[fpc-pascal]Problem with TReader and integer idents

[Mattias Gaertner]

On Tue, 6 Aug 2002 13:45:00 +0200 (W. Europe Daylight Time)
Michael Van Canneyt <michael.vancanneyt at wisa.be> wrote:

> 
> 
> On Tue, 6 Aug 2002, Mattias Gaertner wrote:
> 
> > Hi all,
> >
> > I have an object stream for a component with a published property Color: TColor. The stream contains the statement:
> > Color = clRed
> > Since 'clRed' is an identifier, TReader lookups for an IdentToInt function via FindIdentToInt declared in classes.pp. FindIdentToInt needs a pointer to a TTypeInfo.
> >
> > The problem is that the TypeInfo pointer differs from unit to unit.
> > A TypeInfo(longint) in classes.pp is not the same as a TypeInfo(longint) in the program.
> > This is incompatible to Delphi.
> > So, the program can't register a valid TypeInfo pointer.
> >
> >
> > I guess, it's not easy to fix this, so I suggest a workaround:
> >
> > The function FindIdentToInt can be extended to not only check for TypeInfo pointers, but also for the values in the typeinfo:
> >
> > function FindIdentToInt(AIntegerType: Pointer): TIdentToInt;
> > var
> >   i: Integer;
> > begin
> >   with IntConstList.LockList do
> >   try
> >     for i := 0 to Count - 1 do
> >       with TIntConst(Items[I]) do begin
> >         if (IntegerType = AIntegerType)
> >         // start of workaround --------------------
> >         or ((PTypeInfo(AIntegerType)^.Name = PTypeInfo(IntegerType)^.Name)
> >           and (PTypeInfo(AIntegerType)^.Kind = PTypeInfo(IntegerType)^.Kind))
> >         // end of workaround --------------------
> >         then
> >           exit(IdentToIntFn);
> >       end;
> >     Result := nil;
> >   finally
> >     IntConstList.UnlockList;
> >   end;
> > end;
> 
> Will this not lead to problems with enumerated values ?
> 
> You can perfectly have the same enumerated name in 2 units for 2
> different enumeration types.
> 
> Unita.EnumA = (enOne,onTwo,EnThree);
> Unitb.EnumB = (enZero,enOne,enTWo);

Here the property types can be EnumA and EnumB, so I think this should work.
But of course you're right, the name and kind is not sufficient to distinguish types.
OTOH FindIdentToInt is currently rather useless. So, I guess no one uses it. With the workaround it works pretty well in the intended way.

 
> So I think a compiler solution is in order, here. Florian, any ideas how
> this should be solved ? Maybe typeinfo should always be included in the
> unit where something is defined ?

Sounds good.

 
> OTOH, longint is a special case, it is an internal compiler type...

Ok, maybe typeinfo(longint) was a bad example. I don't want to register longint.
 

Mattias