Re: Should contrib modules install .h files?

>>>>> "Andres" == Andres Freund <andres(at)anarazel(dot)de> writes:

Andres> I'm a bit surprised that you decided to push to the 11 branch -
Andres> the consensus in this thread seem to have gone the other way by
Andres> my read?

There were also quite a few non-objections or "don't care"s to the idea
of pushing it to 11beta, and no real justification for not doing so was
ever given other than "it's a new feature", which I don't really agree
with (it's certainly fixing a defect, even if you don't want to call it
a bug). I actually think an argument could be made for backpatching it
even further (though I don't intend to pursue that).

The reason for making the change in the first place is to support
out-of-tree development (particularly of PLs and their transforms) which
isn't strongly tied to pg's development and release cycle; not pushing
to 11-beta would mean delaying the issue for another year and forcing
yet another cycle of workarounds. (Right now for example I can take
advantage of the fact that hstore.h didn't change between 9.5, 9.6 and
10 except in non-significant whitespace; if I can rely on hstore.h being
actually installed in pg11 onwards, then I can draw a line under the
issue and move on. Otherwise, I have to deal with the fact that hstore.h
_does_ change between 10 and 11 and then wait for 12 to see what that
ends up doing.)

(Obviously I'm biased by the fact that this is an issue that impacts me
personally.)

--
Andrew (irc:RhodiumToad)