magdev/php-qml
1
Fork 0
Code Issues Pull Requests Actions Releases 4 Activity

release prep v0.1.2: collapse audit fixes into v0.1.2
Some checks failed
CI / Quality (push) Failing after 5m8s
Details
Release / Linux AppImage (push) Failing after 5m4s
Details

Browse Source 
The previous commit (0cceefc) staged the audit-driven fixes under a
hypothetical v0.1.3 since v0.1.2 hadn't tagged yet. Tagging cadence
decision: ship them all as v0.1.2 — there's no point spending a tag
on the clean-shutdown fix alone when v0.1.3 was already lined up
behind it. PLAN.md §13 + CHANGELOG.md collapse v0.1.2 + v0.1.3 into
the single v0.1.2 entry, dated 2026-05-03; orphaned [0.1.3] link
reference at the foot of CHANGELOG removed.

No code changes — the four fixes from 0cceefc + f132c3c stand as-is.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Marco Grätsch
2026-05-03 16:35:26 +02:00
parent 0cceefc890
commit 8b2fc4dd06
2 changed files with 10 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
CHANGELOG.md
View File

@@ -10,21 +10,16 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/)
- (none yet — next changes land here)
## [0.1.3] — TBD
## [0.1.2] — 2026-05-03
Non-breaking bugfixes from the post-v0.1.2 architecture audit. No behaviour change for existing apps; no public API additions or signature changes.
### Fixed
- **`bridge.qml_path` is now actually configurable.** The `BridgeResourceMaker` and `BridgeWindowMaker` docstrings claimed the QML scaffold path was settable via the bundle's `qml_path` option, but the bundle's `configure()` was empty and the constructor default (`'../qml/'`) was the only knob. `BridgeBundle::configure` now defines a `qml_path` scalar node; `loadExtension` exposes it as the `bridge.qml_path` container parameter; `services.yaml` binds it into both makers. Apps can override with `config/packages/bridge.yaml`: `bridge: { qml_path: ../qml/ }`. Default unchanged.
- **`SessionAuthenticator`: problem+json on the entry-point path.** `onAuthenticationFailure` already returned RFC 7807 `application/problem+json` for *bad-token* requests, but Symfony's default `AuthenticationEntryPointInterface::start` fired for *no-token* requests, returning a Form-flavoured 302/401 with the wrong shape for QML's `RestClient` error mapping. The authenticator now implements `AuthenticationEntryPointInterface` and routes both paths through a shared `problemJson()` helper so QML sees one error shape regardless of which firewall path was taken. New test covers the entry-point response.
- **`CorrelationKeyListener::onTerminate` sub-request guard.** `onRequest` already guarded with `isMainRequest()`, but `onTerminate` cleared unconditionally — a sub-request finishing mid-controller would wipe the main request's correlation key, causing the matching Mercure echo to lose its `correlationKey` field and the optimistic UI to never reconcile. FrankenPHP worker mode does not currently emit sub-requests so the user-visible impact is nil, but the asymmetry was a defensive bug.
## [0.1.2] — TBD
Bugfix release. Bundles the v0.1.1 follow-up that surfaced during the v0.1.2 cycle (bundled-mode supervisor cleanly SIGTERMs its child on host exit) with three non-breaking fixes from a post-v0.1.1 architecture audit.
### Fixed
- **Bundled supervisor: clean child shutdown.** `BackendConnection`'s destructor was the only path that called `teardownChild()`, but it ran during stack unwinding *after* `app.exec()` returned — by then the Qt event loop was already mid-shutdown and `QProcess::waitForFinished` couldn't reliably reap the child. Symptom: Qt logged `QProcess: Destroyed while process ("...frankenphp") is still running`, frankenphp + its PHP workers became orphans. The constructor now also connects `QCoreApplication::aboutToQuit` → `teardownChild`, so the child is SIGTERM'd while the event loop is still active. The bundled-supervisor integration test gained a clean-shutdown assertion (no Qt warning + no orphan frankenphp under the host's PGID after SIGTERM).
- **`bridge.qml_path` is now actually configurable.** The `BridgeResourceMaker` and `BridgeWindowMaker` docstrings claimed the QML scaffold path was settable via the bundle's `qml_path` option, but the bundle's `configure()` was empty and the constructor default (`'../qml/'`) was the only knob. `BridgeBundle::configure` now defines a `qml_path` scalar node; `loadExtension` exposes it as the `bridge.qml_path` container parameter; `services.yaml` binds it into both makers. Apps can override with `config/packages/bridge.yaml`: `bridge: { qml_path: ../qml/ }`. Default unchanged.
- **`SessionAuthenticator`: problem+json on the entry-point path.** `onAuthenticationFailure` already returned RFC 7807 `application/problem+json` for *bad-token* requests, but Symfony's default `AuthenticationEntryPointInterface::start` fired for *no-token* requests, returning a Form-flavoured 302/401 with the wrong shape for QML's `RestClient` error mapping. The authenticator now implements `AuthenticationEntryPointInterface` and routes both paths through a shared `problemJson()` helper so QML sees one error shape regardless of which firewall path was taken. New test covers the entry-point response.
- **`CorrelationKeyListener::onTerminate` sub-request guard.** `onRequest` already guarded with `isMainRequest()`, but `onTerminate` cleared unconditionally — a sub-request finishing mid-controller would wipe the main request's correlation key, causing the matching Mercure echo to lose its `correlationKey` field and the optimistic UI to never reconcile. FrankenPHP worker mode does not currently emit sub-requests so the user-visible impact is nil, but the asymmetry was a defensive bug.
## [0.1.1] — 2026-05-03
@@ -78,8 +73,7 @@ First public preview. Phases 0 through 4a in PLAN.md are complete plus the Phase
- The bundle ships without `composer.lock` (it's a library); the skeleton and the todo example carry their own.
- Licensed under **LGPL-3.0-or-later** (`LICENSE` + `LICENSE.GPL` at the repo root). Chosen to align with Qt 6's LGPLv3 licensing — see PLAN.md §12 for the relinkability obligations the AppImage build already honours.
[Unreleased]: https://src.bundespruefstelle.ch/magdev/php-qml/compare/v0.1.3...HEAD
[0.1.3]: https://src.bundespruefstelle.ch/magdev/php-qml/releases/tag/v0.1.3
[Unreleased]: https://src.bundespruefstelle.ch/magdev/php-qml/compare/v0.1.2...HEAD
[0.1.2]: https://src.bundespruefstelle.ch/magdev/php-qml/releases/tag/v0.1.2
[0.1.1]: https://src.bundespruefstelle.ch/magdev/php-qml/releases/tag/v0.1.1
[0.1.0]: https://src.bundespruefstelle.ch/magdev/php-qml/releases/tag/v0.1.0

