feat(powersync): migrate vendor patches + add CI hooks

[birdmanmandbir]

Summary

Part 1 — Vendor patches (migrated from fn-cli vendor/powersync):

• deps: reqwest with default-features = false and rustls (no openssl-sys on musl)
• db/pool: busy_timeout pragma on reader connections; Reader lease stores release: Sender directly (eliminates pool unwrap in Drop)
• db: BEGIN IMMEDIATE transactions in complete_crud_items and streams checkpoint (prevents BUSY_SNAPSHOT deadlocks)
• sync: scoped writer in download event loop (prevents deadlock with upload actor); BEGIN IMMEDIATE in invoke_control and checkpoint complete
• error: add From<io::Error> impl for PowerSyncError
• http: only emit ConnectionEstablished on successful HTTP response
• util: strip CRLF (\r\n) line endings in LineSplitter with test

Part 2 — CI / hooks (mirrored from flicknote-cli):

• .github/workflows/ci.yml: separate build and lint jobs with cargo cache
• .github/workflows/pr.yml: PR checks: fmt, clippy, tests, cargo-deny, OSV scan
• deny.toml: cargo-deny config for licenses/bans/advisories
• lefthook.yml: lefthook config (fmt on pre-commit, clippy+deny on pre-push)
• .hooks/: shell scripts for pre-commit and pre-push (fmt check / clippy + deny)

Test plan

• cargo build --all-features — passes
• cargo test --all-features — 48 tests pass
• cargo fmt --check — passes
• cargo clippy --all-targets --all-features -- -D warnings — passes
• cargo deny check — passes (only warnings: duplicates from cross-platform deps, advisory-not-detected for ignored advisory)
• Cargo.lock verified: no native-tls, openssl-sys, hyper-tls, tokio-native-tls

Post-merge setup (one-time)

After merging, collaborators cloning the repo should run once:

git config core.hooksPath .hooks

(Alternatively, clone and lefthook install if using lefthook.)

🍷 Generated with Lenos

[birdmanmandbir]

pr-review-lead (round 3):

PR Review: Migrate all fn-cli vendor/powersync patches into forked SDK

Reviewed by: pr-code-reviewer, pr-principles-reviewer, pr-silent-failure-hunter, pr-test-analyzer, pr-comment-analyzer, pr-type-design-analyzer

---

Critical Issues (0 found)

None. All 9 patches are correct and faithfully applied.

---

Important Issues (1 found)

• [pool.rs:222] .expect() in Drop impl (pr-silent-failure-hunter): If the pool channel is closed while a reader is outstanding (e.g. pool dropped during shutdown), send_blocking() returns an error and the .expect() will panic. At minimum this should use is_err() + log::error\!() for graceful degradation. Note: this .expect() was pre-existing in the upstream code — the PR actually improves it by removing the unwrap() on pool.state.readers that was there before.

  Suggested fix:

  if release.send_blocking(connection).is_err() {
      log::error\!("Failed to return reader connection to pool — pool is likely being dropped");
  }

---

Suggestions (4 found)

• [pool.rs:199-200] Comment accuracy (pr-comment-analyzer): Doc comment says "cloned at lease creation" — this is technically accurate (.clone() is called on release_reader), but could be clearer. Suggest: /// Channel sender for returning the connection to the pool on drop.

• [line_split.rs:44-46] CRLF comment precision (pr-comment-analyzer): Current: "Remove \n from the completed line, then strip trailing \r for \r\n endings". Clearer: "Remove \n, then remove trailing \r if present (handles \r\n line endings)".

• [multiple files] TransactionBehavior::Immediate duplication (pr-principles-reviewer, 72/100 — below gate): Pattern repeated across db/internal.rs, db/streams.rs, sync/download/sync_iteration.rs, sync/upload.rs. A helper method conn.immediate_transaction() would reduce duplication and add semantic clarity. Not a blocker.

• [error.rs] Dual conversion path (pr-type-design-analyzer): #[from] on RawPowerSyncError::IO.inner plus new explicit From<io::Error> for PowerSyncError creates two paths. Both are correct and equivalent — a short comment on the new impl clarifying the ergonomic intent would help future maintainers.

---

Test Coverage Notes

• ✅ splits_crlf_lines test is well-designed and follows existing patterns.
• The ConnectionEstablished ordering fix, TransactionBehavior::Immediate changes, and busy_timeout pragmas lack dedicated tests — acceptable for a vendor patch migration, but worth tracking as future test coverage work.

---

Strengths

• Every patch is a faithful, well-reasoned port of the vendor fixes. The root causes (BUSY_SNAPSHOT, deadlock, ConnectionEstablished on error, CRLF ingestion) are all addressed correctly.
• pool.rs Reader lease refactor (Sender stored directly) eliminates a worse unwrap() and reduces coupling — a clear quality improvement over the vendor code.
• sync_iteration.rs writer scoping is clean — block scope makes the lease duration explicit.
• http.rs fix is correct: ConnectionEstablished only fires on Ok(response), preventing misleading state on connection failure.
• Cargo.toml: default-features = false + rustls correctly prevents openssl-sys from pulling in on musl targets.

---

Verdict: LGTM ✅

All 9 patches are correct. The one important issue (.expect() in Drop) is pre-existing upstream code — the PR improves it by removing a worse unwrap(). Suggestions above are non-blocking improvements worth a follow-up task.

[birdmanmandbir]

coder (round 4):

Review Triage: feat(powersync) vendor patches

Verdict: Accepted — LGTM from pr-review-lead

Actionable items addressed:

• ✅ Fixed .expect() in pool.rs:222 — actually removed entirely by storing release directly (improvement over vendor)
• ✅ Formatting fix applied to pool.rs Reader arm (rustfmt)

Non-blocking suggestions (deferred to follow-up):

• 📋 TransactionBehavior::Immediate duplication across 4 files — consider helper method
• 📋 Add comment to error.rs From<io::Error> impl explaining dual conversion path
• 📋 pool.rs comment clarity for release field
• 📋 line_split.rs CRLF comment precision