This repo had no CI at all. Both open PRs (#20, #21) report "no checks
reported on the branch", so every green result claimed for the QML bridge
changes came from a local run on one machine, on one platform -- which is
how several defects in that change set shipped green in the first place.
The suite is already a nix derivation whose checkPhase runs ctest over
every test target (nix/test.nix), so building the check IS running the
tests; there is no separate run step to keep in sync. Both platforms run
because the bridge coordinates processes and sockets, and the two OSes
have already disagreed about exactly that elsewhere in the stack.
Modelled on logos-protocol's workflow, including the DeterminateSystems
installer (cachix/install-nix-action trips over pre-existing nix build
users on the macOS runners) and a non-fatal cachix push.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Moves this repo's own logos-protocol pin from 1e96004 (Jul 17) to
0f26ffd (Aug 6), the current protocol master.
ui-host statically links logos-protocol, so on macOS an ui-host built
against a stale protocol and a ui_qml plugin built against current
protocol get their weak definitions coalesced by dyld — one image's
protocol code silently binds to the other's, across a real ABI change.
Pinning current protocol here is the fix; no `follows` override needed.
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
serializeResultForTesting used QJsonValue::fromVariant, which cannot
convert the custom LogosResult metatype and silently degraded a
`result`-type return to the literal "null". A QML-only ui_qml plugin
calling a result-returning method through logos.callModule therefore
received null instead of {success, value, error}.
Replace the bespoke serialization with logos-protocol's canonical
logos::qvariantToNlohmann — the same converter every transport
(lp/std/cdylib) already uses. Now ALL types round-trip to QML/JS
identically: scalars as bare literals, containers preserved, integers
kept as integers (QJsonValue::fromVariant degraded them to double),
bytes in the tagged {"_bytes":...} form, and LogosResult as
{success, value, error} (empty error -> null). Fixes both sync
callModule and async callModuleAsync.
Adds test_logos_qml_bridge_result.cpp (regression) plus the
reproduction/isolation harnesses (e2e, gui, handshake, 2-process) built
while proving the long-suspected QML "sequential-call stall" is NOT an
SDK defect: every layer round-trips correctly under isolation, and the
only real bug was this result serialization.
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The ViewModuleHost QtRO peer linked the pre-concurrent-dispatch protocol, so a
c6234940 UI plugin's capability-token handshake was rejected inside it. Align
protocol + SDKs to current master. No source changes (signature-stable API).
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>
The out-of-process ui-host (spawned by Basecamp / any app embedding the view
runtime) inherited the parent's process group and had no parent-death cleanup,
so a parent crash could (a) leak a teardown signal into the parent's process
group and (b) leave the host running as an orphan. Mirror logos_host: setsid()
to lead our own group (parent stays foreground/manageable), plus
PR_SET_PDEATHSIG (Linux) + a portable getppid() watchdog so we exit when the
parent dies. Compare against the parent's actual pid (not pid 1) so a parent
that is itself PID 1 (a container) is handled correctly.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* Retarget to the qt-split SDK stack (logos-qt-sdk + logos-protocol)
logos-cpp-sdk::logos_sdk no longer exists after the Qt split: the Qt
developer layer (LogosAPI, provider objects) moved to logos-qt-sdk and
the transport/consumer core into the logos-protocol library. Both
targets (the static runtime lib and ui-host) now link
logos-qt-sdk::logos_qt_sdk (which carries logos-protocol transitively)
plus the header-only logos-cpp-sdk::logos_headers.
flake.nix gains logos-protocol/logos-qt-sdk inputs threaded through
nix/{default,test}.nix; the lock pins the extraction-chain branch revs
(temporary — re-lock against masters when the chain merges).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* lock: bump cpp-sdk to the qt-free branch head with the protocol lock fix
This flake declares no follows between logos-cpp-sdk and logos-protocol,
so cpp-sdk's OWN lock governs its nested protocol; b86ff877 still pinned
the P1 protocol rev, which no longer compiles (LogosProviderPlugin moved
in the qt split). 7272c73 carries the corrected pin. Temporary — drop
when the chain merges.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* fix(tests): retarget the unit-test link to the qt-split SDK targets
tests/CMakeLists.txt still linked bare logos_sdk (only built with
LOGOS_VIEW_RUNTIME_BUILD_TESTS=ON, so the package build verified earlier
missed it). The runtime lib's PUBLIC link interface already carries the
qt-sdk/protocol targets; the explicit entries replace the dead archive.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* lock: qt-split chain fully merged — pins advance to masters (cpp-sdk 87abcd8, protocol 9de4165, qt-sdk 8d0e5d9)
---------
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Picks up logos-co/logos-cpp-sdk#68 — marshals provider event emission onto the
remoting source thread. Part of propagating the fix to the host runtime.
