bearophile wrote:
Walter Bright:
Java does do some escape analysis to try and allocate heap objects on
the stack instead, but I don't know how effective this is, and even that
won't help if you try to embed a value aggregate into a class:
struct S { int x, y, z; }
class C
{
S v; // in Java this would require a second heap allocation
}
I have discussed with LDC devs an improvement of that idea (it was discussed in
this newsgroup too):
class C1 { int x, y, z; }
class C2 {
scope C1 c;
}
This idea raises few problems, but I think they can be solved.
You can special-case the management of such objects scope-allocated inside
other objects. Or you can just add an invisible field to that C2 class, the
reference to c.
Bye,
bearophile
I would love to see such a feature added in D, would make object
composition much more convenient and faster.
You'd just need the compiler to force a call into C1's constructor
within C2's and generate a call to C1's destructor in the epilog of
C2's. I can't see how implementing idea that could be any harder, other
than the parser related support.
No need to store a reference within the object, that would still require
another heap allocation killing the need for scope in the first place.
