On Tue, Nov 5, 2019 at 4:18 AM Julien Rouhaud <rjuju...@gmail.com> wrote: > On Mon, Nov 4, 2019 at 11:13 AM Julien Rouhaud <rjuju...@gmail.com> wrote: > > On Mon, Nov 4, 2019 at 4:58 AM Thomas Munro <thomas.mu...@gmail.com> wrote: > > > Here are some problems to think about: > > > > > > * We'd need to track dependencies on the default collation once we > > > have versioning for that [...] > > Another problem I just thought about is how to avoid discrepancy of > collation version for indexes on shared objects, such as > pg_database_datname_index.
I didn't look closely at the code, but I think when "name" recently became collation-aware (commit 586b98fd), it switched to using C_COLLATION_OID as its typcollation, and "C" doesn't need versioning, so I think it would only be a problem if there are shared catalogs that have "name" columns that have a non-type-default collation. There are none of those, and you can't create them, right? If there were, if we take this patch set to its logical conclusion, we'd also need pg_shdepend.refobjversion, but we don't need it AFAICS.
