Boston D Meetup: Strawman Structs

[Nicholas Wilson via Digitalmars-d-announce]

On Sunday, 23 July 2017 at 02:15:18 UTC, Steven Schveighoffer 
wrote:
> On 7/21/17 8:49 PM, Nicholas Wilson wrote:
>> On Friday, 21 July 2017 at 21:55:01 UTC, Steven Schveighoffer 
>> wrote:
>>> On 7/2/17 6:35 AM, Steven Schveighoffer wrote:
>>>> I'll have a short presentation on a weird trick I discovered 
>>>> while writing some MySQL serialization code. Hope you can 
>>>> attend!
>>>>
>>>> https://www.eventbrite.com/e/d-lang-presentation-strawman-structs-tickets-35120523431
>>>>
>>>
>>> I've set up a live stream for this. No guarantees of the 
>>> quality, it will be audio + slideshow.
>>>
>>> Waiting for people to arrive, so probably won't start until 
>>> at least 6:30.
>>>
>>> https://www.youtube.com/watch?v=ZxzczSDaobw
>>>
>> 
>> Great talk!
>
> Thanks!
>
>> 
>> Regarding the inferred attribute problem with the concepts 
>> like Straw-man usage, this should not be a problem with my 
>> attributes DIP, since all special attributes become normal 
>> attributes and you could just have an AliasSeq of the required 
>> values.
>
> Maybe I'm misunderstanding you, but my concern is that 
> something like this:
>
> struct StrawmanRange(T)
> {
>   ...
>   void popFront() {}
> }
>
> So popFront would be inferred to be pure, @safe, and nothrow. 
> However, since really we want to only do what was specified, we 
> don't want the compiler inferring this. More specifically, I 
> wouldn't want the `implements` function generating requirements 
> that a suitable range struct must have @safe nothrow pure 
> popFront. I don't think introspection can tell whether the 
> attributes were specified or inferred.
>
> I don't see how being able to combine attributes is going to be 
> able to prevent compiler inference of them. Or maybe I am 
> missing something?
>
> -Steve

Its the combining with AliasSeq in conjunction with being normal 
(albeit compiler recognised) attributes that makes it work. It 
doesn't matter what the compiler infers because the struct knows, 
with RequiredAttributes, what attributes are _actually_ required 
E.g.

struct RequiredAttributes(Values...) 
if(AllSatisfy!(isCoreAttributeValue, Values))
{
     alias values = AliasSeq!(Values);
}

  struct StrawmanRange(T)
  {
     @RequiredAttributes!(): // i.e. no attributes required
     // or use
     //@RequiredAttributes!(safe,nothrow,nogc,pure): for very 
strict functions
     // can apply this on a per symbol basis too.

     @property T front();
     bool empty();
     void popFront() {}
  }

and reflect on the RequiredAttributes.values to force the 
"correct" attributes in `implements` and `describeDifferences`.

// roughly and ignoring optional methods
bool Implements(Strawman,T)()
{
     foreach(m; __traits(allMembers, Strawman)
     {
         static if (!hasMember!(T,m)
             return false;
         else static if (!isCovariantWith!(__traits(getMember, T, 
m),getUDA!(__traits(getMember, Strawman, m), RequiredAttributes). 
values)
             return false;
     }
     return true;
}