Add warning about GC to taggedPointer and taggedClassRef docs

[schuetzm]

http://forum.dlang.org/thread/ildhefvrvxxuphkfckcy@forum.dlang.org

[schuetzm]

Includes trivial typo fix in the same section.

[schveiguy]

I think this was a mistake. Pointer math is allowed, and pointer bitops with given preconditions are equivalent to pointer math.

I think we should revert this and fix the docs.

[dnadlinger]

Pointer bitops that do not respect the alignment of the pointed-to type are not equivalent to normal (+, -, etc.) pointer arithmetic. This change is correct.

[schveiguy]

It's not equivalent to pointer arithmetic, but we do allow pointer arithmetic to get the same result:

void *ptr = ptrToSomeInt;
ptr++;

This is allowed, and fundamentally must be. There is no difference here. In fact, the code could be reimplemented this way, and the result, as far as the GC cares, is the same. The GC docs are wrong IMO.

[schveiguy]

I'll ask this: if the pointer is stored as a void *, does this warning go away? I am not actually sure it isn't, I can't tell from the code, because it's so meta.

[deadalnix]

That doesn't make any sense. The code specifically checks so that this is safe to do (you can only tag pointer withing the alignment, meaning you can't get a pointer past the end of the object).

Can we fix the GC doc rather than making this needlessly scary to use ? The GC cannot work if it can't handle that to begin with.