model-writtenrun 4986 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Andres Freund proposed smaller relation segments to exercise multi-segment code in regression tests, initially replacing the gigabyte options with explicitly named megabyte options to avoid silently changing existing builds #1.
- Tom Lane argued that even megabyte segments were too large for many regression tables and proposed a separate page-count option while retaining the existing gigabyte interface; Bharath Rupireddy suggested calling the units blocks and documenting the developer-only purpose #2 #4 #5.
- Andres found that segments of five blocks or fewer broke a pg_basebackup corruption test because its checksum-corruption helper did not understand segment boundaries, while six-block segments passed all tests; Michael Paquier endorsed the additional flexibility #6 #7.
- When Andres raised how to display sub-gigabyte sizes, Tom suggested choosing the display units according to the option supplied, which the revised patch adopted #8 #9 #10.
- The revised patch retained gigabyte settings, added a block-count override with precedence warnings and developer documentation, corrected Meson segment-size calculations and reporting, and configured eight-block and six-block segments in the autoconf and Meson CI tasks respectively #10.
- Andrew Dunstan questioned imposing a minimum segment size, but Andres and Tom defended one-block segments as useful for testing; Andrew accepted that reasoning, leaving the known corruption-test limitation without a helper fix in this patch #11 #12 #13 #14 #15.
- Andres committed the revised patch with an additional Meson documentation entry as d3b111e320, then closed the thread by confirming that both small-segment CI configurations remained enabled because their cost was low #10 #16.