On 9/11/2011 12:53 PM, dsimcha wrote:
On 9/11/2011 12:39 PM, Martin Nowak wrote:
RegionAllocator should have a dup function (better name required),
which will call newRegionAllocator on it's stack.
This is needed to create a nested scope without knowing the used stack.
martin
Good idea. This does suggest a little refactoring, though, because I
really don't want to add a **third** way to create a RegionAllocator.
Hmm, just off the top of my head, this might solve tons of problems:
1. Take RegionAllocatorStack out of the public API. It would still
exist conceptually, though.
2. newRegionAllocator() -> newRegionAllocator(newStackSize = 0); If
newStackSize == 0, return a RegionAllocator that uses the thread-local
stack. If newStackSize > 0, create a new stack (hidden from the user)
and return a new RegionAllocator on that stack.
3. Add dup(). Maybe call it newInstance() instead, or even just
newRegionAllocator(). Someone please suggest a good name. If using
explicit stacks the workflow would be:
void fun() {
// The whole stack is freed on exiting from fun, along with arr.
auto baseAlloc = newRegionAllocator(42 * 1024);
auto arr = baseAlloc.newArray!(int[])(666);
gun(baseAlloc);
}
void gun(RegionAllocator alloc) {
auto alloc2 = alloc.dup();
// arr2 is allocated on alloc2, freed on exit.
auto arr2 = alloc2.newArray!(int[])(42);
}
