On 5/30/13 2:25 PM, Maxim Fomin wrote:
In neighbor thread (especially from 5 page
http://forum.dlang.org/thread/[email protected]?page=5)
there is discussion about current state of definitions in D related to
slices, slice expressions and arrays. There is significant contradiction
between documentation in different parts of the D site and people who
interpret it.
Problem boils down to following:
- in array and type official spec page, dynamic array is defined as T[]
type as "Dynamic arrays consist of a length and a pointer to the array
data.". The page also describes what slicing is. Also expression page
defines what SliceExpression is. Internally dmd follows these conventions.
- in articles part of the site there is article "D Slices" written by
Steven Schveighoffer, which abolishes current relevant parts of current
spec. According to the article, dynamic array is runtime managed memory
which in implementation specific manner provides some set of operation
related to arrays. According to the article T[] is by no means a dynamic
array, but a slice. The article explicitly claims that spec is wrong.
So, there is contradiction between what T[] is. Either it is a slice
(and what is more important, not a dynamic array type) which point by
druntime managed dynamic array, or is object of type dynamic array,
which may point to heap or stack memory.
Discussion shows that there is no clear consensus on this, so there is
need for Walter and Andrei to comment on this.
T[] is both. You can use it to take a slice of any piece of memory,
including freshly GC-allocated arrays. But when you try e.g. to append
to it with ~=, that only works if the slice originated as a GC-allocated
array.
Andrei
