sandbox targets

[Tom Lane]

Dan Williams <dcbw at redhat.com> writes:
> We need to strike a balance here between people who might use the build
> system as a personal development machine (if they don't have PPC for
> example) and those people who do actually have a reasonable expectation
> that the package will build for them.

> A small little barrier like requiring a tag, while not much, is a
> psychological handbrake on dumping stuff into the buildsystem.

Certainly it's reasonable to expect people to have successfully built
a package locally before they dump it into the build system.  But you
can't expect them to have built it on all N architectures beforehand
--- so there is *always* some question in the mind when launching a
build.  (And that's supposing that the build system doesn't impose
any hazards of its own ... which isn't the case for Red Hat's internal
beehive system at least :-()

Not sure about the best tradeoffs here.  One thing I think ought to be
more easily possible is to launch a build on a "scratch" basis, and then
transfer the result into the "real results" arena only after you've
checked it out.

			regards, tom lane