[fpc-devel] Questions on RTTI

[Sven Barth]

Am 23.10.2026 um 10:48 schrieb Martin Frb via fpc-devel:
> - A class itself has always some RTTI of its own. (but that does not 
> describe all the fields/members)

Correct

> - other types may also have RTTI (even if they are outside a published 
> section, or even outside a class)

All other types also have RTTI (except *currently* TP-style objects and 
file types, but I do intend to change that). In addition to that managed 
types may have a Init RTTI that's different from the full RTTI.

Whether the RTTI is included in the final binary or not depends on 
whether it's used (e.g. by using TypeInfo() or indirectly by other RTTI) 
and the linker settings (smart linking).

> - fields/members can be found via RTTI  if they are published. That is 
> the field/member can be looked up by name within the RTTI of the class.
>
>    (and that returns the RTTI for the field's type / and the location 
> or getter/setter of the field)

Correct. Though this is also the case if extended RTTI for 
methods/fields/properties is enabled for the corresponding visibility 
sections while a class, record or object is parsed.

> So for the "published" case the DW_TAG_Member needs to have the flag 
> (indicating the presence of RTTI)
>
> (actually indicating that the containing type has type info for that 
> field, and its type)
>
>
> I'll have to double check the below with the fpc team, to make sure I 
> got the nitty gritty parts correct.
>
>
> RTTI may also be available for a type. Then the type would want to 
> have that flag, but that would not relate to the published concept.
>
> Some types have always RTTI for the type. Such types could be declared 
> scoped inside a class. But that would not mean they are published.

The RTTI for types also exists for non-structured types, so it makes no 
sense to put the RTTI below the type.

>
>
> Afaik (and as I said, I am double checking that)
>
> type
>
>    TFoo = class
>
>    public type
>
>        TMyEnum =(mon,tue,wed,thu,fri);
>
>   public
>
>        FMyAnonEnumVal = (Jan, Feb, Mar, Apr);
>
>    published
>
>        Value: integer;
>
>    end;
>
>
> TMyEnum has rtti for the type. But can't be looked up by name. (afaik 
> enums always produce RTTI, becaues they can be printed (with writeln) 
> using the name, rather than the stored ordinal val)

Once we have support for iterating through all types it would indeed be 
possible to look up a type's RTTI by name (assuming it's linked in).

Regards,
Sven