Implements a unified action system that serves as a single source of truth
for all operations in Yaak (Tauri app, CLI, plugins, deep links, MCP server).
Key features:
- ActionExecutor: Combined registry and execution engine with async RwLock
- ActionHandler: Trait-based handlers using async closures
- Context system: RequiredContext and CurrentContext for action availability
- Action groups: Organize related actions
- TypeScript bindings: Auto-generated via ts-rs for frontend use
Design highlights:
- Handlers are closures (no dependencies on other yaak crates)
- Registration requires both metadata and handler (prevents orphan actions)
- Flexible return values via serde_json::Value
- All methods are async using tokio
All 33 tests passing. Ready for integration with yaak-core and yaak-app.
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
The skip_cache flag in services() called reflect(), but reflect() had its
own cache check that returned early. Simplified by removing skip_cache and
always invalidating the pool in cmd_grpc_reflect, since that command is
only called when fresh schema is needed.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Events from previous WebSocket/gRPC connections and HTTP responses were
persisting in the store and displaying in new connections. Added filter
parameter to mergeModelsInStore that clears old events when switching
connections, plus render-time filtering as a safety net.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add new_xplatform_command() helper in yaak-common that creates a
tokio::process::Command with CREATE_NO_WINDOW flag set on Windows.
Also converts git commands to async for consistency.
Events stream in via model_write listener while also being fetched
from the database. If the DB fetch completed before all events were
persisted, replaceModelsInStore would wipe out events that came in
via model_write.
Added mergeModelsInStore that adds fetched events without removing
existing ones. Applied to HTTP, gRPC, and WebSocket event hooks.
Previously, when a gRPC streaming message failed to deserialize (e.g., wrong
type like int instead of string), the error was silently logged and the message
was dropped. Now errors are surfaced to the UI as GrpcEventType::Error events.
Changed the streaming/client_streaming methods to accept an on_message callback
that handles both success (logs ClientMessage) and error (logs Error) cases,
rather than logging the client message prematurely before deserialization.
The gRPC streaming code was using tokio::runtime::Handle::current().block_on()
inside filter_map closures, which caused a panic ('Cannot start a runtime from
within a runtime') when called from an async context.
Fixed by replacing the pattern with .then(async move { ... }).filter_map(|x| x)
which properly handles async operations in stream pipelines.
This fixes the gRPC Ping/Pong freeze issue and restores request cancellation.
