model-writtenrun 9390 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Geoghegan reported that long remains 32 bits on 64-bit Windows, allowing B-tree index-build progress counters to overflow, and subsequently identified the same bug in hash index builds #1 #5.
- Andres Freund advocated eliminating long and preventing its reintroduction, while Tom Lane explained that memory-sizing conventions and translatable integer formats made a blanket replacement difficult; Peter proposed addressing simpler cases first #4 #7 #10 #11.
- Alvaro Herrera submitted a focused patch widening both index-build counters to uint64, having checked progress-reporting callers for similar mistakes #12.
- Peter Geoghegan recommended int64, while David Rowley favored uint64 for consistency with CopyFrom; Alvaro chose int64 to match the progress-reporting API and committed the adjusted patch as 9a83afecb7 #14 #16 #17.
- Peter Eisentraut reopened the formatting question in May, suggesting that PostgreSQL's own snprintf could help, but identified non-GNU gettext compatibility concerns; Tom Lane rejected inventing a format modifier that compiler checks would not understand #19 #20 #22.
- Peter Eisentraut proposed defining int64 as long long everywhere, then suggested using long long directly where storage layouts were unaffected; Robert Haas argued that widely shared values should retain explicit widths to avoid later storage or protocol mistakes #25 #27 #28 #32.
- Tom Lane offered a narrower solution: keep int64 values and cast them to long long when printing with %lld, relying on C99's minimum width guarantee #30.
- Peter Eisentraut submitted a new two-patch series combining explicit casts and selected long long declarations to remove temporary formatting buffers, plus a separate ECPG cleanup using %zu directly for size_t #31.
- Tom Lane reviewed the series, considered preserving all int64 declarations but accepted the submitted choices, and marked it ready for committer #33.
- Peter Eisentraut announced committing the work; 6a1cd8b923 reproduces the six-file formatting patch from #31 unchanged, while the companion ECPG patch is not represented by a supplied candidate #34.
- The thread ended with further discussion of supported gettext implementations: Peter Eisentraut said Solaris gettext remained supported in theory, with past compatibility complaints, while no blanket elimination of long or universal integer-type policy was implemented in the supplied patches #35 #36.