On 15/01/16 5:11 AM, Andrei Alexandrescu wrote:
> On 01/14/2016 10:46 AM, Jacob Carlborg wrote:
>> On 2016-01-14 15:52, Andrei Alexandrescu wrote:
>>
>>> I don't see how this categorization works. E.g. there's no need for
>>> special tooling related to the GC. -- Andrei
>>
>> It was just a couple of suggestions. I don't know exactly what you're
>> going to include in this module.
>
> Me neither, which adds to the problem :o). -- Andrei

I can already suggest one thing.

I'm currently working on a library solution to my managed memory concept. I should have it completed within a day or so.

I will be replacing the dummy ref counted solution I built for alphaPhobos with this. As it allows me a LOT more control over the memory.

And it will be wonderful for containers with secondary ownership and all that :)

http://wiki.dlang.org/User:Alphaglosined/ManagedMemory

On Thursday, 14 January 2016 at 00:35:53 UTC, Andrei Alexandrescu wrote:
> Hey folks, I want to push things forward with artifacts dedicated to avoiding the GC, and of course my main worry is finding the right name.
>
> An obvious choice is std.experimental.nogc but we know from Marketing 101 that expressing something as a positive is better than a negative. Another possibility is std.experimental.rc, but that's imprecise because the artifacts in there will contain a variety of things in addition to reference counting-related artifacts.
>
> Ideas?
>
>
> Andrei

Will there also be an article about how to integrate the new memory model(s) into existing code? Something along the lines of "How to make existing code GC free". It'd be great to have a guide that shows how to do it properly.

On Thursday, 14 January 2016 at 07:44:16 UTC, Jacob Carlborg wrote:
> On 2016-01-14 01:35, Andrei Alexandrescu wrote:
>> Hey folks, I want to push things forward with artifacts dedicated to
>> avoiding the GC, and of course my main worry is finding the right name.
>>
>> An obvious choice is std.experimental.nogc but we know from Marketing
>> 101 that expressing something as a positive is better than a negative.
>> Another possibility is std.experimental.rc, but that's imprecise because
>> the artifacts in there will contain a variety of things in addition to
>> reference counting-related artifacts.
>
> std.experimental.memory with submodules for the different use cases:
>
> std.experimental.memory.rc
> std.experimental.memory.gc
> std.experimental.memory.manual // or something

+1

On Saturday, 16 January 2016 at 15:56:39 UTC, karabuta wrote:
> On Thursday, 14 January 2016 at 07:44:16 UTC, Jacob Carlborg wrote:

>> std.experimental.memory.rc
>> std.experimental.memory.gc
>> std.experimental.memory.manual // or something
>
> +1


Simplifies things IMO

On Thursday, 14 January 2016 at 14:52:19 UTC, Andrei Alexandrescu wrote:
> On 01/14/2016 09:50 AM, bachmeier wrote:
>> On Thursday, 14 January 2016 at 07:44:16 UTC, Jacob Carlborg wrote:
>>
>>> std.experimental.memory.rc
>>> std.experimental.memory.gc
>>> std.experimental.memory.manual // or something
>>
>> +1
>
> I don't see how this categorization works. E.g. there's no need for special tooling related to the GC. -- Andrei

That's not true. All of http://dlang.org/phobos/core_memory is about the GC. If some of the low-level stuff are put in druntime, then this categorization would look nice:

core.memory.gc (previously just core.memory)
core.memory.manual (something along the lines of https://github.com/D-Programming-Language/phobos/pull/3031)
core.memory.rc (Unique/Isolated and Shared smart pointers)

On Thursday, 14 January 2016 at 07:44:16 UTC, Jacob Carlborg wrote:
> std.experimental.memory.rc
> std.experimental.memory.gc
> std.experimental.memory.manual // or something

This "rc" is "reference counting"? Reference counting is also a form of garbage collection ("gc"). So I don't like this classification. If we really want "rc", then something like "trace" instead of "gc" should be the alternative.

On 2016-01-18 09:40, qznc wrote:

> This "rc" is "reference counting"? Reference counting is also a form of
> garbage collection ("gc"). So I don't like this classification. If we
> really want "rc", then something like "trace" instead of "gc" should be
> the alternative.

They are suggestions. More an indication that he doesn't need to come up with a weird name that can fit anything.

-- 
/Jacob Carlborg