Fix setPrimArray for Float & Double

[konsumlamm]

Fix the bug mentioned in #265 by generating different functions for Float and Double that check x == 0.0 && !signbit(x) to ensure that we only use memset for +0.0 (and not -0.0).

I also removed the special case for sizeof(Hs ## TYPE) == sizeof(int)*2 since it didn't yield better performance in my benchmark. I also don't see why one would expect performance improvements from that in the first place, since a single access should be faster than two.

[andrewthad]

This looks correct to me. And also, it looks like the tests suite would catch a mistake in the implementation of Prim for Float or Double. Are you alright with me merging this?

[konsumlamm]

And also, it looks like the tests suite would catch a mistake in the implementation of Prim for Float or Double.

Yes, I added primLaw tests for Float and Double since they were missing. I also added separate tests for the -0.0 bug, since as far as I can tell the Arbitrary instances don't generate -0.0.

Are you alright with me merging this?

Yes, go ahead!

[treeowl]

Yes, I added primLaw tests for Float and Double since they were missing. I also added separate tests for the -0.0 bug, since as far as I can tell the Arbitrary instances don't generate -0.0.

Should that be considered a bug in QuickCheck? Has it been reported?

[konsumlamm]

Yes, I added primLaw tests for Float and Double since they were missing. I also added separate tests for the -0.0 bug, since as far as I can tell the Arbitrary instances don't generate -0.0.

Should that be considered a bug in QuickCheck? Has it been reported?

0.0 and -0.0 behave the same in most cases and == can't distinguish them, but I think it would still make sense to generate it. It is mentioned in nick8325/quickcheck#98.

[treeowl]

That QuickCheck ticket is old enough to go to kindergarten.