model-writtenrun 1233 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Chao Li reported that LZ4 decompression advanced its output pointer by cumulative bytes rather than the current iteration's output, supplied a fix, and demonstrated failure followed by successful backup verification using an artificial output-size restriction #1.
- Xuneng Zhou agreed that the cumulative advancement could leave holes in the output buffer and approved the proposed fix #3.
- Tom Lane committed the pointer fix as 4548e87466, reorganizing the buffer updates and comments; he reported that he could not reproduce the crash but considered the defect clear #4.
- While reviewing the code, Tom Lane identified a separate problem: pg_log_error only logged compression failures and allowed execution to continue, prompting his proposal to use pg_fatal where recovery was absent #4.
- Tom Lane submitted a broader cleanup covering streamer error exits and pg_dump LZ4 close-result handling, while deliberately leaving apparently intentional nonfatal cases alone #5.
- Chao Li subsequently reproduced the original failure on an old branch and verified that the updated master succeeded with the same reproduction hack #6.
- At the thread's final message, Chao Li approved the error-handling cleanup and reported finding no additional similar sites; the pointer fix had landed and the cleanup awaited commitment #7.
- The subsequent commit 9a42888a32 landed the streamer fatal-error conversions from #5, but its diff excludes the proposed pg_dump close-result changes and Zstandard long-distance-mode cleanup; the thread provides no explanation or final disposition for those omitted changes.