[Nix-dev] Re: Stdenv compilation times

[Peter Simons]

Hi Michael,

 >> Because I want my OS to work as reliably as possible, by default.
 >
 > I want updates to take finite amount of time, so problems get fixed in
 > reasonable time by default.

these two interests don't necessarily contradict each other.

 > I described approach that will let you know of any test failures
 > anyway.

Your approach has its merits, but it also has a downside. When tests are
run in the same expression as the actual build, then a test suite
failure is essentially a build failure. Consequently, expressions that
fail the test suite cannot be used in production by accident. The scheme
you described, however, does not accomplish that, because a failure in
"foobar_checked" doesn't prevent "foobar" from being installed.


 > And look at bison - it is important for GCC, yet tests are getting
 > turned off because the test suite is incorrect in some places. You
 > really hope for a better treatment of PPL?

Do you have concrete information that suggests that PPL's test suite is
unreliable or is going to be unreliable in the foreseeable future? It
sounds to me like you're speculating, or rather, you're making an
invalid generalization. Just because bison's test suite is (apparently)
unreliable, it doesn't mean that PPL's will be.

Take care,
Peter