model-writtenrun 8341 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Odin Ugedal proposed huge_page_size in #1 so PostgreSQL could select preallocated huge-page sizes independently of the system default, particularly for Kubernetes and benchmarking, with zero preserving existing behavior.
- Thomas Munro suggested architecture-specific documentation and identified Windows support as possible future work in #2; Odin incorporated the wording and limited explicit mmap size flags to nondefault sizes in #3.
- Thomas requested capability detection and rejection of nonzero settings on unsupported platforms in #4; Odin added configure checks and a GUC validation callback in #6, confirming that huge_pages=try still falls back to ordinary pages.
- Thomas reported documentation and Windows build failures, an incorrect allocation-size error hint, and benchmark results in #7; Odin fixed the three concrete issues in #8, while both left reclaiming wasted space from large-page rounding for separate work.
- Andres Freund questioned the benchmark's TLB assumptions and suggested investigating worker scheduling and mixed page sizes in #9, without a subsequent implementation of those ideas in this thread.
- Thomas committed #8 as d2bddc2500 after replacing configure checks with direct preprocessor checks and editing the documentation; his closing message #10 confirms the feature landed and leaves broader portability and documentation cleanup for future work.