forst-lang / forst / 29642793206

75%

main: 76%

Build Type

Pull #146

github

Committed by haveyaseen

Commit Message

fix(nodert): recover dead supervisors and harden interop CI

Replace sync.Once supervisor init with mutex-backed respawn when the Node
process or RPC read loop dies, so GetClient can bootstrap or reattach again
instead of returning a corpse forever. Watch bootstrap children, detect dead
connections via readLoopExited, and add integration coverage for in-flight
ErrNodeRuntimeDied plus respawn on the next CallSync.

Lower DefaultCallTimeout from five minutes to thirty seconds when ftconfig
does not override it, so stuck exports fail fast in production-shaped deploys.

Treat intentional SIGKILL after shutdown grace as success in managedProcess
terminate, fixing host-mode Shutdown failures when stdout spam or keepalive
shims prevent graceful exit.

Harden multipackage reload e2e: wait for node.sock.ready, trigger reload via
main.ft content change, detect generation from reload marker and health,
assert host exits after forst dev SIGINT, restore main.ft, and kill stale
6321 listeners after embedded-invoke run.

Simplify multi-package-dev host.mjs to a keepalive shim; fix register.cjs to
run host startup through Effect.runPromise with ForstNodeRuntimeLayer.

Tighten allowlist widen rejection to true supersets only (retain all frozen
exports plus new ones), with a lifecycle unit test for non-widen fingerprint
changes.

Use short per-test Unix socket paths under /tmp to stay within macOS path
limits in host integration and unit tests.

Pull Request Pull Request #146: test(ci): gate interop deploy paths and harden invoke coverage

Coverage Stats

85 of 115 new or added lines in 7 files covered. (73.91%)

5 existing lines in 3 files now uncovered.

35783 of 47424 relevant lines covered (75.45%)

1169.39 hits per line