Re: Strange failure in LWLock on skink in REL9_5_STABLE

On Mon, Nov 5, 2018 at 2:28 AM Michael Paquier <michael(at)paquier(dot)xyz> wrote:
> Hm. Don't we need to worry about anybody potentially using these APIs
> in a custom module on platforms where it was actually working? I
> imagine that their reaction is not going be nice if any code breaks
> suddenly after a minor release. No issues with removing the interface
> on HEAD of course.

+1.

The fact that the code exists there at all is my fault. I thought it
might be useful someday, but now I don't think so any more. Thomas's
solution -- in the DSA machinery -- of allocating entirely new
segments seems like a better approach for now, and in the long run, I
think we should convert the whole backend to use threads,
nonwithstanding the TODO entry that says otherwise. Even if we never
do that, extending a segment in place is pretty difficult to make
practical, since it may involve remapping the segment, which
invalidates cached pointers to anything in the segment. And then
there are the portability problems on top of that. So I'm not very
optimistic about this any more.

But ripping it out in the back branches seems unnecessary. I'd just
leave the bugs unfixed there. Most likely nobody is using that stuff,
but if they are, let's not break it.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company