AA with complex keytype?

[Stewart Gordon]

Manfred Nowak Wrote:

<snip>
> Why are classes required to implement opCmp( Object) for AA's to 
> function properly?

Apparently because Walter doesn't like the idea of providing an alternative implementation of AAs for unordered classes, but hasn't to my knowledge given a truly convincing reason.

My library implementation

http://pr.stewartsplace.org.uk/d/sutil/

uses only toHash and opEquals, thus works on unordered types.

> Currently AA's misbehave if opCmp(Object) is not overwritten for a 
> class, but no error is thrown.

Currently?  Not under DMD 1.005 as I try it.  Please supply a code example that demonstrates this behaviour.

> For the transitivity of an ordering it suffices for opCmp(Object) 
> to always give "==" if the objects are equal, and give "<" if they 
> are not.

So if x != y, then both x < y and y < x?  That wouldn't make sense at all.

> It should be easy to implement this as the default opCmp(Object).

OUAT, Object.opCmp was set up to compare the memory addresses, but this behaviour was removed to prepare for copying/compacting GC, and probably partly to eliminate the confusing behaviour it created.

Stewart.