On Monday, 22 December 2014 at 21:09:21 UTC, Meta wrote:
There are only a couple of constructs in D that allocate, so it
may be worthwhile to let the allocator control the primitives,
i.e.:
auto gc = new GCAllocator(...);
gc.Array!char a = ['x', 'y'];
a ~= ['a', 'b']; //use allocatorStack.top.realloc(...);
//I don't remember the proposed API for
//allocators, but you get the idea
gcAlloc.DestroyAll();
This looks a bit uglier, but it also doesn't require any new
language constructs. The downside is, how do you manage things
like closures doing it this way?
In this case you have to make all things template on allocator or
even more on Allocator!T making Cartesian product of allocators
and types. The point is to provide method to handle different
memory management strategies by user in run time.
It is ease to give up GC using annotation and run time stack. Set
an allocator for a class by annotation(apply for all methods) and
it force std library to use the allocator whenever you use it in
the class.
