[fpc-devel] specialization with type alias, bug or feature
Mattias Gaertner
nc-gaertnma at netcologne.de
Fri Aug 18 00:09:14 CEST 2023
On 17.08.23 20:08, Michael Van Canneyt wrote:
> [...]
>> Both FPC and Delphi create different classes for TDoubleBird and
>> TMyDoubleBird.
>>
>> In Delphi they are not assign compatible, in FPC they are. For example:
>> var
>> a: TDoubleBird;
>> b: TMyDoubleBird;
>> begin
>> b:=TMyDoubleBird.Create;
>> a:=b; // forbidden in Delphi
>> writeln(a is TDoubleBird); // writes false
>>
>> Bytewise the assignment is ok, but logic wise "a" is no longer a
>> TDoubleBird.
>>
>> Is this a bug or a feature?
>
> Depends on how you look at it. Double and TMyDouble are assignment
> compatible.
> There is no reason why the same should not hold true for TDoubleBird and
> TMyDoubleBird ?
I guess that is true for specialized arrays, records (non advanced),
proc types and pointers, unless I forgot something (maybe helpers?).
But classes and interfaces have "is", "as", "class of", and "class
vars", where the difference matter.
>[...]
> Of course, the method addresses are different. Other than that and the
> different RTTI (which has 100% the same structure)
The different RTTI has 100% the same structure?
For clarification for others to follow the discussion here is an example:
type
TBird<T> = class(TPersistent)
private
FWing: T;
published
property Wing: T read FWing;
end;
This is allowed in Delphi, but FPC refuses generic templates in published.
The TBird<double>.Wing has a different PPropInfo than
TBird<TMyDate>.Wing. That is probably only relevant in some rare cases.
> I am inclined to go for feature, but maybe there are arguments to tip the
> balance in the direction of bug.
One point where it could wrong, is when you port Delphi code to FPC.
For example this is fine in Delphi:
type
TBird<T> = class
class var Cnt: integer;
end;
TDoubleBird = TBird<double>;
procedure Fly(b: TDoubleBird);
begin
inc(b.Cnt); // under Delphi this must be TBird<double>.Cnt
// not so under FPC
...
if TDoubleBird.Cnt=1 then ...
end;
Mattias
More information about the fpc-devel
mailing list