[MLton] starting builds

[Stephen Weeks]

Thanks for the logs.

> There seems to be some hppa funkiness with reals.
> However, I think it was the same before.

Normally we don't worry too much about the low bits in the real
regression.  But in this case, the failure is the section

  Testing copySign, sameSign, sign, signBit

Once Fail "bug" is raised there, the test terminates and we don't even
get to compare the rest.

Could you look into that test in particular and see why Fail "bug" is
being raised?  It could be something more serious than precision.

Also, the testdyn1 problem is troubling, since that is simply testing
floor -23.23= -24.  Could you look at that as well?  Thanks.