pmcgee

@Ali Dehban can you expand on that ? I'm trying to picture why if you have an obj ( where the type = IMyInterface<xxx> ) anywhere in your code, you couldn't use obj.DoSomething . Isn't that the point of the interface?

@Ali Dehban I like the idea of the area you were investigating here. Stretching my brain to a function that can return any type is something I have thought about in the last year or so. It took a while before I started to wrap my head around enough it to have questions, 'tho. The first was that IMyInterface<T> = interface function DoSomething: T; end; looks a lot like the definition of an anonymous function ... and then I replaced it with records. But I'll skip that for now. My second eventual question is why have function UseInterface<T> ( obj: IMyInterface<T> ) : T; begin Result := obj.DoSomething; end; vs just ? obj.DoSomething; So, as a first step I ended up with the following : (I realise you would have been simplifying the code above from your real use case) {$APPTYPE CONSOLE} program Project3; uses System.SysUtils, System.Variants, System.Classes, System.Rtti; type IWithAny<T> = interface function DoSomething: T; end; TWithAny<T> = class(TInterfacedObject, IWithAny<T>) function DoSomething : T; end; { TWithAny<T> } function TWithAny<T>.DoSomething: T; begin var val: TValue := TValue.Empty; var typ: T := default(T); case GetTypeKind(T) of tkString, tkUString : val := 'Hello'; tkInteger : val := 20; tkClass : if TValue.From<T>(typ).IsType<TStringList> then begin val := TStringList.Create; val.AsType<TStringList>.Add('Hello from StringList'); end; end; Result := Val.AsType<T>; end; var obj1 : IWithAny< Integer >; obj2 : IWithAny< String >; obj3 : IWithAny< TStringList >; begin obj1 := TWithAny< Integer > .Create; obj2 := TWithAny< String > .Create; obj3 := TWithAny< TStringList >.Create; writeln( obj1.DoSomething ); writeln( obj2.DoSomething ); var lvstr := obj3.DoSomething; writeln(lvstr.Text); lvstr.Free; readln; ReportMemoryLeaksOnShutdown := True; end. What I'd kinda like to see is the interface return a (managed record?) holding the <T>, made to clean up it's own memory.

@EugeneK There is a short section in (at least) the Delphi 11 Alexandria version of Marco Cantu's Handbook (and probably earlier versions) called : Implementation of Anonymous Methods

That wouldn't be strange in C++ ... int i = 42; int main() { const int j = i; const int* p = &j; }

Is it correct that each of your threads adds its own ThreadID to a main-thread Dictionary? Could there be any race issues there, or is access to it via some type of lock?

Is it running very quickly for everyone else as well, now?

I just meant this type of situation where the C declarations are impenetrable, and Delphi is so much clearer:

I have no problem with pointer to record ... I have been pretty scathing about C-style function declarations that are not broken down into more readable sub-types. We could separate ownership from access with something like : begin var o := TRecObject.Create; begin var p:PRec := @o.r; // use p for whatever end o.Free; end That pointer could be copied, passed to functions, whatever ... and simply pass out of scope. It wouldn't have ownership of the data object, and has no responsibility to release it. [edit - forgot] Or of course, you can have methods in the class to control and implement the access to the data object. I think this also highlights that the outer wrapper can be something other than manually managed.

Yes. 100% I'm not arguing about it requiring management as is, ... or added structure to automate that. But of the code below, the new / dispose is (imo) ugly code that is unsuited to 2023, and to the long-term goal of regaining wider recognition of Delphi as a modern and relevant language. My standard rant / pedestal is that over the coming years we need to see Delphi improve it's language .. and it's practice ... to not be left behind by the general progress of other languages. Currently Delphi doesn't really qualify as a good teaching language any more - which I think is really sad. Without some more modern language facilities, it would be unfair to modern (say university level) students, and I'd like to see that change. begin var p:PRec := New(p); try p.x := 5; finally Dispose(p); end; end; begin var o := TRecObject.Create; try o.r.x := 5; finally o.Free; end; end

Yes. I was arguing from a consistency and code-style view point. I don't agree. I don't think it is the generally accepted meaning of 'raw pointer'. Maybe you are thinking of 'void pointer' ? >> A raw pointer is a pointer whose lifetime isn't controlled by an encapsulating object, such as a smart pointer. A raw pointer can be assigned the address of another non-pointer variable, or it can be assigned a value of nullptr. Microsoft Learn - Raw Pointers

Hi Ian. A nullable or option type basically adds a single possible "uninhabited" state to a return type of any kind. You can think of a function that can throw an exception in almost exactly the same way ... it has two possible states ... the normal return or the exception. I was operating on the assumption that the boolean you returned from your original function was indicating whether you successfully created a date range.

Not to be negatively critical, but can I suggest that there's a discussion to be had as to whether out parameters should be an out-dated usage style. @Ian Branch Can I take it that what you want returned from the function is EITHER a failure condition/error result OR a tuple of (startdate, enddate) ? If so, this is really a function with a single return type ... function GetWeekDates(const GivenDate: TDateTime; var SOWDay: string = 'SU') -> FAIL | (startdate, enddate) This could be known as a Nullable type, or Option type ... in Rust it is the Result type. When you return from this function, you either have (effectively) no result, or an inhabited tuple. There are quite a few published examples in Delphi of Nullable types .. .from Allen Bauer through to Spring4D. I'm not sure all of them capture the idea in the most functional way.

I would say, in probably every language, the recommendation should be to code in a way that is familiar and idiomatic for that language. If we wanted stack-allocation and hands-off memory safety, then we can use records .. even custom managed records. But if we want to use heap memory, then I think it only makes sense to do it in the same idiomatic language as all our other Delphi code. I don't think it's a wild opinion in 2023 that pretty much nobody outside of C code should be cooking up raw pointers. At least C++ has unique pointer and shared pointer. I guess we could cook up a smart pointer record to look after the heap-allocated memory ... that would still be avoiding raw pointers. In summary, I think the lesson from the C++ ecosystem is: pointers can be ok ... but {owning, raw} pointers are something to avoid.

Wouldn't this be better expressed as a class containing a record? type TMyRec = record ... end; TMyRecC = class r : TMyRec; end; If C++ can declare "no raw pointers", then we certainly should avoid it in Delphi, right? This way you can maintain the usual convention with TMyRecC.Create and .Free, have constructor & destructor ... or use an interface ... or create your own smart pointer class ...

Well. I learnt about something today. Recursive regexes.