Re: Strange failure in LWLock on skink in REL9_5_STABLE

On Sat, Sep 22, 2018 at 2:28 AM Andres Freund <andres(at)anarazel(dot)de> wrote:
>
> On 2018-09-22 08:54:57 +1200, Thomas Munro wrote:
> > On Fri, Sep 21, 2018 at 4:43 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> > > Thomas Munro <thomas(dot)munro(at)enterprisedb(dot)com> writes:
> > > > On Fri, Sep 21, 2018 at 4:06 PM Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> > > >> Why would we fix it rather than just removing it?
> > >
> > > > I assumed we wouldn't remove an extern C function extension code
> > > > somewhere might use. Though admittedly I'd be surprised if anyone
> > > > used this one.
> > >
> > > Unless it looks practical to support this behavior in the Windows
> > > and SysV cases, I think we should get rid of it rather than expend
> > > effort on supporting it for just some platforms.
> >
> > We can remove it in back-branches without breaking API compatibility:
> >
> > 1. Change dsm_impl_can_resize() to return false unconditionally (I
> > suppose client code is supposed to check this before using
> > dsm_resize(), though I'm not sure why it has an "impl" in its name if
> > it's part of the public interface of this module).
> > 2. Change dsm_resize() and dsm_remap() to raise an error conditionally.
> > 3. Rip out the DSM_OP_RESIZE cases from various places.
> >
> > Then in master, remove all of those functions completely. However,
> > I'd feel like a bit of a vandal. Robert and Amit probably had plans
> > for that code...?
>
> Robert, Amit: ^
>

I went through and check the original proposal [1] to see if any use
case is mentioned there, but nothing related has been discussed. I
couldn't think of much use of this facility except maybe for something
like parallelizing correalated sub-queries where the size of outer var
can change across executions and we might need to resize the initially
allocated memory. This is just a wild thought, I don't have any
concrete idea about this. Having said that, I don't object to
removing this especially because the implementation doesn't seem to be
complete. In future, if someone needs such a facility, they can first
develop a complete version of this API.

[1] - https://www.postgresql.org/message-id/CA%2BTgmoaDqDUgt%3D4Zs_QPOnBt%3DEstEaVNP%2B5t%2Bm%3DFPNWshiPR3A%40mail.gmail.com

--
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com