I was chatting on IRC. One of the topic was about a "@nogc"
thread. I suggested something like "core.thread.NoGcThread" that
would take a "@nogc" function but then i realized that this
hypothetical class could still be allocated on the GC heap.
So "new" and "delete" would have to be disabled...Despite of
being deprecated it appears that class allocators and
deallocators can be used for that purpose, without compiler
changes:
---
class Foo
{
@disable new (size_t size){return null;}
@disable delete (void* p){}
}
void main()
{
static assert(!__traits(compiles, new Foo));
import std.experimental.allocator;
import std.experimental.allocator.mallocator;
auto f = make!Foo(Mallocator.instance);
dispose(Mallocator.instance, f);
// auto g = new Foo, // doesn't work
}
---
What's the point ?
- It's fun to discover that.
- Libraries that don't want their aggregates to reside on the GC
heap can use this, with a mixin. (note: although the GCAllocator
can still be used...)
