#291 fix(dev): resolve meridian binary to repo build in dev to stop migration drift

[Anonymous]

Originally created by: adityaharishch

Problem

In local dev, the Tasks → Sync button failed with Sync failed: tasks-sync: open db: failed to run migrations.

Root cause is a dev-only binary mismatch:

• dev-start.sh runs the daemon as cargo watch -x 'run --bin meridian' → target/debug/meridian. That build is rebuilt on every save and migrates meridian.db forward as new migrations land.
• UI route handlers (/api/tasks/sync, OAuth, triage) resolved a separately-installed binary under ~/.meridian/app/bin/meridian via selectMeridianBinary(meridianCandidates()). That artifact does not track the repo.

So the instant a migration was added, the daemon advanced the DB while the installed CLI stayed behind, and sqlx refused to open the ahead-of-it DB — surfacing as the opaque "failed to run migrations".

Fix

Make meridianCandidates() environment-aware:

• NODE_ENV === 'development' (npm run dev): prepend <repo>/target/debug/meridian then target/release/meridian ahead of the installed paths (next dev runs with cwd = <repo>/ui, so repo root is one level up). The UI now shells out to the same binary the daemon runs — they can't drift.
• Production (NODE_ENV !== 'development'): unchanged — installed-only, native-first, preserving the launchd PATH-parity guarantee from the original env: node: No such file or directory fix.

Also: print the full anyhow source chain ({e:#}) on tasks-sync open-db failure so the real sqlx cause is visible next time instead of the top-level context (src/main.rs).

Tests

• Added dev-mode coverage in ui/__tests__/meridian-bin.test.ts: repo build prepended, debug-before-release, node wrapper still last, production no-op.
• Pinned the existing selectMeridianBinary tests to 'production' for determinism.
• bun test → 12 pass / 0 fail; UI build clean; full pre-push suite passed.

🤖 Generated with Claude Code

[Anonymous]

Originally posted by: Akarsh-Hegde

🔍 Code review — resolve meridian binary to repo build in dev

High-effort pass, verified against source. Clean — no bugs found. Nice fix for a real dev/prod drift trap.

The change adds an env/cwd-parameterised dev branch to meridianCandidates(): under NODE_ENV === 'development' it prepends target/debug/meridian then target/release/meridian ahead of the installed paths, so the dashboard shells out to the same binary cargo watch runs — closing the window where a freshly-added migration advances meridian.db past what the installed CLI knows, and sqlx aborts task-sync with the opaque "failed to run migrations".

What I checked:

• Safe production default — any env !== 'development' (incl. undefined) returns the installed-only list unchanged. ✅
• Safe degradation if the cwd heuristic is wrong — resolve(cwd, '..') assumes cwd = <repo>/ui; if it isn't, the prepended target/* paths simply aren't executable and selectMeridianBinary skips them. No crash, falls through to installed. ✅
• launchd-parity preserved — native binaries still precede the node wrapper; the wrapper stays last. The new test 'still lists the node wrapper last' guards it.
• main.rs {e:#} correctly prints the full anyhow source chain (the sqlx migration cause) instead of just the top-level context — directly useful for the drift this PR targets.
• Tests cover dev ordering (debug < release < installed), the production no-prepend case, and the existing selection logic (now explicitly passing 'production').

LGTM.

_{🤖 Assisted review via Claude Code.}

[Anonymous]

Ticket changed by: adityaharishch

• status: open --> closed

[Anonymous]

Originally posted by: adityaharishch

🎉 This PR is included in version 1.53.1 🎉

The release is available on:

• GitHub release
• v1.53.1
• npm package (@latest dist-tag)

Your semantic-release bot 📦🚀