Re: pg_xlogdump

On Tue, Feb 26, 2013 at 12:02 PM, Peter Eisentraut <peter_e(at)gmx(dot)net> wrote:
> On 2/26/13 11:45 AM, Tom Lane wrote:
>> But let's not break the cases that do work. One
>> of the functions of contrib/ is to serve as models/skeletons for
>> external modules. If we pull out the "useless" PGXS support then we'll
>> just be making it that much harder to copy a contrib module and start
>> doing something useful.
>
> Well, this is exactly the problem. Because of this skeleton idea, most
> external extension modules do not build unless you set USE_PGXS=1 before
> building, because they think that they live in contrib by default, which
> is completely bizarre and user-unfriendly.
>
> We could have an actual example or skeleton whose purpose is to teach
> extension authors. The actual contrib module makefiles should just do
> their job and don't pretend to teach things that are misguided and/or
> don't work.

I couldn't agree more. I can't think how much time I've wasted by
forgetting (or remembering) to type USE_PGXS=1 over the years, or not
realizing that I needed to. I'm sure there are many other people in
the same boat.

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