum; TestSentinelFlooredTablesAreCreatedByTheSeries asserts each
floored table is created strictly above its floor and that the floor
excludes ignored/0007. The sqlmock probe sequences in lock_test.go and
cursor_reality_test.go expect the extra INFORMATION_SCHEMA.TABLES probes.
The commit applied to main without conflicts; main's ignored series still
creates the three tables at 0019 and 0022.
Co-authored-by: Julian Knutsen <julianknutsen@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* test(schema): pin cursorRealityFloor's floored-table probe error
cursorRealityFloor returns a floored-table probe error instead of
reading the table as absent or skipping it, but no test reached that
return. TestCursorProbeErrorIsNotTreatedAsContradicted fails every
table probe, so it returns at the floor-0 wisps probe first. Turning
the return into `continue`, or into "absent", left the whole schema
package green.
TestFlooredTableProbeErrorIsNotTreatedAsContradicted fails exactly one
floored table per subtest (events, bd_events_journal, bd_events_seq),
with every other sentinel table present. It asserts that the error
propagates, that the cursor is not reported as contradicted, and that
no column sentinel is probed afterwards. stubSentinelTables applies one
error to every probe, so the test stubs sentinelTableExists inline
rather than widening a helper the other tests share.
Mutated under a go test -overlay, each of `continue`, "absent", and
returning limited=true alongside the error fails only this test's
three subtests. The unmutated package passes 474 of 474 tests and
subtests, none skipped.
The test is additive and sits between two existing tests. It adds no
conflict hunk to #7213, #7297, or #7409, and #7050's duplicate port
commit still replays as empty on top of this PR, so a rebase drops it.
For that reason the review's suggested rewording of the floored-table
rationale comments is left for after #7050 lands. The comments stay
identical to release commit aa3c43b6bd.
Agent-Signature: claude-code-claude-opus-5-5-max on behalf of Julian Knutsen
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Jim Wordelman <jim@wordelman.name>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>