On 25.01.23 22:16, Jeff Davis wrote:
> I am highlighting this case because the existence of a single non-
> contrived case or regression suggests that we may want to explore
> further and tweak heuristics. That's quite natural when the heuristics
> are based on a complex dependency like a collation provider. The
> sort_abbreviated_keys GUC makes that kind of exploration and tweaking a
> lot easier.
Maybe an easier way to enable or disable it in the source code with a
#define would serve this. Making it a GUC right away seems a bit
heavy-handed. Further exploration and tweaking might well require
further source code changes, so relying on a source code level toggle
would seem appropriate.