Repository navigation
chore: add back tck modules - #861
Conversation
Copy the generated code from a2a-tck/sut/a2a-java Signed-off-by: Jeff Mesnil <jmesnil@ibm.com>
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request introduces a new TCK (Technology Compatibility Kit) module to the Java SDK, providing a System Under Test (SUT) implementation. It includes a Quarkus-based application with CDI producers for the agent card and executor, handling various TCK scenarios such as streaming artifacts and task lifecycles. Feedback suggests ensuring consistent artifact IDs when streaming chunked data to prevent the creation of separate artifacts and using the emitter's reject method instead of throwing an error to maintain correct task state transitions.
There was a problem hiding this comment.
When append is set to true, the artifactId must be consistent across calls to indicate that the chunks belong to the same artifact. Passing null causes AgentEmitter to generate a new random UUID for each call, which results in the creation of two separate artifacts instead of appending to one. This will likely cause the TCK test for chunked artifacts to fail.
| if (messageId.startsWith("tck-stream-artifact-chunked")) { | |
| emitter.startWork(); | |
| emitter.addArtifact(List.of(new TextPart("chunk-1 ")), null, null, null, true, false); | |
| emitter.addArtifact(List.of(new TextPart("chunk-2")), null, null, null, true, true); | |
| emitter.complete(); | |
| return; | |
| } | |
| if (messageId.startsWith("tck-stream-artifact-chunked")) { | |
| emitter.startWork(); | |
| String artifactId = "tck-chunked-artifact"; | |
| emitter.addArtifact(List.of(new TextPart("chunk-1 ")), artifactId, null, null, true, false); | |
| emitter.addArtifact(List.of(new TextPart("chunk-2")), artifactId, null, null, true, true); | |
| emitter.complete(); | |
| return; | |
| } |
| if (messageId.startsWith("tck-reject-task")) { | ||
| throw new A2AError(-1, "rejected", null); | ||
| } |
There was a problem hiding this comment.
For a task rejection scenario, it is more appropriate to use emitter.reject() to transition the task to the REJECTED state. Throwing an A2AError is intended for exceptional conditions and will typically result in a FAILED state, which may not match the expectations of the tck-reject-task test case. Additionally, the AgentExecutor documentation recommends using the emitter for business logic outcomes.
| if (messageId.startsWith("tck-reject-task")) { | |
| throw new A2AError(-1, "rejected", null); | |
| } | |
| if (messageId.startsWith("tck-reject-task")) { | |
| emitter.reject(A2A.toAgentMessage("rejected")); | |
| return; | |
| } |
References
- Adhere to the A2A protocol specification for error code mappings and state transitions, especially in TCK implementations.
Copy the generated code from a2a-tck/sut/a2a-java Signed-off-by: Jeff Mesnil <jmesnil@ibm.com>
2 participants
Copy the generated code from a2a-tck/sut/a2a-java