model-writtenrun 2314 - may be wrong, links go to the source messages. Commit-by-commit LLM analysis is in the Commits dialog.
- The thread began with Tomas Vondra announcing basic Linux NUMA support, authored by Jakub Wartak and Bertrand Drouvot, with an optional libnuma dependency, a SQL availability check, and batched memory-page queries using move_pages #1.
- Daniel Gustafsson reported a build failure caused by pg_numa.c including backend headers while being compiled for frontend use; Andres Freund reproduced it with unlimited parallelism, and he and Tom Lane agreed to move the SQL and page-size functions out so the portability code could use c.h #2 #3 #4 #5 #6.
- Wolfgang Walther traced a separate suspicious ninja warning to dogfish actually using the samurai replacement; Andres accepted keeping that build variation, and Wolfgang added a buildfarm note identifying it #7 #8 #9.
- Daniel then supplied the thread’s only attached patch to restore omitted autoheader output in pg_config.h.in, adding the unused HAVE_LIBNUMA macro and correcting the USE_LIBNUMA comment without changing runtime behavior #10.
- Jacob Champion confirmed that the patch matched his independently generated autoreconf changes, and Tomas Vondra also approved it #11 #12.
- Daniel committed the attachment unchanged as 1fd3566ebc and confirmed the push in the final reply, completing the autoheader cleanup; this thread does not show implementation of the earlier proposed frontend/backend code separation #13.