Heap allocation and internal pointers

[Dmitry Olshansky]

20-Jan-2014 23:41, Steven Schveighoffer пишет:
> On Mon, 20 Jan 2014 14:22:34 -0500, monarch_dodra
> <monarchdodra at gmail.com> wrote:
>
>> On Monday, 20 January 2014 at 12:27:41 UTC, Dicebot wrote:
>>> Quick dlang.org investigation has found this:
>>>
>>> http://dlang.org/phobos/std_exception.html#.pointsTo
>>>
>>> Note:
>>> Evaluating pointsTo(x, x) checks whether x has internal pointers.
>>> This should only be done as an assertive test, as the language is
>>> free to assume objects don't have internal pointers (TDPL 7.1.3.5).
>>>
>>> I guess you may want to check mentioned TDPL part ;)
>>
>> Yeah, I remember the pull that added that comment. As a matter of
>> fact, it was I that brought up said part in the original pull.
>>
>> It basically says "dmd assumes no internal pointers for its move
>> semantics" :/
>
> Um.. that should only apply to stack-allocated items. Heap allocated
> items are fine to point to themselves, even structs.
>
> Consider that cycles are allowed, and planned for, in the GC. An
> internal pointer is essentially a 1 element cycle!

Makes sense... Then allocating on GC heap is fine.

>
> A simple logical evaluation of how a moving GC would work reveals that
> it should update internal pointers as well as all other pointers.
>
> -Steve


-- 
Dmitry Olshansky