Changes to Trunk (
http://source.squeak.org/trunk.html) in the last 24 hours:
http://lists.squeakfoundation.org/pipermail/packages/2014-March/006981.htmlName: Kernel-nice.843
Ancestors: Kernel-nice.842
Correct the bug I introduced for large integer sqrtFloor.
If receiver was of the form 2^2n*u
we did answer 2^n *E(sqrt(u))
But sqrt(u) = E(sqrt(u)) + residue
where 0<=residue<1
And result is rather
E(2^n*sqrt(u))
= E( 2^n*(E(sqrt(u)) + residue))
= 2^n*E(sqrt(u)) + E(2^n*residue)
As 2^n * residue can be big (superior to 1), we miss the correct value by default.
If we want to correct this, an approximation of residue is (u-E(sqrt(u))^2) / 2 E(sqrt(u))
This is more or less like super Newton-Raphson inner loop...
It would be possible to duplicate super work, but I feel like it's adding too much complexity for small reward, so I prefer to remove offending code for now
=============================================
http://lists.squeakfoundation.org/pipermail/packages/2014-March/006982.htmlName: KernelTests-nice.264
Ancestors: KernelTests-nice.263
A non regression test for just found sqrtFloor bug.
=============================================