September 06, 2004
Suppose you have a property defined like this:

----------
class Qwert {
    int yuiop() { ... }

    void yuiop(int i) { ... }
}
----------

We have a property getter and setter.  A statement like

    asdfg = qwert.yuiop = 42;

is not valid, since the yuiop setter returns void.  No problem:

----------
class Qwert {
    int yuiop() { ... }

    int yuiop(int i) {
        ...
        return i;
    }
}
----------

However, sometimes it isn't as simple as that.  The retrieval type of the property might be different from the type by which it is set.  The value might need some conversion or clipping that involves a time-consuming process.  Then you're stuck with something like:

----------
class Qwert {
    char[] yuiop() { ... }

    char[] yuiop(int i) {
        ...
        return yuiop;
    }
}
----------

which is wasteful when you're not using the value re-retrieved from the property.  If the getter calls an external API, the compiler cannot optimise it away, as it can't be sure that the API call has no essential side effects.  Moreover, the getter might allocate a new object that will be cached, which again would be wasteful if never used.

It would therefore be handy if, where a property setter's return type is void, the value of an AssignExpression is taken to be the result of getting the property value, which is evaluated only if it's actually used.  So,

    asdfg = qwert.yuiop = 42;

would be equivalent to

    qwert.yuiop = 42;
    asdfg = qwert.yuiop;

except that if qwert is an expression, it is evaluated only once.

And similarly,

    hjkl(qwert.yuiop = 42);

equivalent to

    qwert.yuiop = 42;
    hjkl(qwert.yuiop);

and the same with other expression forms.

Of course, property setters would be still allowed to return something, which would override the implicit chaining.  This would preserve the efficiency of existing code, allowing for the cases when returning the new value can be done in a way that bypasses the getter function (e.g. by passing the argument straight through, or if the setter API happens to return the converted/clipped value as an extra).

Stewart.

-- 
My e-mail is valid but not my primary mailbox.  Please keep replies on the 'group where everyone may benefit.
Top | Discussion index | About this forum | D home