Dario Gabriel LipicarandClaude Opus 5
e9b50b7635
The repo was published as logos-plugin-nim, mirroring logos-plugin-qt. That name describes the wrong axis: what this backend actually builds is the Nim half of a VIEW module, so it belongs to the logos-view-module family rather than to a general plugin one. Renamed on GitHub (the old URL keeps redirecting), remote repointed, and the three self-references updated: the README title, the flake description, and the vanilla test's pname. Both checks green. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
logos-view-module-nim
The Nim backend for Logos modules: builds a module's Qt plugin from Nim instead
of C++, as a sibling to logos-plugin-qt.
The output is an ordinary Qt plugin. ui-host loads it with QPluginLoader and
drives it through the meta-object system, with no idea it is not C++ —
nimside supplies the meta object that
moc would otherwise generate, and nim-seaqt
the Qt bindings underneath.
What a module looks like
metadata.json
nim_counter.rep the interface, and its single definition
nim/nim_counter_impl.nim plain Nim procs: the whole hand-written backend
nim/nim_counter_plugin.nim entry point; Q_PLUGIN_METADATA and nothing else
qml/Main.qml
rep2nim (shipped in nimside) generates the qobject: block from the .rep,
and logos_replica_factory() (in logos-plugin-qt) builds the typed replica QML
talks to from the same .rep — so the two sides cannot drift.
See repos/logos-nim-ui-example for a complete one.
Two shapes
With a .rep you get the stock typed replica: from QML the module is
indistinguishable from a C++ one. This is the recommended shape.
Without one, ui-host dynamic-remotes the plugin and LogosQmlBridge
acquires a dynamic replica. No C++ is involved anywhere, at the cost of QML
seeing a runtime-built meta object.
Why the patches
patches/nimside/ is a git format-patch series against the pinned nimside
revision, applied in lib/nimside.nix. Keeping it as patches rather than a
vendored fork means the delta stays reviewable and upstreamable. It currently
carries three changes:
- property accessors are no longer published as slots, matching
moc— they were showing up on QtRemoteObjects replicas and in QML's property cache; qproperty(write = true, notify = true)no longer reads asfalse— the pragma is parsed untyped, sotruearrives as an identifier andboolValquietly returned false for it, silently dropping the setter and the notify signal;rep2nimitself.
Platform notes
nim-seaqt locates Qt through pkg-config, and on nixpkgs the .pc files live
in the main output rather than -dev. Two flags are also needed that
nim-seaqt's own docs do not mention, both set by lib/common.nix:
--passC:-F<qtbase>/lib— nixpkgs builds Qt as frameworks andpkg-config --cflagsemits only-I, so<QtCore/qglobal.h>does not resolve when Nim compiles the generated C++ shims;--passL:-lc++— Nim drives the link withclang, notclang++, even though those shims are C++.