clay RFC: rename bitcast to refcast, reimplement bitcast as a bitwise copy

What do you guys think of renaming the current bitcast primitive, which returns a reference of a new type to the same memory, to refcast, and having bitcast actually bitwise-copy a new value? bitcast could then be less restrictive about the types it casts, since the destination type could have less strict alignment than the source.

Oct 15 '12 02:10 jckarter

Intuitively I would take any 'cast' to be in place. How about a bitcopy primitive for the latter case?

Oct 15 '12 11:10 ghost

My worry is that the alignment requirement for bitcast is too restrictive, and might hinder casts you'd expect to work on some platforms, for instance if an Int64 is 4-byte-aligned but a Double is 8-byte-aligned as in some x86-32 ABIs.

Oct 15 '12 16:10 jckarter

I guess my concern is not so much with the change to refcast which I think works just fine, but using a name that more accurately represents whats happening under the bonnet with the 'cast with copy to ensure correct alignment' case. copycast, alignedcast, or whatever. Maybe that's too much syntactic bike-shedding . . .

Oct 15 '12 16:10 ghost

Change of heart on this one, alternatives aren't really preferable to keeping bitcast. +1 for proposed changes.

Nov 10 '12 12:11 ghost