[racket] shared and immutable structures

At Tue, 1 Feb 2011 12:38:44 -0800, David Herman wrote:
> Just thinking out loud, but is there any way to use the unsafe operations to 
> make the shared library work for immutable data? If you can prove it's 
> actually safe, then there'd be no harm to users. (After all, as you say, 
> that's what it's doing with pairs.)

The difference is that pairs are defined a priori as allowing the
possibility of cycles, so you know that possibility exists when working
with pairs.

If my module gives yours a structure type, however, I may not want to
automatically give yours the capability to create cyclic instances of
the structure type. In that sense, creating a cycle is not always safe.

Furthermore, implementing the cycle by creation plus mutation would
mean introducing mutation for a supposedly immutable datatype, which
isn't necessarily safe if the intermediate state is visible. Attempting
to hide the intermediate state creates all sorts of new problems for
features like properties and guards.

Along those lines, granting the capability to create cyclic instances
seems to be just about tantamount to granting mutability, so `shared'
might as well require mutability.