On Tue, Mar 07, 2023 at 02:13:07PM -0500, Tom Lane wrote: > What I am maintaining is that no extension author is actually going > to write such a script, indeed they probably won't trouble to write > any downgrade-like actions at all. Which makes the proposed design > mostly a foot-gun.
What I'm maintaining is that such authors should be warned about the risk, and discouraged from installing any wildcard-containing script UNLESS they deal with downgrade protection. PostGIS does deal with that kind of protection (yes, could be helped somehow in doing that by PostgreSQL). > I'm not unsympathetic to the idea of trying to support multiple upgrade > paths in one script. I just don't like this particular design for that, > because it requires the extension author to make promises that nobody > is actually going to deliver on. Would you be ok with a stricter pattern matching ? Something like: postgis--3.3.%--3.3.ANY.sql postgis--3.3.ANY--3.4.0.sql Would that be easier to promise something about ? --strk;