model-writtenrun 9241 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andrey Borodin proposed explicit corruption error codes for TOAST and B-tree failures so automated monitoring would not need to interpret diagnostic text, attaching the initial patch #1.
- Alvaro Herrera suggested separately identifying permanent system failures #2; Andrey questioned conflating hardware failures with corruption #3, while Michael Paquier favored case-by-case improvements using existing error categories, identified indentation problems, and considered the patch unsuitable for backpatching because it was not a bug fix #4 #6.
- Peter Eisentraut raised whether the converted messages should remain untranslated through errmsg_internal() #7, and Peter Geoghegan supported that choice because expert diagnostics are most useful as searchable English strings #9.
- Andrey reported additional corruption diagnostics observed with a faulty kernel page cache #8, then expanded the patch to cover five more cases while retaining errmsg(); he excluded the block-read failure because errcode_for_file_access() already classified it appropriately and deferred broader page-integrity mechanisms to another discussion #10.
- Peter Eisentraut committed the expanded patch as fd6ec93bf8, preserving its scope while switching to errmsg_internal(), repairing compilation errors, correcting one B-tree error code, and adjusting formatting; his announcement noted that the submitted patch did not compile #11.
- The thread ended with the change landed and Andrey acknowledging that his check-world run had missed unsaved edits, as well as thanking Peter for correcting ERRCODE_DATA_CORRUPTED to ERRCODE_INDEX_CORRUPTED #12.