What are the exit conditions for graduating from std.experimental.* to std.*?

Has anything graduated yet?

-- 
Adam Wilson
IRC: LightBender
import quiet.dlang.dev;

On Wednesday, 4 January 2017 at 07:32:34 UTC, Adam Wilson wrote:

> Has anything graduated yet?

No

On 1/3/17 11:55 PM, Ilya Yaroshenko wrote:
> On Wednesday, 4 January 2017 at 07:32:34 UTC, Adam Wilson wrote:
>
>> Has anything graduated yet?
>
> No

So at what point well we? I mean that is the point after all...

-- 
Adam Wilson
IRC: LightBender
import quiet.dlang.dev;

On Wednesday, 4 January 2017 at 10:02:34 UTC, Adam Wilson wrote:
> On 1/3/17 11:55 PM, Ilya Yaroshenko wrote:
>> On Wednesday, 4 January 2017 at 07:32:34 UTC, Adam Wilson wrote:
>>
>>> Has anything graduated yet?
>>
>> No
>
> So at what point well we? I mean that is the point after all...

ndslice is deprecated https://github.com/dlang/phobos/pull/4946
It is hard to maintain both Phobos and Mir forks.

Allocators are useful but they are not betterC. I will fork them anyway if I will need to use them for production.

On Wednesday, 4 January 2017 at 07:32:34 UTC, Adam Wilson wrote:
> What are the exit conditions for graduating from std.experimental.* to std.*?

I think it should have been accepted by the community and there shouldn't be any design concern or usability issues.
However, this is a chicken-and-egg problem as most people don't want to use `experimental` thoroughly in production software.
Furthermore the exact criteria aren't defined which lead to the weird state that `std.experimental.allocator` is considered stable by most of the community.

The real issue is that moving from `experimental` to Phobos means that the API effectively can't be changed anymore due to aversion against deprecations and imho it's quite difficult to guarantee that an API is superb.

> Has anything graduated yet?

1) allocator
------------

You may want to see this discussion:

http://forum.dlang.org/post/hxwadyrnvzvutrdcgsdp@forum.dlang.org

2) logger
----------

I think sth. like the capability to do async logging is needed.

[1] https://github.com/dlang/phobos/pull/3194


3) typecons
------------

Both `Wrap` and `Final` were added as part of 2.072.

> ndslice is deprecated https://github.com/dlang/phobos/pull/4946
> It is hard to maintain both Phobos and Mir forks.

Yep, I fully agree with Ilya here. Whenever there's a change in Mir, it will be released almost instantly with a new version. This means that thanks to DUB a user can easily upgrade to check out the greatest & latest, but also stay on a stable version if he wishes so.
With Phobos one is usually tied to the bundled Phobos version of one's DMD compiler release.
You may also see Ilya's proposal for a more modular standard library:

http://forum.dlang.org/post/phexetutyelrssyrucvw@forum.dlang.org

On Wednesday, 4 January 2017 at 07:32:34 UTC, Adam Wilson wrote:
> What are the exit conditions for graduating from std.experimental.* to std.*?
>
> Has anything graduated yet?

Promotion of std.logger has been officially stalled until reference counted strings are part of D. I think this was done because the current code uses the GC a lot; not sure.

RC strings were being pushed heavily by Walter and Andrei until it was realized that in order for it to be @safe, there had to be a lot of holes in @safe plugged and new additions to the compiler.

Walter is still periodically working on @safe, but I'd say RC strings landing in master is two years away minimum.

On Wednesday, 4 January 2017 at 19:33:13 UTC, Jack Stouffer wrote:
> Promotion of std.logger has been officially stalled until reference counted strings are part of D.

When? That is the first I hear about that.

On Wednesday, 4 January 2017 at 13:47:53 UTC, Ilya Yaroshenko wrote:
> On Wednesday, 4 January 2017 at 10:02:34 UTC, Adam Wilson wrote:
>> On 1/3/17 11:55 PM, Ilya Yaroshenko wrote:
>>> On Wednesday, 4 January 2017 at 07:32:34 UTC, Adam Wilson wrote:
>>>
>>>> Has anything graduated yet?
>>>
>>> No
>>
>> So at what point well we? I mean that is the point after all...
>
> ndslice is deprecated https://github.com/dlang/phobos/pull/4946
> It is hard to maintain both Phobos and Mir forks.
>
> Allocators are useful but they are not betterC. I will fork them anyway if I will need to use them for production.

We should be able to take care of this with simple modularization techniques. And there's little worry about breakage because it's still experimental. Could you please submit fine-grained bug reports a la "this can't be done because this static constructor is in the same module as this independently useful artifact"? Or whatever the matter is. Then each issue can be looked at and decided upon properly. Thanks. -- Andrei

On Wednesday, 4 January 2017 at 21:59:29 UTC, Andrei Alexandrescu wrote:
> On Wednesday, 4 January 2017 at 13:47:53 UTC, Ilya Yaroshenko wrote:
>> [...]
>
> We should be able to take care of this with simple modularization techniques. And there's little worry about breakage because it's still experimental. Could you please submit fine-grained bug reports a la "this can't be done because this static constructor is in the same module as this independently useful artifact"? Or whatever the matter is. Then each issue can be looked at and decided upon properly. Thanks. -- Andrei

Have not reviewed std.allocator code for betterC mode. Basic idea is here https://issues.dlang.org/show_bug.cgi?id=17060

Thanks,
Ilya

On Wednesday, 4 January 2017 at 21:44:16 UTC, Robert burner Schadek wrote:
> On Wednesday, 4 January 2017 at 19:33:13 UTC, Jack Stouffer wrote:
>> Promotion of std.logger has been officially stalled until reference counted strings are part of D.
>
> When? That is the first I hear about that.

???

You were the one who told me about it: https://github.com/dlang/phobos/pull/1500#issuecomment-155457980