[bitc-dev] Problem: mutable literals

[David Hopwood]

Jonathan S. Shapiro wrote:
> HOWEVER: the partial evaluation of *immutable* content is actually
> important. When we see:
> 
>   (let ((v (vector (the int 1) 2 3))) ...)
> 
> is the compiler obligated to allocate a new vector on each entry?
> 
> It is tempting to imagine that no operator will reveal the partial
> evaluation, but consider:
> 
>   (define v-save:(mutable (vector int)) (vector 1))
> 
>   (define (f)
>     (let* ((v (vector 1 2 3))
>            (result (eq? v v-save)))
>       (set! v-save v)
>       result))
> 
> If partial evaluation is permitted, the first invocation of F will
> return #f, but the second will return #t.

One possible viewpoint is to consider this to be a problem with 'eq?'.
If a language only has Henry Baker's 'egal?' operator:

  Equal Rights for Functional Objects, or The More Things Change,
    The More They Are the Same
  <http://citeseer.ist.psu.edu/baker93equal.html>

then it cannot observe whether a reference to a deeply immutable
data structure is shared or not.

The difficulty then is that 'egal?' is not constant time, and sometimes
(not very often), you really want a constant-time reference equality
test.

-- 
David Hopwood <david.nospam.hopwood at blueyonder.co.uk>