Discussion: Change the specification of rotateL for non-finite Bits, or move rotations to FiniteBits

[David Feuer]

On Sep 29, 2014 11:41 AM, "Edward Kmett" <ekmett at gmail.com> wrote:
>
> It does seem like it should rotate in 1s, but if it does so, what happens
when it rotates out something that isn't all 1s into a negative number or
all 0s into a positive number?
>
> We'd expect `rotateL n . rotateR n = id = rotateR n . rotateL n` for the
finite cases.
>
> but you can't rotate out the digits "up at infinity".
>
> So even this is messy. =/

You're absolutely right. Integer couldn't support that. You could, however,
have a Bits instance
newtype RotInteger = RotInteger {number::Integer, atInfinity::Integer}
That would be able to support rotations into and out of the bits "at
infinity".

> Ideally we would move rotate/rotateL/rotateR to FiniteBits, but there is
a huge upgrade/migration cost associated with that. All user specified
instances break in a way that requires CPP and it diverges from the
standard further.

Well, the approach taken for bitSize would work here too. Deprecate the
three names, leaving them in place. Add three new ones to FiniteBits, using
correct but somewhat slow defaults (implementing rotations using shifting
and masking). The FiniteBits dictionary is already growing from one member
to three, so adding another few should be okay, right?
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.haskell.org/pipermail/libraries/attachments/20140929/ead91be1/attachment.html>