I agree with Peter that "We don't tab-complete everything we possibly could", but using tabs after "SET ROLE TO " provides "DEFAULT" as an option which seems wrong. This patch adds list of roles over there, which I guess good to have than giving something unusual.
...
But back to this topic. I am thinking so it is little bit different due fact so we support two very syntax for one feature. And looks little bit strange, so one way is supported by autocomplete and second not.
Yeah, it's a bit strange. We have a specific autocomplete rule for "SET ROLE", but "SET ROLE TO" is treated as a generic GUC. With your patch, we'd also lose the auto-completion to "SET ROLE TO DEFAULT".
I think we want to encourage people to use the SQL-standard syntax "SET ROLE ..." rather than the PostgreSQL-specific "SET ROLE TO ...". On the whole, this just doesn't seem like much of an improvement. I'll mark this as 'rejected' in the commitfest app.
PS. I note that the auto-completion for "SET XXX TO ... is pretty lousy in general. We have rules for DateStyle, IntervalStyle, GEQO and search_path, but that's it. That could be expanded a lot. All enum-type GUCs could be handled with a single rule that queries pg_settings.enumvals, for example, and booleans would be easy too. But that's a different story.
I wrote a patch for fallback tabcomplete for bool and enum GUC variables