11
PLAN.md
View File

@@ -1,6 +1,6 @@
# php-qml — Plan for a Symfony/FrankenPHP + Qt/QML Desktop Framework
> **Status (2026-05):** v0.1.0 + v0.1.1 shipped 2026-05-03 (LGPL-3.0-or-later). v0.1.2 + v0.1.3 in progress on `dev`. Planning is version-based — see §13.
> **Status (2026-05):** v0.1.0 + v0.1.1 + v0.1.2 shipped 2026-05-03 (LGPL-3.0-or-later). Planning is version-based — see §13.
>
> **Where else to look:**
>
@@ -541,14 +541,11 @@ Closed the four shakedown follow-ups identified during v0.1.0 shipping:
- **Caddyfile fmt.** `framework/skeleton/Caddyfile` and `examples/todo/Caddyfile` reformatted per `caddy fmt`; the "Caddyfile input is not formatted" boot warning is gone.
- **Cache-wipe on bundled launch** (added during v0.1.1 shakedown). Symfony bakes `kernel.project_dir` into its compiled cache; the AppImage's FUSE mount path changes per launch, so cache from launch N is stale by N+1. Supervisor now wipes `var/cache/` on every `initBundledMode`. Build-time cache warmup is the v0.2.0 follow-up.
### v0.1.2 — in progress on `dev`
### v0.1.2 — shipped 2026-05-03
Bugfix release. Bundles the v0.1.1 follow-up that surfaced during the cycle (clean child shutdown) with three non-breaking fixes from a post-v0.1.1 architecture audit:
- **Bundled supervisor: clean child shutdown.** The destructor's `teardownChild()` only ran during stack unwinding *after* `app.exec()` returned, by which point Qt's event loop was already mid-shutdown — so `QProcess::waitForFinished` couldn't reliably reap the child and Qt warned `QProcess: Destroyed while process is still running`, leaving an orphan frankenphp + its workers behind. Fix: connect `QCoreApplication::aboutToQuit` to `teardownChild` in the constructor, so the child is SIGTERM'd while the event loop is still active. Bundled-supervisor integration test gained a clean-shutdown assertion (no Qt warning, no orphan frankenphp under the host's PGID after SIGTERM).
### v0.1.3 — in progress on `dev`
Non-breaking bugfixes surfaced by the post-v0.1.2 architecture audit:
- **`bridge.qml_path` is now actually configurable.** `BridgeResourceMaker` and `BridgeWindowMaker` carried docstrings claiming the QML output dir was settable via the bundle's `qml_path` option, but the bundle never wired one — the constructor default was the only knob. `BridgeBundle::configure` now defines a `qml_path` node (default `../qml/`); `loadExtension` exposes it as the `bridge.qml_path` container parameter; `services.yaml` binds it into both makers. Apps configure with `config/packages/bridge.yaml`: `bridge: { qml_path: ../qml/ }`. Default is unchanged so existing skeleton/example apps need no edit.
- **`SessionAuthenticator` problem+json on entry-point path.** `onAuthenticationFailure` already returned RFC 7807 `application/problem+json` for *bad-token* requests, but Symfony's default entry point fired for *no-token* requests — yielding a Form-flavoured 302/401 instead. Implemented `AuthenticationEntryPointInterface::start`, factored the response into a `problemJson()` helper, so QML's RestClient sees one shape regardless of which path the firewall takes. Added test coverage.
- **`CorrelationKeyListener::onTerminate` sub-request guard.** `onRequest` already had `isMainRequest()`, but `onTerminate` cleared unconditionally — so a sub-request finishing mid-controller would wipe the main request's correlation key, causing the matching Mercure echo to lose its `correlationKey` field and the optimistic UI to never reconcile. Defensive: real-world impact is low (FrankenPHP worker mode does not currently emit sub-requests), but cheap to fix and the obvious correctness bug.