model-writtenrun 5990 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Mark Dilger proposed cheap TOAST-pointer checks for excessive raw size, apparent compression expansion and invalid compression methods, explicitly excluding retrieval and decompression because of their performance cost #1.
- Justin Pryzby objected that the oversized-attribute comment misdescribed TOAST's purpose #2; Mark corrected it #3, and Aleksander Alekseev approved the patch while clarifying that his failed review checkboxes were accidental #4 #5.
- Heikki Linnakangas questioned the size limit #6, and Mark explained that VARLENA_SIZE_LIMIT equals MaxAllocSize and that impossible size relationships indicate corruption regardless of whether detoasting is attempted #7 #8.
- Greg Stark considered the patch committable but questioned the constant's documentation and compression-method switch #9; Mark explained that omitting a default catches future enum additions through compiler warnings and deferred broader TOAST-chunk diagnostics as a larger project #10.
- Greg accepted the switch rationale and requested an explanatory comment #11, which Mark proposed inline #12; however, the next attachment remained identical to the earlier patch and omitted those expanded comments #13.
- Robert Haas requested tests #14, and Mark added two raw-size corruption cases while leaving compressed-data corruption tests out because he could not make them compact, deterministic and portable #15.
- Robert committed the final attachment unchanged as bd807be693 and announced its landing #16.
- Tom Lane reported unstable buildfarm output #17, and Mark's investigation confirmed that the tests themselves appeared faulty in the thread's last message #18.
- Robert subsequently removed both new tests in ccf289745d pending investigation, leaving the TOAST-pointer checks committed; the supplied thread contains no repaired tests or resolved explanation of their instability.