* fix(dolt): honor --dolt-auto-commit batch/off in SQL-server mode (bd-4wamg)
In server mode the auto-commit mode was declared but unenforced: the CLI's
policy hooks were all embedded-gated, and the storage layer minted one Dolt
commit inside every write transaction unconditionally — measured at one
commit per write through all five configuration paths (per-invocation flag,
config.yaml, env, bd config set, no-daemon). A constellation running ~482
versioned commits/h on a shared server could not use the documented
coalescing relief.
- issueOpsContext and transactHonoringAutoCommit are now mode-driven, not
embedded-only: batch/off set the deferred-version-commit context / blank
the commit message in server mode too.
- The storage layer's three per-write commit sites (doltAddAndCommitInTx,
doltAddAndCommit, the deleter's inline site) honor the deferral and leave
writes in the working set for an explicit commit point (bd dolt commit /
CommitPending). Explicit commit APIs (Commit, CommitWithConfig,
CommitPending) and the federation sync point are deliberately unguarded.
- The server-mode default flips off->on — a spelling change, not behavior:
the old default resolved to 'off' while every value behaved like 'on'.
With batch/off now real, 'on' is the compatible default. Help text stops
claiming 'Default: off'.
- The embedded-only flush points (PersistentPostRun, create post-writes)
keep their embedded gate via embeddedWritesCommitNow, now layered on the
shared writesCommitNow.
On a shared server, staging is table-level: an on-mode writer's commit may
sweep up another client's deferred rows. Deferral bounds who creates
commits, not commit contents — acceptable for coalescing, documented in the
changelog.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
* fix(delete): thread batch/off auto-commit deferral through bd delete (bd-4wamg)
Fable adversarial review blocker: the deleter's VersionCommitDeferred
guard was dead code — delete.go passed a bare rootCtx on both the
single-ID and batch paths, so server-mode batch still minted one Dolt
commit per bd delete. Thread issueOpsContext exactly as update.go does.
Coverage: new TestDeferredVersionCommitDeletePath at the storage layer
(plain ctx advances history, deferred ctx leaves the delete in the
working set for CommitPending); batch-mode delete added to the embedded
container test; scope docs per review (proxied-server routes never
apply auto-commit policy — batch/off inert there, follow-up filed) and
the shutdown-flush help line qualified to live batch-mode processes.
---------
Co-authored-by: Claude Fable 5 <noreply@anthropic.com>