model-writtenrun 3342 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- jian he reported that EXPLAIN JSON used bytes for planning memory but kB for serialization output, asking whether non-text serialization should instead use bytes #1.
- Daniel Gustafsson supported consistent, documented units #2, while Michael Paquier proposed explicit unit metadata as a possible PostgreSQL 18 improvement rather than a PostgreSQL 17 change #3.
- David Rowley proposed preserving EXPLAIN’s established round-up convention through a shared conversion macro and changing MEMORY output to kB #4; his first patch also replaced a potentially narrow long in hash memory reporting and updated regression expectations #5.
- jian he found that the first patch left non-text MEMORY output in bytes #6, prompting David’s revision to share converted values across output formats and use unsigned hash memory reporting #7.
- Michael Paquier endorsed kB and consistent rounding while reiterating the documentation gap #8, and Tom Lane agreed to preserve existing round-up behavior despite preferring round-to-nearest in a new design #9.
- jian he questioned apparently duplicated planning and serialization buffer counters #10 and sought a nonzero serialization example #12; David explained that the counters measure independent work #11 and demonstrated serialization buffer accesses caused by detoasting #13.
- David committed #7 with only variable-name and whitespace cleanup as 0de37b5106 and confirmed the push in the final message #14; unit documentation and explicit unit metadata remained outside the committed patch.