morningman opened a new pull request, #70720:
URL: https://github.com/apache/doris/pull/70720

   ### Background
   
   Reviewer assignment collects authoritative PR data in Python, asks Codex to 
choose modules, then samples candidates and applies recoverable GitHub 
mutations. The previous prompt supplied file names, so classification depended 
on the model choosing to read them. CI returned an empty result for #70665 
despite a populated Iceberg pool.
   
   ### What problem does this PR solve?
   
   Issue Number: None
   Related PR: #70695 (introduced assignment), #70712 (permissions and initial 
diagnostics), #70665 (reproduction PR)
   PR Type: Build/CI/Chore
   
   Problem Summary:
   
   The [CI run for 
#70665](https://github.com/apache/doris/actions/runs/44268326033/job/134678307977)
 returned `{"modules":[]}` with five eligible Iceberg candidates. Build the 
complete classifier input in Python and supply it through stdin, leaving only 
module classification to the model. Add shared local replay, richer 
diagnostics, offline branch checks and a read-only hosted dispatch mode so 
changes can be verified before merge.
   
   ### Details
   
   - **Change map:** Ten files: +382/-51 for helpers and workflows, +219/-31 
for tests, and +91/-23 for documentation. The input builder/runner removes 
model-side data gathering; validation rejects empty results with eligible path 
candidates; replay and dispatch provide verification paths; diagnostics retain 
input fingerprints, decisions and execution failures.
   - **Affected:** Reviewer assignment on Ubuntu 24.04 and local Linux/macOS 
replay with Python 3.9+, GitHub CLI and the pinned Codex runtime.
   - **Developer and release impact:** Python supplies PR metadata, complete 
diff, change statistics, routing instructions and all module pools. Codex runs 
in an empty directory with shell, apps, plugins, web search and subagents 
disabled, returning module IDs and a brief classification reason. Input hashes 
are checked before invocation. New empty classifications with eligible matched 
pools fail without mutating the PR; valid manual fallback and saved checkpoints 
remain supported. The RNG and GitHub assignment sequence remain in Python. 
Manual dispatch runs selection only; PR/push checks use offline fixtures 
without secrets.
   - **License and third-party:** None; no new runtime dependency or change to 
the pinned Codex package.
   - **Rollback:** Revert this PR. No cache, secret or data migration is needed.
   - **CI permissions and secrets:** The selection job remains read-only. Both 
the workflow condition and Python entry point prohibit assignment for manual 
events. Local replay rejects non-GET GitHub calls. Diagnostic artifacts retain 
hashes, versions, classification reason, validation status and filtered 
command/MCP/errors, while excluding credentials, raw reasoning and complete 
model sessions.
   
   ### Testing
   
   - [Ubuntu 24.04 branch 
checks](https://github.com/apache/doris/actions/runs/44275079517) passed all 
120 reviewer-assignment tests and all 16 GitHub API diagnostic tests, including 
the real-gh loopback fixture.
   - [Read-only hosted branch 
dispatch](https://github.com/apache/doris/actions/runs/44275109656) passed 
before merge using the pinned Linux CLI and existing auth pool. For #70665 it 
selected `connector.iceberg` and `optimizer`, emitted a reason and produced two 
eligible reviewers in about five seconds. The assignment job was skipped and 
diagnostics/usage/plan artifacts were uploaded successfully.
   - A real local CLI 0.166.0 replay using live GitHub reads selected the same 
modules. Local macOS and hosted Ubuntu produced identical 78,279-byte prompts 
(SHA256 `7cab4e99f4df436032fa195a490b5a4f3aafc0510788fa8c498054ce8f725179`) and 
matching implementation fingerprints. Both runs recorded no shell/MCP calls or 
model errors.
   - Offline fixtures cover complete inline input, hash reproducibility and 
altered-input rejection, credentials, model failure/timeout, empty-output 
rejection, saved checkpoints, local replay and manual-run mutation prevention. 
Local Docker's image store currently reports I/O errors; the hosted branch run 
supplies the real Linux/auth/artifact verification instead.
   
   ### Release note
   
   None
   
   ### Check List (For Author)
   
   - Behavior changed: No
   - Does this need documentation: No
   - Pre-submission status: Ready
   
   ### AI Disclosure
   
   - Tool: Codex (GPT-6).
   - Generated: Implementation, tests, workflow updates, documentation and this 
description.
   - Human review: Not stated.
   


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to