How to forward a Class in {$mode tp} in 2 different 'type's?

[Hartmut]

The documentation says in https://www.freepascal.org/docs-html/current/ref/refse35.html

A class forward definition is simply the name of the class, with the keyword Class, as in the following example:
...
When using a class forward definition, the class must be defined in the same unit, in the same section (interface/implementation). It must not necessarily be defined in the same type section.

But the sentence I have marked green seems to be only valid in {$mode objfpc} and not in {$mode tp}

Code: Pascal  [Select][+][-]

1. {$mode tp}{$H+}      {this gives Error: Type "Class1" is not completely defined}
2. { $mode objfpc}{$H+} {this compiles}
3. {$modeswitch class}  {allows classes in '$mode tp'}
4.  
5. type
6.    Class1 = class; {forward declaration}
7.  
8.    Class2 = class
9.      C1: Class1;
10.    end;
11.  
12. {... a lot of code ...}
13.  
14. type // without this separate 'type' it would compile in {$mode tp}
15.    Class1 = class
16.      C2: Class2;
17.    end;
18.  
19. begin
20. end.

In {$mode tp} the compiler shows Error: Type "Class1" is not completely defined. If I would delete the 2nd 'type', it would work. But to move both declarations into the same 'type' I a) would have to rearrange a lot of Code and b) the readability / structure would become worse.

The documentation of {$mode tp} in https://www.freepascal.org/docs-html/current/prog/progse73.html says nothing about this restriction :-((

My question: is there a {$modeswitch xyz} which I can add in {$mode tp} to solve the problem? Or something else?
Thanks in advance.

[marcov]

No. Turbo Pascal has no classes, so this is not supported behaviour IMHO

That forward class declarations must be resolved within a type block is normal.

[Thaddy]

IIRC forward object declarations are also not allowed in mode TP.
You can tinker with the modeswitches, but then you get kind of a bastard TP mode.

[Hartmut]

No. Turbo Pascal has no classes, so this is not supported behaviour IMHO

To support classes I had added {$modeswitch class}. I did expect, that this would enable classes completely, including forward class definitions in different 'type' sections.

That forward class declarations must be resolved within a type block is normal.

Did you see that the documentation (see link above) for this says:

... It must not necessarily be defined in the same type section.

[Thaddy]

But it must be defined in A type section. Not anywhere else.
Note even forward class declarations is fairly recent anyway.

[Hartmut]

But it must be defined in A type section. Not anywhere else.

As so often your post helps nobody - only you to increase your post counter.
Of course a class must be defined in a type section!
If you would have had a short look at my demo, you would have seen this.

Note even forward class declarations is fairly recent anyway.

That's wrong. In {$mode objfpc} forward class declarations work - also separated in 2 different type sections - even in my very oldest FPC which is 2.6.4 (!) from March 2014. Thats not "fairly recent".

[PascalDragon]

Note even forward class declarations is fairly recent anyway.

Huh? If you consider 20 years fairly recent...

[Bart]

Huh? If you consider 20 years fairly recent...

Well, that just feels like yesterday to me 

Bart

[Thaddy]

Yes, I was not clear. I meant to explain that the following is illegal in mode TP or TP itself:

Code: Pascal  [Select][+][-]

1. {%FAIL%}
2. {$ifdef fpc}{$mode tp}{$endif}
3. { pure TP }
4. type
5.   TMyObject = object;  // try forward
6.   TmyObject2 = object
7.     obj:TMyObject;  // try use it
8.   end;
9.  
10.   TMyObject = object  // try implementation of forward
11.   end;
12. begin
13. end.

As explained, TP has no concept of class. Adding {$modeswitch class} mutilates the intend.
And this is legal code:

Code: Pascal  [Select][+][-]

1. {$ifdef fpc}{$mode objfpc}{$endif}
2. type
3.   TMyObject = class;
4.   TmyObject2 = class
5.     obj:TMyObject;
6.   end;
7.  
8.   TMyObject = class
9.   end;
10. begin
11. end.

Just because of the mode and class concept.
I might add that Harmut's code should work if the modeswitch is in the right place, since this compiles:

Code: Pascal  [Select][+][-]

1. {$ifdef fpc}{$mode tp}{$modeswitch class}{$endif}
2. type
3.   TMyObject = class;
4.   TmyObject2 = class
5.     obj:TMyObject;
6.   end;
7.  
8.   TMyObject = class
9.   end;
10. begin
11. end.

If it is sane, though, I have no opinion, since I also sometimes create bastard modes.
Modeswitches must be added AFTER choosing the default mode, otherwise they get reset.

Since his snippet has the right order I am a bit bemused Hartmut did not get it to work.

[zamtmn]

TP allow only pointer forward declaration

[Hartmut]

Since his snippet has the right order I am a bit bemused Hartmut did not get it to work.

Sorry, again you did not understand the question.

In {$mode tp} the compiler shows Error: Type "Class1" is not completely defined. If I would delete the 2nd 'type', it would work.

As I wrote in my 1st post, class forward definitions do work in {$mode tp}.
But in {$mode tp} they must be defined in the same type section.
While in {$mode objfpc} they must not be defined in the same type section.
The documentation (see link above) for this says:

... It must not necessarily be defined in the same type section.

without including a restriction in {$mode tp}.

TP allow only pointer forward declaration

Sorry, this is wrong, as I wrote in my 1st post and as Thaddy tested in his 3rd example.

Again my question:
is there a {$modeswitch xyz} (or a Compiler switch? or something else?) which I can add in {$mode tp} to allow class forward definitions in {$mode tp} not only in the same type section, but in different type sections, as in {$mode objfpc} is possible and as the documentation says?

[PascalDragon]

Again my question:
is there a {$modeswitch xyz} (or a Compiler switch? or something else?) which I can add in {$mode tp} to allow class forward definitions in {$mode tp} not only in the same type section, but in different type sections, as in {$mode objfpc} is possible and as the documentation says?

No, there is no modeswitch for that, because that is a speciality of the two FPC modes.

[Hartmut]

Thank you PascalDragon for that clear answer.