model-writtenrun 8477 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- Peter Eisentraut proposed annotating allocation functions with alloc_size to enable compiler bounds diagnostics, accompanied by fixes that split the varlena union and ensure sufficient allocation for pg_trgm, numeric, and bootstrap structures #1.
- Zsolt Parragi spotted a stale va_4byte member access in the big-endian compressed-varlena macro after the union was split #2.
- Tristan Partin suggested reusable flexible-array allocation helpers and supplied an independent pahole-based TAP test for catalog trailing padding; he tentatively favored explicit placeholder fields over separate on-disk and in-memory structures but did not implement either solution #3.
- Peter acknowledged the untested big-endian mistake #4 and corrected it in the revised series, which also introduced WIP flex_size and allocation wrappers that cover trailing padding and converted additional variable-sized allocations that compiler warnings had missed #5.
- Peter kept the bootstrap fix focused on allocating sizeof(FormData_pg_attribute), matching tupdesc.c, and deferred broader catalog-padding issues; he suggested generated static assertions as a simpler, more portable alternative to the proposed pahole test, while noting that resolving their failures could be substantial work #5.
- Tristan supported the helper concept and the non-WIP patches, requested overflow-safe arithmetic, suggested an empty-flex-array convenience macro, and agreed to separate the broader padding discussion; the series remained under review with another committer's opinion requested and no linked commit #6.