kandev/docs/plans/go-dev-launcher/task-05-run-dev.md

id, title, status, wave, depends_on, plan, spec

id	title	status	wave	depends_on	plan	spec
05-run-dev	runDev orchestration	done	2	01-dev-paths 02-db-backup 03-web-child 04-dev-command	plan.md	../../specs/platform/requirements/go-dev-launcher.md

Task 05: runDev orchestration

Assemble tasks 01–04 into the working kandev dev command.

Acceptance

• runDev performs, in order: repo-root discovery → dev port selection → dev backend env resolution → production-DB backup (abort on failure) → startup banner → supervisor and signal handlers → backend launch → health wait (600s dev timeout, dumping captured backend output on failure) → Vite launch → URL readiness wait → browser open → waitForAppExit. Every failure path shuts the supervisor down before returning.
• The backend child is make -C apps/backend dev with CWD at the repo root, launched through launchRestartableBackend so the restart control socket and launch manifest behave exactly as they do in start.
• launchRestartableBackend takes the supervisor home directory as an explicit parameter instead of calling resolveHomeDir() internally, so dev's supervisor state lands under <repoRoot>/.kandev-dev/supervisor/ and not in the ambient KANDEV_HOME_DIR (which, inside a Kandev task workspace, is the parent backend's). runStart and runInstalled pass resolveHomeDir() and are behaviorally unchanged.
• --headless skips the browser open, as in start.

Verification

Use TDD with the existing seam pattern in start.go:12-21 (package-level function vars swapped in tests) so runDev is exercised without spawning real processes. Then:

cd apps/backend && go test ./internal/launcher/ -race && go build ./... && golangci-lint run ./internal/launcher/... --timeout=5m

Files

• apps/backend/internal/launcher/dev.go (new)
• apps/backend/internal/launcher/dev_test.go (new)
• apps/backend/internal/launcher/supervisor.go, supervisor_test.go — homeDir parameter
• apps/backend/internal/launcher/start.go, start_test.go — updated call site
• apps/backend/internal/launcher/run.go — updated call site
• apps/backend/internal/launcher/launcher.go — real CommandDev dispatch

Inputs

• apps/cli/src/dev.ts:26-110 — the orchestration order and the exact log lines ([kandev] starting backend..., [kandev] backend ready at ..., [kandev] starting web..., [kandev] open: ...). Keep them; people grep for these.
• apps/backend/internal/launcher/start.go:86-124 — runManagedApp, the closest existing shape. Consider whether runDev can reuse it with a web-child hook rather than duplicating the backend/health/exit sequence.
• docs/decisions/0019-restart-supervisor.md — the manifest and control-socket contract.

Risks

• launchRestartableBackend already takes eight parameters; adding homeDir pushes it past readability and the lint limits. Introduce a backendLaunchConfig struct in the same change rather than appending an argument.
• The dev backend child is make, not the launcher binary, so dumpLogs captures make's output too. Verify the failure dump is still readable when the Go build itself fails — that is the most common dev startup failure and the output must make it obvious.
• Do not re-exec self __backend for dev. That would drop rebuild-on-restart; see the plan's design decisions.