[fpc-devel] generic constraints...

[Michael Van Canneyt]

On Tue, 29 Jul 2025, Mattias Gaertner via fpc-devel wrote:

>> 
>> And each list needs to be
>>    TMyFinalItem1 =
>>         specialize TFeatureXForList<
>>             specialize TFeatureYForList<
>>                 specialize TFeatureZForList< specialize TFpgList<TItem1>
>>    >
>>            >
>>        >;
>
> IMO that does not look human friendly for reading and debugging.

Let's hope it was a contrived example :-)

>
>
>> Currently I can swap the order, and git it working by creating a non 
>> generic base.... But that introduces dependencies that were not wanted...
>> 
>> 
>> Also, the error by the compiler is
>>       Error: Generics without specialization cannot be used as a type 
>> for a variable
>> 
>> But
>>     generic FooChild<F: TFoo> = class(F)
>> 
>> - F is not a variable
>> - TFoo in not a type (for somethnig), TFoo is a constraint
>
> The problem is the ambiguity of TFoo, because Delphi (aka modeswitch) 
> distinguishes TFoo, TFoo<> and TFoo<,>:
>
> type
>   generic TFoo<A> = class
>      procedure Bar; virtual;
>   end;
>   generic TFoo<A,B> = class
>      procedure NoBar; virtual;
>   end;
>
>   generic FooChild<F: TFoo> = class(F)
>      procedure Bar; override;
>   end;
>
> So you have to be more specific:
>
>   generic FooChild<A,F: specialized TFoo<A>> = class(F)
>      procedure Bar; override;
>   end;
>
>
>> If that syntax was allowed, it should mean that
>> -  F must be of any type that is (a descendand of) a specialization of 
>> TFoo.
>> 
>> So this is NOT to allow FooChild to be specialized with another generic, 
>> but with a (any) specialization of that generic.
>
> AFAIK there is currently no "Any" placeholder in constraints. It would 
> be something like
>   generic FooChild<F: specialized TFoo<>> = ...
>
> Although that looks like something was forgotten. Maybe better:
>
>   generic FooChild<F: specialized TFoo<*>> = ...

it seems to me that 'specialized' is not compatible with 'any', 
as specialized needs a specific type: it actually creates the type.

So that would need to become something like
   generic FooChild<F: TFoo<*>> = ..

Michael.