Hi,

When combining ref counting and GC allocated classes, weird bugs might happen. Is it possible to add a new member in `core.memory.GC`, like `isCollecting` or `isRunning`, to avoid testing if the GC is running, without allocating like this?

https://p0nce.github.io/d-idioms/#GC-proof-resource-class


Thanks,
Bogdan

On Thursday, 12 December 2019 at 08:56:37 UTC, bogdan wrote:
> Hi,
>
> When combining ref counting and GC allocated classes, weird bugs might happen. Is it possible to add a new member in `core.memory.GC`, like `isCollecting` or `isRunning`, to avoid testing if the GC is running, without allocating like this?
>
> https://p0nce.github.io/d-idioms/#GC-proof-resource-class
>
>
> Thanks,
> Bogdan

I agree that this is a good idea. It shouldn't be hard to implement as the GC implementation already has this information available, so it's just a matter of exposing it through the layers.

On 2019-12-12 10:00, Petar Kirov [ZombineDev] wrote:

> I agree that this is a good idea. It shouldn't be hard to implement as the GC implementation already has this information available, so it's just a matter of exposing it through the layers.

In Tango for D1, Object has an additional method, `dispose`,  which is called by explicit `delete` or when the scope exists and the class variable is declared `scope`.

-- 
/Jacob Carlborg

On Thursday, 12 December 2019 at 09:00:38 UTC, Petar Kirov [ZombineDev] wrote:
> On Thursday, 12 December 2019 at 08:56:37 UTC, bogdan wrote:
>> Hi,
>>
>> When combining ref counting and GC allocated classes, weird bugs might happen. Is it possible to add a new member in `core.memory.GC`, like `isCollecting` or `isRunning`, to avoid testing if the GC is running, without allocating like this?
>>
>> https://p0nce.github.io/d-idioms/#GC-proof-resource-class
>>
>>
>
> I agree that this is a good idea. It shouldn't be hard to implement as the GC implementation already has this information available, so it's just a matter of exposing it through the layers.

https://issues.dlang.org/show_bug.cgi?id=17563

-Steve

On Sunday, 15 December 2019 at 22:33:58 UTC, Steven Schveighoffer wrote:
> On Thursday, 12 December 2019 at 09:00:38 UTC, Petar Kirov [ZombineDev] wrote:
>> On Thursday, 12 December 2019 at 08:56:37 UTC, bogdan wrote:
>>> Hi,
>>>
>>> When combining ref counting and GC allocated classes, weird bugs might happen. Is it possible to add a new member in `core.memory.GC`, like `isCollecting` or `isRunning`, to avoid testing if the GC is running, without allocating like this?
>>>
>>> https://p0nce.github.io/d-idioms/#GC-proof-resource-class
>>>
>>>
>>
>> I agree that this is a good idea. It shouldn't be hard to implement as the GC implementation already has this information available, so it's just a matter of exposing it through the layers.
>
> https://issues.dlang.org/show_bug.cgi?id=17563
>
> -Steve

This looks like an old issue, maybe we won't have this very soon... Is a DIP necessary for such changes?

On 17/12/2019 2:23 AM, Bogdan Szabo wrote:
> On Sunday, 15 December 2019 at 22:33:58 UTC, Steven Schveighoffer wrote:
>> On Thursday, 12 December 2019 at 09:00:38 UTC, Petar Kirov [ZombineDev] wrote:
>>> On Thursday, 12 December 2019 at 08:56:37 UTC, bogdan wrote:
>>>> Hi,
>>>>
>>>> When combining ref counting and GC allocated classes, weird bugs might happen. Is it possible to add a new member in `core.memory.GC`, like `isCollecting` or `isRunning`, to avoid testing if the GC is running, without allocating like this?
>>>>
>>>> https://p0nce.github.io/d-idioms/#GC-proof-resource-class
>>>>
>>>>
>>>
>>> I agree that this is a good idea. It shouldn't be hard to implement as the GC implementation already has this information available, so it's just a matter of exposing it through the layers.
>>
>> https://issues.dlang.org/show_bug.cgi?id=17563
>>
>> -Steve
> 
> This looks like an old issue, maybe we won't have this very soon... Is a DIP necessary for such changes?

Only if it requires language changes (like adding a parameter to the destructor).

On 12/16/19 8:23 AM, Bogdan Szabo wrote:
> On Sunday, 15 December 2019 at 22:33:58 UTC, Steven Schveighoffer wrote:
>> On Thursday, 12 December 2019 at 09:00:38 UTC, Petar Kirov [ZombineDev] wrote:
>>> On Thursday, 12 December 2019 at 08:56:37 UTC, bogdan wrote:
>>>> Hi,
>>>>
>>>> When combining ref counting and GC allocated classes, weird bugs might happen. Is it possible to add a new member in `core.memory.GC`, like `isCollecting` or `isRunning`, to avoid testing if the GC is running, without allocating like this?
>>>>
>>>> https://p0nce.github.io/d-idioms/#GC-proof-resource-class
>>>>
>>>>
>>>
>>> I agree that this is a good idea. It shouldn't be hard to implement as the GC implementation already has this information available, so it's just a matter of exposing it through the layers.
>>
>> https://issues.dlang.org/show_bug.cgi?id=17563
>>
> 
> This looks like an old issue, maybe we won't have this very soon... Is a DIP necessary for such changes?

No. It's completely a library change, and backwards compatible.

-Steve

On Monday, 16 December 2019 at 13:23:56 UTC, Bogdan Szabo wrote:

> [..]
>
> This looks like an old issue, maybe we won't have this very soon... Is a DIP necessary for such changes?

Elapsed time is not always a good indication of the complexity or feasibility of a task, especially in open-source ;)

https://github.com/dlang/druntime/pull/2873

On Monday, 16 December 2019 at 23:36:21 UTC, Petar Kirov [ZombineDev] wrote:
> On Monday, 16 December 2019 at 13:23:56 UTC, Bogdan Szabo wrote:
>
>> [..]
>>
>> This looks like an old issue, maybe we won't have this very soon... Is a DIP necessary for such changes?
>
> Elapsed time is not always a good indication of the complexity or feasibility of a task, especially in open-source ;)
>
> https://github.com/dlang/druntime/pull/2873

P.S. Thanks for reminding me of this issue Bogdan, sometimes communication is all that's necessary :)

On Monday, 16 December 2019 at 23:38:15 UTC, Petar Kirov [ZombineDev] wrote:
> On Monday, 16 December 2019 at 23:36:21 UTC, Petar Kirov [ZombineDev] wrote:
>> On Monday, 16 December 2019 at 13:23:56 UTC, Bogdan Szabo wrote:
>>
>>> [..]
>>>
>>> This looks like an old issue, maybe we won't have this very soon... Is a DIP necessary for such changes?
>>
>> Elapsed time is not always a good indication of the complexity or feasibility of a task, especially in open-source ;)
>>
>> https://github.com/dlang/druntime/pull/2873
>
> P.S. Thanks for reminding me of this issue Bogdan, sometimes communication is all that's necessary :)

This is awesome!
Thank you for the fix!