Re: costly foreign key ri checks (4)

Fabien COELHO <coelho(at)cri(dot)ensmp(dot)fr> writes:
> I'll keep that in mind. However, from other projects I've worked on, I
> found that large regression tests are not wasted.

> Maybe two level of tests wouldn't be bad, as when you're about to release
> a new version, it's better to pass large tests, but when installing some
> light checks are enough just to check that all functionnalities are there.
> I think I already saw something like that somewhere in the sources?

There is a provision for "big tests" in the regression test
infrastructure. The only additional test at present is a rerun of the
numeric test with greater precision choices; which personally I think is
quite useless. But we could imagine using the mechanism for other stuff.

regards, tom lane