* wip(storage): events table -> dolt_ignored plane, migration 0062 + ignored 0019 (bd-red8u)
Events-only flip per the mitigation-first ruling (comments hold on
bd-1quyq). Main 0062 copies rows aside, drops the tracked table, commits
the drop out of HEAD (conditional DOLT_ADD + --skip-empty DOLT_COMMIT),
registers the dolt_ignore pattern, recreates + restores. Ignored 0019 is
the fresh-clone creator (PREPAREd conditional CREATE — the __temp__+RENAME
dance collides on the database-unique fk_events_issue name). Version-gated
dolt_ignore seeding (cursor >= 62), doctor ignored-table lists, embedded
flip/fresh-clone contract tests, and existing events-at-HEAD assertions
retargeted to the new plane contract.
KNOWN OPEN: server-path (tx-pinned) flip does not detach — see bead.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix(storage): force-add pierces dolt_ignore in 0062 detach; rekey fixture commits events seed (bd-red8u)
The 0062 replay conditional used plain DOLT_ADD('events'), which is a
silent no-op when the 'events' pattern already exists in dolt_ignore
while the table is still tracked at HEAD (the branch-per-test shared DB
materializes exactly that state) — the drop never committed and the
recreate netted out to a permanently-dirty tracked table. '-f' pierces
the ignore exactly when a tracked-events drop is pending; on the normal
path no pattern exists yet, and on phase-3+ replays the status probe
reads zero so no add runs at all.
The aux-rekey fixture simulates a pre-flip lineage, where events rows
were committed; DOLT_COMMIT('-Am') now skips the ignored table, so the
seed is staged explicitly with '-f' before the commit.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(storage): events-at-HEAD assertions tolerate the branch-test materialized shell (bd-red8u)
The shared branch-per-test database deliberately commits an empty events
shell to HEAD so branches inherit the schema
(testutil.MaterializeLocalTableSchemasForBranchTests), so asserting the
table absent from committed history — the production shape, pinned by
the embedded contract tests — fails there. Assert the invariant that
holds on both shapes instead: no events ROW ever reaches HEAD. The
dolt_status checks drop events for the same reason: tracked-at-HEAD plus
working-set audit rows legitimately shows as a modification on test
branches, while production's untracked table never appears at all.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* test(domain/db): recreate working-set events table after per-test hard reset (bd-red8u)
DOLT_RESET('--hard') swaps the tracked issues table's backing object,
which silently severs fk_events_issue enforcement on the untracked
dolt_ignored events table (verified on dolt-sql-server 2.2.0 — bogus
issue_id inserts start succeeding while labels' FK keeps enforcing).
The suite's baseline commit no longer contains events at all, so
SetupTest recreates the table from DDL captured at SetupSuite: the
fresh table re-links the FK against the reset root and clears audit
rows the reset orphaned. Production hard-reset sites tracked in
bd-7bpkd.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Migration 0051 (bd-2rd37) dropped DEFAULT (UUID()) from comments.id and
events.id, breaking every raw test INSERT that relied on the default:
fk_cascade_merge_test.go (FK settle coverage), create_issue_commit_test.go
(three dirty-side-table cases), and git_remote_test.go (round-trip and
post-pull seeds, integration-tagged). All failed with "Field 'id' doesn't
have a default value" since this morning; they went unnoticed because the
dolt package run exceeded its timeout in the wave's validation. Explicit
UUID() ids now, consistent with the app's own insert sites (bd-6dnrw.18).
The embedded twin (pull_settle_test.go) was repaired in d1a2fc75a.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Fixes#4259 by replacing clone-local dependency UUIDs with deterministic IDs derived from the dependency edge. Existing rows are backfilled, the random default is removed so missed insert paths fail loudly, and same-edge merge conflicts become normal resolvable row conflicts instead of unmergeable duplicate logical edges.
Agent-Signature: kilocode-gpt-5.5-unknown-reasoning on behalf of maphew