565c2114be551ec3491e30a780a8485a97fee7d4
1 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
| 565c2114be |
Fase 8 (1/N): delete the worker-side remote sync protocol
Deletes frontend.worker.sync/* wholesale (apply_txs, asset_db_listener,
assets, auth, client_op, const, crypt, download, handle_message,
large_title, log_and_state, presence, temp_sqlite, transport, upload,
util), worker/sync.cljs, worker/handler/sync.cljs, and
worker/ui_request.cljs -- the remote CRDT sync protocol and its E2EE
password-request plumbing, both made obsolete by the server-client
daemon architecture built in Fases 1-7. Deviates from the plan's KEEP
list on sync/presence.cljs and sync/transport.cljs: neither has any
caller left once the rest of the cluster is gone (confirmed by grep),
because Fase 4's WS resume-by-revision work already built its own
self-contained module (db_worker_node_ws.cljs) instead of adopting
these as the plan assumed it would.
This touched far more than a mechanical deletion:
- frontend.worker.sync.client-op wasn't sync-only. Its client_ops
SQLite table is also the idempotency marker for db_core.cljs's
built-in-property repair migration and the forward/inverse-outliner-
ops relay undo/redo reads back through history-action-ops-by-tx-id.
Trimmed (not deleted) into frontend.worker.local-tx-log, keeping only
the tx-entry/local-tx functions those two callers need and dropping
the genuinely sync-only surface (pending-upload counts, checksums,
graph-uuid, sync-conflict tracking, asset-op queues).
- frontend.worker.sync.apply-txs's apply-history-action! looked
sync-specific (it lived next to the remote-tx-apply/rebase machinery)
but turned out to be the *only* implementation ever registered for
worker-undo-redo/*apply-history-action! -- undo_redo.cljs calls it on
every db-transact undo/redo, local edits included, not just
remote-origin ones. Deleting it outright would have silently broken
undo/redo app-wide the next time anyone hit Cmd-Z. Ported the
general-purpose slice (replaying a recorded entry's forward/inverse
outliner ops against the current db state, plus the raw-tx-data
fallback for edits that bypass the structured outliner-op layer) into
a new frontend.worker.undo-redo.replay, along with persist-local-tx!
(derive history ops on every local commit, stash them in
local-tx-log, hand off to gen-undo-ops!) -- dropping only the
remote-rebase-specific rest (CRDT conflict resolution, upload-queue
bookkeeping). Wired into db_listener.cljs as a renamed
:persist-local-tx handler (was :db-sync, a name that stopped meaning
anything once sync was gone).
Caught by running the full existing undo_redo_test.cljs suite (30
deftests written for the old mechanism) against the new module after
fixing its test harness to use the real db_listener.cljs dispatch
instead of a hand-rolled listener -- all 29 tests port over passing
with no behavior change, which is the actual evidence this preserves
undo/redo correctness rather than just compiling.
- db_core.cljs's own 8 :thread-api/db-sync-* handlers (remote-graph-
download import, close-db/invalidate-search-db/recreate-lock
sync-specific variants) are deleted along with their sync-crypt/
sync-download calls; :thread-api/unsafe-unlink-db, :close-db, and
:release-access-handles keep their real (non-sync) bodies.
db_worker_node.cljs's non-repo-methods set drops from 22 to 7
entries (the 15 removed were all e2ee/db-sync/ui-request keywords
backed by deleted files).
- worker/handler/export.cljs's :thread-api/recompute-checksum-diagnostics
(and worker/db/validate.cljs's backing recompute-checksum-diagnostics,
which pulled in logseq-db-sync.checksum) compared a local checksum
against a remote one that no longer exists -- deleted outright, not
ported, since there's nothing left to diff against.
- worker/handler/block.cljs's :block.temp/sync-conflicts metadata
(backed by the deleted client-op/get-all-sync-conflicts) is gone;
its client-side consumer (a retry-upload/clear-conflicts UI in
components/block.cljs) is dead code now and still needs a follow-up
pass -- flagging here since worker-side cleanup alone can't remove
it without breaking that UI's thread-api calls at runtime.
Verified: app/electron/db-worker-node/mobile all compile clean. Full
worker test suite green (db_core_test 131, db_listener_test 11,
db_worker_test 8, db_worker_node_test 59, db_worker_node_lock_test 10,
local_tx_log_test 4, undo_redo_test 29, subs_test 50, subs_loader_test
2, config_test 6 -- 0 failures across all of them). The entire
src/test/frontend/worker/sync/ test directory (7 files) and
db_sync_test.cljs/db_sync_sim_test.cljs/node_sync_test.cljs (all
testing only-deleted namespaces) are deleted wholesale rather than
ported.
Known follow-up, not yet done: the client-side (frontend.handler.*,
frontend.components.*) RTC/E2EE/db-sync UI and its dead thread-api call
sites (components/block.cljs, components/repo.cljs,
components/settings.cljs, handler/user.cljs, handler/assets.cljs,
db/restore.cljs, handler/common/developer.cljs, and the
handler/db-based/{sync,rtc-flows,rtc-background-tasks}.cljs +
handler/events/rtc.cljs + components/rtc/* + db/rtc/* files themselves)
-- a separate commit.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
