What's the story with @nogc exceptions?
There was DIP1008 or whatever, I didn't follow it because I never use
exceptions... but I kinda need it.

IIRC, the idea was the throw statement transfer ownership of the exception to the runtime, and then use a `scope` catch block, such that the runtime can be sure it owns the only reference at the end of the catch block and it can free the exception rather than transfer it to the GC?

How did it fail?

On Sunday, 19 May 2019 at 04:24:39 UTC, Manu wrote:
> What's the story with @nogc exceptions?
> There was DIP1008 or whatever, I didn't follow it because I never use
> exceptions... but I kinda need it.

It used to not work, but it does now.

> IIRC, the idea was the throw statement

Specifically it works with `throw new [My]Exception(args);` and only that form (since otherwise you could leak it).

> transfer ownership of the exception to the runtime, and then use a `scope` catch block, such that the runtime can be sure it owns the only reference

Well 	
    try throw new Exception("foo");
    catch(scope Exception ex) {}

fails with

Error: basic type expected, not scope
Error: found scope when expecting )
Error: semicolon expected, not )
Error: found ) instead of statement
Error: unrecognized declaration

but I think that was done for compatibility with other code that could store the exception.
But regardless

    try throw new Exception("foo");
    catch(Exception ex) {}

will allocate a reference counted exception with malloc(?) (you can probably edit druntime to make it use some other scheme) which will deallocate at the end of the scope.

> at the end of the catch block and it can free the exception rather than transfer it to the GC?

> How did it fail?

It used to not actually do the thing it was supposed to do (as in it would still call the GC). It was fixed.

On Sat, May 18, 2019 at 9:55 PM Nicholas Wilson via Digitalmars-d <digitalmars-d@puremagic.com> wrote:
>
> On Sunday, 19 May 2019 at 04:24:39 UTC, Manu wrote:
> > What's the story with @nogc exceptions?
> > There was DIP1008 or whatever, I didn't follow it because I
> > never use
> > exceptions... but I kinda need it.
>
> It used to not work, but it does now.

Okay...How do I do it? I just get:
  throw new Exception("ex");
  > error : cannot use `new` in `@nogc` function

On Sunday, 19 May 2019 at 05:05:32 UTC, Manu wrote:
> Okay...How do I do it? I just get:
>   throw new Exception("ex");
>   > error : cannot use `new` in `@nogc` function

pass -preview=dip1008 to the compiler

On Sat, May 18, 2019 at 10:20 PM Nicholas Wilson via Digitalmars-d <digitalmars-d@puremagic.com> wrote:
>
> On Sunday, 19 May 2019 at 05:05:32 UTC, Manu wrote:
> > Okay...How do I do it? I just get:
> >   throw new Exception("ex");
> >   > error : cannot use `new` in `@nogc` function
>
> pass -preview=dip1008 to the compiler

Oh right. Is this a breaking change or something?

On Sunday, 19 May 2019 at 05:59:11 UTC, Manu wrote:
> On Sat, May 18, 2019 at 10:20 PM Nicholas Wilson via Digitalmars-d <digitalmars-d@puremagic.com> wrote:
>>
>> On Sunday, 19 May 2019 at 05:05:32 UTC, Manu wrote:
>> > Okay...How do I do it? I just get:
>> >   throw new Exception("ex");
>> >   > error : cannot use `new` in `@nogc` function
>>
>> pass -preview=dip1008 to the compiler
>
> Oh right. Is this a breaking change or something?

Yes, at minimum it changes the type checking of the `throw new <Ex>` expression (e.g. it's no longer not @nogc). It also makes such exceptions not garbage collected, but reference counted, so the user should not keep an exta reference around as the compiler and the runtime can't track that.
But most of all, such changes needs to be tested with a larger set of third-party libraries before we can consider tuning them on by default. Some projects have already started doing that, but we're still in the early stages.

On Sunday, 19 May 2019 at 06:09:41 UTC, Petar Kirov [ZombineDev] wrote:
> On Sunday, 19 May 2019 at 05:59:11 UTC, Manu wrote:
>> On Sat, May 18, 2019 at 10:20 PM Nicholas Wilson via Digitalmars-d <digitalmars-d@puremagic.com> wrote:
>>>
>>> On Sunday, 19 May 2019 at 05:05:32 UTC, Manu wrote:
>>> >   > [...]
>>>
>>> pass -preview=dip1008 to the compiler
>>
>> Oh right. Is this a breaking change or something?
>
> Yes, at minimum it changes the type checking of the `throw new <Ex>` expression (e.g. it's no longer not @nogc). It also makes such exceptions not garbage collected, but reference counted, so the user should not keep an exta reference around as the compiler and the runtime can't track that.
> But most of all, such changes needs to be tested with a larger set of third-party libraries before we can consider tuning them on by default. Some projects have already started doing that, but we're still in the early stages.

Well as long as druntime and Phobos don't compile with it, it's unlikely that third-party projects will pick it up.

See e.g. https://github.com/dlang/dmd/pull/8508