On Sunday, 6 November 2016 at 02:12:12 UTC, Alexandru Caciulescu
wrote:
On Friday, 4 November 2016 at 09:37:39 UTC, Shachar Shemesh
wrote:
On 03/11/16 23:00, Yuxuan Shui wrote:
On Wednesday, 2 November 2016 at 03:36:42 UTC, Andrei
Alexandrescu wrote:
[snip]
* The implementation cannot use Phobos' allocator because
it's in
druntime.
How about let the user provide the memory block when they
calls reserve()?
I think that's just a halfhearted attempt at introducing STL's
allocators.
I see this topic started a clash of opinions regarding the
future of AAs. After Andrei suggested a free-list
implementation I had a good idea of how to proceed but now I am
not so sure since this discussion isn't converging to a single
idea/implementation.
To make a concrete proposal - For near-term, against current AAs,
add a 'reserve()' function that pre-allocates the bucket array,
but not the memory for elements. This would be consistent with
the current implementation. It looks like it could be done via a
call to the 'resize()' function. It would be similar to the
'rehash()' function (see _aaRehash() in aaA.d). Memory allocation
and management for inserted elements is unchanged, it continues
to operate as current. This approach would be similar to how the
C++ std::unordered_map::reserve() function works.
This still permits a longer term approach adding template-based
hash tables in the future, with various options for memory
management, etc.
I think this suggestion is consistent with Steve Schveighoffer's
suggestion earlier in the thread.
--Jon
