Better signature for copy constructors.

[Thierry Delisle]

Copy constructors now have a weird C copy as a parameter. This is due to historical reasons and should use const reference instead.

I.e. this:

void ?{}(char &, char);
char ?=?(char &, char);

Instead of

void ?{}(char &, char const &);
char const & ?=?(char &, char const &); 

[ajbeach]

There was a typo in the original copy assignment signature.

After a discussion about wanting the return value of an assignment to be value like, it was agreed that we should change it to be a constant reference. You could still take the address (unless a check is added to prevent it) but it is logically closer to a value.

There was some talk of allowing different signatures be used for different types (for instance, C primitive types could accept and return values) but this interferes with polymorphism and properly lowering from Cforall to C in the last stages of the compiler should take care of performance issues in the remaining cases.

If we allowed operators to not be 1-to-1 with the underlying functions (which is a big change), then we could translate "a = b" into "?=?(a, b), a", that is to say, have the operator return void and provide the return value by comma expression. This forces the behaviour to be consistent, means the user does not have to worry about it and may give more control to the optimizer.

[ajbeach]

Diff from attempt to make this change.

[ajbeach]

I am attaching my attempt to make this change. It does not work, it never worked and now there is also an infinite loop that I do not recall being there when I was working on it actively. It should be enough to help whoever picks it up later.