only provide traits like Eq/Ord on @T and @mut T when they're known to be non-cyclic

[thestinger]

Based on #5767.

We should really only be providing these default implementations when we know they're valid.

An easy way of doing it is only implementing them by default on T: Const. It's still possible to make user-defined implementations for types containing non-Const @T, you just can't use deriving and implicitly end up with a faulty implementation.

[thestinger]

Nominating for the backwards compatible milestone.

[huonw]

Triage visit: it would be neat if we could somehow implement it so that == worked perfectly on possibly cyclic structures, e.g. Python handles it fine:

>>> L = []
>>> L.append(L)
>>> L == L
True
>>> x = [L, 1]
>>> y = [L, 2]
>>> x == y
False

At the very least, for @ and @mut, using TLS and keeping track of the nodes(memory locations?) already examined, and substituting in the appropriate value for the trait being implemented rather than recursing again.

(This would presumably have to be careful in the presence of a possibly-moving GC, but I'd guess that such a function call would freeze the pointers in memory?)

[catamorphism]

Just a bug

[thestinger]

This will be dealt with by the move to Gc<T>.