model-writtenrun 4777 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Tom Lane proposed initdb -c NAME=VALUE and --set in #1 so installations that cannot start with default server settings could initialize successfully; overrides applied during probes and were initially persisted in postgresql.auto.conf.
- Andres Freund examined probe precedence, suggested testing override priority, and raised backpatching in #2; Tom confirmed that user settings must override probe command-line values in #3, but the thread did not settle backpatching.
- Peter Eisentraut, Isaac Morland, and Robert Haas argued for storing permanent initialization settings in postgresql.conf in #4, #5, and #6; Peter also reported that enabling track_commit_timestamp during bootstrap triggered an assertion.
- Robert proposed replacing configuration values by parameter name while retaining adjacent comments in #8, and Tom agreed to implement that approach in #9; Tom rejected David G. Johnston's appended-settings section from #10 because settings should remain near their documentation, reserving append-at-end behavior for unmatched custom parameters in #11.
- Tom's revised patch in #12 introduced the configuration editor, wrote overrides into postgresql.conf, and converted existing substitutions to avoid dependence on exact sample-file values.
- Peter endorsed the revised patch and suggested include and port cleanups in #14; Tom incorporated those suggestions and small output-formatting cleanups when committing #12 as 3e51b278db.
- While testing the feature, Tom found roughly a megabyte of leaked string-array data and independently changed the helpers to modify arrays in place in 4fe2aa7656, explaining the cleanup in #16.
- Andres reported that initdb --wal-segsize 64 now failed a min_wal_size check in #17; Tom identified the mistaken treatment of WAL-size settings as compile-time constants in #18 and restored active assignments in b48af6d174.
- At the thread's end, the feature and both follow-up changes had landed, while Peter's boolean-parameter testing in #19 left track_commit_timestamp as an outstanding bootstrap bug, found the expected failure for default_transaction_read_only, and left broader nonboolean testing open.