# Remove Unused db-sync Cycle Repair

## Problem

`deps/db-sync/src/logseq/db_sync/cycle.cljs` implements approximately 370 lines
of client-side DataScript repair for `:block/parent` and
`:logseq.property.class/extends` cycles. Its approximately 220-line unit test
namespace is the only current consumer.

The client remote-apply path stopped requiring and calling `fix-cycle!` in
commit `905e131770` (`fix: no need to fix parent cycle for blocks`) on
2026-03-20. Current `frontend.worker.sync.apply-txs` repairs duplicate outliner
orders but has no cycle-repair call. Repository-wide searches find no other
production, script, documentation-contract, reflection, registry, or generated
source reference. The `@logseq/db-sync` package is private and does not export
the namespace as an external API.

The unused utility therefore preserves a behavior model, transaction metadata,
and tests that are not part of runtime behavior. The implemented runtime guide
must also carry a special warning explaining that the utility is not integrated.

## Proposal

Delete:

- `deps/db-sync/src/logseq/db_sync/cycle.cljs`; and
- `deps/db-sync/test/logseq/db_sync/cycle_test.cljs`.

Remove the `Cycle repair utility` subsection from
`docs/agent-guide/implemented/architecture/2026-08-24-logseq-runtime-and-engineering-guide.md`.
The runtime guide should continue to describe the active duplicate-order repair
where relevant, without documenting a removed test-only implementation.

Do not change `frontend.worker.sync.apply-txs`, transaction application order,
outliner semantics, or server-side validation. Generated
`deps/db-sync/worker/dist/worker-test.js` is excluded from source edits and will
be regenerated by the db-sync test build.

## Alternatives considered

### Reconnect `fix-cycle!` to remote apply

Rejected as a simplification because it would add cycle-repair transactions and
change synchronized graph behavior. If runtime evidence shows cycle repair is
needed, that work belongs to a separate bugfix or architecture decision with
new behavior tests.

### Keep the utility as a test-only design reference

Rejected because version control already preserves its history, while the
current source and test suite should describe executable behavior. Keeping it
also forces the runtime guide to distinguish implemented code from the active
path.

### Extract a generic graph-cycle library

Rejected because there is no second production consumer and the extraction
would relocate unused complexity instead of deleting it.

## Acceptance criteria

- The cycle repair source and its self-contained unit test namespace are
  removed.
- Repository-wide search finds no `logseq.db-sync.cycle`, `cycle/fix-cycle!`, or
  equivalent build/test registration reference in tracked source.
- `frontend.worker.sync.apply-txs` retains its current remote apply and
  duplicate-order repair behavior unchanged.
- The implemented runtime guide no longer documents the removed test-only
  utility or suggests that active client sync repairs reference cycles.
- `pnpm --dir deps/db-sync test` passes and regenerates its test bundle without
  the deleted namespace.
- `bb lint:carve` passes.
- `bb dev:lint-and-test` passes.

## Risks

- The deleted implementation includes algorithms for two kinds of cycles, so a
  future runtime bugfix cannot reuse it directly from the working tree. Its Git
  history and focused tests remain recoverable from commits before this
  decision.
- The removal assumes that the March 2026 production-path deletion was
  intentional for both supported attributes. Reintroducing any client-side
  repair would be behavior-changing and requires a new decision rather than a
  reversal hidden inside this simplification.

## Questions

- None.