model-writtenrun 7692 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Alexey Bashtanov proposed bit_xor aggregates for four input types in #1, motivated by fast, simple checksums over large unordered sets where collision resistance was less important.
- Kyotaro Horiguchi noted that users could already define these aggregates with CREATE AGGREGATE but considered built-in support consistent with bit_and and bit_or in #2; Peter Eisentraut supported the checksum use case and requested documentation explaining it in #3.
- Ibrar Ahmed removed an unrelated OpenSSL configuration-comment change that prevented the patch from applying in #4, and Alexey refreshed the patch against master and added the requested checksum documentation in #5.
- Peter Eisentraut committed #5 as f9a0392e1c, with light documentation edits and a catalog-version bump, and announced completion in #6.
- After the commit, David Fetter proposed requiring an unambiguous ORDER BY in #7, but withdrew the objection in #11 after Vik Fearing pointed out that XOR does not depend on input order in #10.
- The discussion broadened into safeguards for order-sensitive aggregates: Pavel Stehule advocated mandatory ordering where results depend on it in #18, while Vik Fearing objected to imposing it on array_agg in #19; Pavel acknowledged compatibility constraints and called the debate off-topic in #22.
- Tom Lane argued that existing documentation sufficiently addresses order-sensitive aggregates in #23, and David Rowley closed the thread with a floating-point sum example and objections to mandatory sorting in #24; bit_xor remained committed, with no further patch or unresolved implementation work in this thread.