Re: Refactor code around GUC default_toast_compression

On Fri, May 01, 2026 at 02:44:07PM +0530, Ayush Tiwari wrote:
> 1. The patch includes an unrelated hunk in
> doc/src/sgml/ref/alter_index.sgml, adding text about `ALTER INDEX ...
> ATTACH PARTITION`. That looks like an accidental carry-over from another
> patch and shouldn't be there ig.

Sorry about that. That feels like a rebase fart.

> 2. The comment in src/include/access/toast_compression.h describing
> default_toast_compression looks stale after this change? It still says
> that the GUC value is one of the char values stored in
> pg_attribute.attcompression, but the patch changes it to use the new
> ToastCompressionGucValue enum values instead.(Maybe I'm
> missing something)

Nope, you are missing nothing. I was re-reading the patch and I think
that we could just remove the whole paragraph. Even by doing so we
lose no information.

> 3. One minor point: CompressionIdToMethod() seems to be added as a public
> helper, but I could not find any callers in this patch.

Oops, removed. I may have used it at some point.

> Also,
> pg_column_compression() still keeps its own cmid-to-name switch. If the
> intent is to centralize these mappings in the registry, perhaps that code
> could use the new helper path as well, otherwise the unused helper may
> not be necessary yet (though it might be in future).

Yes, this is part of the extra tweaks that would be needed when added
a new compression method. This part looks at a varlena pointer,
retrieves the on-disk ID. So this is left as-is on purpose, like the
direct TOAST decompress business based on varlena pointers.

Attached is a v2, to keep the CI happy for as long as we can use it.
--
Michael