On 29.05.2012 10:52, Era Scarecrow wrote:
On Tuesday, 29 May 2012 at 06:30:37 UTC, Dmitry Olshansky wrote:
You surely haven't looked at the source code did you? :)
It's conceptualy non per bit '-', it's a set difference...
I recall looking at it, but to me that just didn't make sense. I could
add subtract back and update it (Not many changes needed to keep it).
Not at all. Once you established that it's not a pointer namely since
every pointer to size_t is word aligned (unless constructed by hand).
You could use it's lowest bit as marker then. It's 0 state won't
disturb pointer usual semantics, when it's set to 1 it's obviously.
I considered that, but then you actually limit your address space to
2^63,
No you don't. Since pointer is already a pointer to word-sized object.
It has 2 last bits == 0. Always. There is no escaping of this fact. And
no your address space is intact. All it has to do is assuming proper
alignment, and you sure have it since you _allocate_ it.
To be more specific most allocator go even farther and provide 8bytes
aligned pointer.
true that seems silly up until in the future when we use all
64bits for memory referencing and suddenly it's seg faulting for no
understandable reason (Yes it's a long ways off, but this will long be
forgotten about by that time). However referring to the internal offsets
it only effects slices; and those are easy to fix.
See the above.
This will likely take a little time to think over and get working, I'd
hate to have to make alternate versions of everything.
Cool. Eager to see that on the way to Phobos too.
We'll see, once I sorta figure this all out.
--
Dmitry Olshansky
