zmuxuny opened a new pull request, #5363:
URL: https://github.com/apache/rocketmq-dashboard/pull/5363

   ## Summary
   
   Refs #2040 and the stale-closed, unmerged proposals #2042 / #2058.
   
   Apply the existing 5 MiB CLI output budget to Claude streaming stdout before 
UTF-8 decoding and JSON line parsing. The aggregate per-run limit also bounds 
tool arguments accumulated across small frames. A producer exceeding the budget 
is terminated promptly and the caller receives the existing 
`llm.provider.output_too_large` error.
   
   Both hosted event streaming and the text-only path use the same reader. 
Exactly-at-limit output, UTF-8 text, and a final frame without a newline remain 
supported. Stderr keeps its existing separate 64 KiB diagnostic cap. README 
documentation is updated in English and Chinese.
   
   ## Verification
   
   - Fail-before on unchanged production code at `6a68042f`: the new 
oversized-frame and aggregate-frame regressions both failed because no 
exception was raised; the exact UTF-8 boundary case passed
   - Final focused backend matrix: **303 tests passed**, 0 
failures/errors/skips: 
`ClaudeCodeAgentProviderTest,ClaudeCodeStreamParserTest,CliAgentProviderTest,AgentEventProjectorTest,AiEventContractTest,AgentRunHandleStopTest`
   - Six new cases cover oversized single frames, many small frames, exact 
UTF-8 byte boundary, one-byte-over UTF-8 boundary, early termination of a 
blocked producer, and hosted tool-argument accumulation
   - Maven compiled production and test sources; Checkstyle reported 0 
violations; `git diff --check` passed
   - Independent code review completed
   
   Local runner: Temurin 21, Maven 3.9.11, one test fork, 512 MiB heaps. 
Mockito was attached at JVM startup because this runner does not allow its 
normal self-attachment. The full backend integration suite, live Claude 
service, frontend and Docker builds were not run.
   
   ## Scope
   
   One bounded-output fix, with no dependency or configuration change. It 
terminates the direct producer; this does not introduce a new process-tree 
lifecycle guarantee.
   
   AI assistance was used for implementation, regression tests and review.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to