let floating
A program transformation used in functional programming to implement full laziness. E.g. the function
f x = x + sqrt 4
f x = let t = sqrt 4 in x + t
but note that t does not depend on the argument x so we can automatically transform this to
t = sqrt 4 f x = x + t
Making t into a global constant which need only be evaluated at most once, rather than every time f is called. The general idea is to float each subexpression as far out (toward the top level) as possible to maximise sharing.