By the way, the test you have proposed in [1] to copy the expression stats from an origin to a target and check the differences was super nice. Could you add that back, with 'exprs' set as a jsonb blob? This was posted in 0003, but got cut due to the initial integration of the feature without expressions.
+1
I have put my hands on the patch, and fixed all these issues, adding tests to cover the holes I have spotted, and addressed a lot of stylistic things. I did not take the time to add the diff test, though when the expressions are cloned. Could you?
+1
The cases of STATISTIC_KIND_RANGE_LENGTH_HISTOGRAM and STATISTIC_KIND_BOUNDS_HISTOGRAM are a bit sad. How about exposing them into pg_stats_ext_exprs as a first patch, then extend the expression restore function to be able to work on them? Modifying the view would be an entirely independent thing. Please note that I do not mind skipping these two cases for now, the proposed patch is already doing a lot. Still if we are able to close entirely the loop for this release that would be nice. Anyway, let's keep this stuff as separate patches on top of this v8 and future reviewed versions.
I was thinking of doing exactly that, with a lot of language around how these extra patches don't need to make it into v19. But having said that, they'll probably be pretty small.
