On Fri, Aug 28, 2026 at 12:04:14PM +0200, Jan Nidzwetzki wrote: > Thanks for the updated patch. Version 5 looks great to me.
Hmm. After studying the patch, I have finished by removing a total of around 200 lines of comments that felt overly talkative (clearly AI generated to me), making the whole leaner by retaining the main information we care about. Same thing has been done in the tests and the C code. I also did not understand the need to force installcheck and _PG_init() requiring to be loaded during shmem setup. The extensible node and custom scan APIs are fine as long as we load a library, something done after calling one function or just creating the extension, so I have removed the s_p_l part and simplified things. The reloption approach would indeed do not work. I forgot that we do not have an anchor for table AMs. We should, actually.. That's annoying to lack this piece of facility while we have these APIs, perhaps Andrew Dunstan's recent work in this area would work, but I did not check it. I still have a small gripe about the requirement of a hardcoded relation name to trigger a custom scan, but at the end, discarded that as being an issue (cannot get excited about extending that beyond non-base relations and such). The module is still useful as a base template and to provide coverage for the extensible node and custom scan APIs. After a few more adjustments, the whole result was looking fine, so applied. -- Michael
signature.asc
Description: PGP signature
