mirror of
https://github.com/mudler/LocalAI.git
synced 2026-05-31 12:07:45 -04:00
c778ad0f6d
The existing unit tests in node_header_test.go drive the wrapper by calling mw(handler)(c) directly against a hand-built echo.Context. That misses regressions where the contract between the real Echo router and the wrapper breaks (e.g. middleware not on the writer chain when the handler runs, or a handler that writes via some surface the wrapper cannot see). Add one integration spec that dispatches a real HTTP request through e.ServeHTTP into a streaming handler shaped like chat.go's streaming branch: set SSE headers, write chunks, Flush. Asserts the X-LocalAI-Node header is on the response map BEFORE the first underlying write commits via an order recorder installed BELOW the wrapper in the writer chain (via a preceding middleware so the wrapper installed by ExposeNodeHeader wraps the recorder). Documents what is NOT exercised: ChatEndpoint and processStream are not wired end-to-end (ChatEndpoint depends on templates.Evaluator, MCP NATS, the LocalAI Assistant holder; processStream lives in core/http/endpoints/openai which imports core/http/middleware, so a regular import would create a cycle). processStream is covered separately in core/http/endpoints/openai/chat_stream_usage_test.go. Signed-off-by: Ettore Di Giacinto <mudler@localai.io> Assisted-by: Claude:claude-opus-4-7[1m]