Repository navigation
Conversation
|
Can you share the search you used to find these errors? I'm unable to reproduce the issue this PR is meant to solve in a live search. |
|
Hi @kirkwat, thanks for reviewing! I identified this through offline tests of the Glassdoor detail-response handling, not from a particular live search. The regression scenario contains a malformed detail alias between valid entries. The existing processing can lose valid details following the malformed entry. The tests also cover aliases that do not belong to the requested IDs. The fix validates aliases individually while preserving the existing batching and transport-failure behavior. The six regression methods in tests/test_oss03_b3.py pass with the patch. This is a defensive improvement for malformed detail responses; I haven’t observed this exact case in live traffic. Happy to provide a smaller example if useful. |
One malformed detail alias interrupts valid aliases later in the response, and an unrequested alias can attach details to an unrelated ID.
Validate the existing small detail shape per alias, ignore unrequested IDs and keep valid neighbors and subsequent batches. Missing/null descriptions remain optional. Use local checks rather than adding detail model classes.
Validation on public main
8dd83cdd194b6e27f6487eb09b2b2a10117d9a1c:python -m unittest discover -s tests -p test_oss03_b3.py -vruff check jobspy tests,ruff format --check jobspy tests, full configuredmypyandgit diff --checkpass with the upstream-declared MCP extra and development dependencies installed.Transport or envelope failure still retains earlier batches and stops later requests as before. The batch size remains 25; no retry is introduced. This patch is independent of optional-field and identity model changes.
This is a narrowly scoped fix in the existing implementation, without architectural changes, new scrapers, dependencies, global models or workflows. No real acquisition was performed. The branch starts directly from public main and includes no private history or audit artifacts. Other contribution groups and published PR #426 are not included.