support overloaded operators in lvalues

[nikomatsakis]

You should be able to write:

foo[x] = y

which should translate to foo.[]=(x, y) or something like that. This would be great as an alias for the insert() function on maps. I can't think of any other overloadable operators that are valid in lvalues... do we allow unary * to be overloaded?

[boggle]

I kind of like ruby's append operator << (which in rust is reserved for lshift), maybe there is a nice alternative?

[graydon]

Not unary star, I think. Likewise not unary ampersand. Too close to the memory model. Same thinking as keeping assign, move, copy as non overloadable.

[nikomatsakis]

An alternative we might consider is to make the call operator () overridable. There is a good argument for having maps (and possibly even vectors) be callable things. We would want to make them coercable to fn&() though. We could still allow foo(x) = y, although it looks odd.

[gwillen]

I urge you to consider allowing the overload of []= as a special case, even if nobody can think of any other operators that need lhs overloading. This may be one of those cases where there's just one specific crucial bit of syntax. (I have been arguing with sully about the need for [] on hashtables, and I'm about to file a bug asking him to implement fn[] on hash tables, readonly for now.)

[bstrie]

Closing in favor of #6515, since graydon has ruled against lvalue overloads other than indexing.