Re: Crash report for some ICU-52 (debian8) COLLATE and work_mem values

I wrote:
> I'm on board with Peter G's proposal: let's distrust ucol_strcollUTF8
> in ICU < 53.  If those folk felt a need to rewrite it completely after
> only three release cycles, that should tell us something.

A bit of googling turned up this document:
http://site.icu-project.org/design/collation/v2
which indicates that the ucol_strcollUTF8 rewrite was really just one
facet of a rather major reimplementation.  But still, that document says
specifically that they knew of multiple bugs and suspected many more.

At this point I'm thinking that really what we ought to do is deprecate
using any pre-53 ICU release for Postgres.  It is very clear that those
versions are an entirely different beast from 53-and-up, and that they
are now abandonware so far as ICU upstream is concerned.

I have not checked, but I wonder whether 53 is also when the large
change in the set of available collations happened.  Maybe rejecting
pre-53 would also be enough to assuage my concerns about disappearing
collations.

			regards, tom lane