Phase 1 sub-commit 6: skeleton wiring — make dev runs end-to-end
All checks were successful
CI / Quality (push) Successful in 5s
All checks were successful
CI / Quality (push) Successful in 5s
Symfony app under framework/skeleton/symfony/: minimal bin/console, public/index.php, MicroKernel-based src/Kernel.php, services.yaml, framework/security/mercure config, and a demo App\Controller\PingController that GETs /api/ping (returning JSON pong) and republishes the same payload to the Mercure topic app://ping. composer.json uses a path repository to symlink the bundle from ../../php so local edits are picked up live. QML app under framework/skeleton/qml/: top-level CMake that add_subdirectory's framework/qml, a main.cpp that creates the Qt process, runs SingleInstance.acquireOrForward before any QML loads, exposes SingleInstance via context property, and loadFromModule's Skeleton.Main. Main.qml uses BackendConnection / RestClient / MercureClient from PhpQml.Bridge and renders status dots, a Ping button, and an event log. Caddyfile binds 127.0.0.1:8765, enables in-memory Mercure with a 256-bit dev JWT (matches symfony/.env, lcobucci/jwt requires this). Makefile wraps build / dev / doctor / clean; scripts/dev.sh starts FrankenPHP --watch and the Qt host together with explicit PID-based teardown (process-group `kill 0` proved unreliable when frankenphp's watch fork reparented). Bug fixes uncovered in this sub-commit: - SingleInstance.acquireOrForward: probe-first, then removeServer + retry-listen. The original loop-with-removeServer-after-failed-bind silently exited on stale sockets from prior runs. - Main.qml: MercureClient does NOT inherit BackendConnection.token — Mercure subscribes anonymously in dev (Caddyfile), and forwarding the bridge bearer made it 401-loop. - /api/ping was 500ing because the dev MERCURE_JWT_SECRET was 144 bits; bumped to 64-char (>=256 bit) to satisfy lcobucci/jwt. - Linked the framework lib (php_qml_bridge) explicitly in addition to the QML plugin so SingleInstance.h resolves. - Auto-generated config/reference.php gitignored. Smoke verified offscreen: /healthz 200, /api/ping 200, 1 publish, 1 subscriber, zero 401s, clean shutdown with no zombies. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
@@ -31,14 +31,10 @@ bool SingleInstance::acquireOrForward(const QStringList& launchArgs)
|
||||
{
|
||||
const QString name = endpointName();
|
||||
|
||||
for (int attempt = 0; attempt < kBindRetries; ++attempt) {
|
||||
if (m_server.listen(name)) {
|
||||
return true; // we are the live instance
|
||||
}
|
||||
|
||||
// Bind failed. Either a real instance is running (forward args), or
|
||||
// a stale socket/pipe is left over from a crashed process (clean up
|
||||
// and retry). Distinguish by trying to connect.
|
||||
// Probe first: if a live instance answers, forward and exit.
|
||||
// This avoids a race where eagerly calling removeServer() would break
|
||||
// a running peer.
|
||||
{
|
||||
QLocalSocket probe;
|
||||
probe.connectToServer(name);
|
||||
if (probe.waitForConnected(kForwardConnectTimeoutMs)) {
|
||||
@@ -50,16 +46,24 @@ bool SingleInstance::acquireOrForward(const QStringList& launchArgs)
|
||||
probe.disconnectFromServer();
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
// No live peer responding — likely stale socket. Remove and retry
|
||||
// with a brief backoff (PLAN.md §3 *Edge cases — Single-instance
|
||||
// launch race*).
|
||||
QLocalServer::removeServer(name);
|
||||
// No live peer answered. The endpoint may not exist, or a stale
|
||||
// file is left over from a crashed process — removeServer() handles
|
||||
// both cases safely.
|
||||
QLocalServer::removeServer(name);
|
||||
|
||||
for (int attempt = 0; attempt < kBindRetries; ++attempt) {
|
||||
if (m_server.listen(name)) {
|
||||
return true;
|
||||
}
|
||||
QThread::msleep(kBindRetryDelayMs);
|
||||
}
|
||||
|
||||
// Exhausted retries without binding and without a live peer. Better to
|
||||
// act as the live instance than to deadlock both processes into exiting.
|
||||
// Exhausted retries — better to continue as a degraded "live" instance
|
||||
// than to deadlock-exit (PLAN.md §3 *Edge cases — Single-instance
|
||||
// launch race*). Subsequent invocations may not be forwarded, but
|
||||
// this process will still run.
|
||||
return true;
|
||||
}
|
||||
|
||||
|
||||
Reference in New Issue
Block a user