[foreign-jextract] Duplicated interface declarations when using typedefs for function pointers

[Maurizio Cimadamore]

This seems odd. But I guess I can see why jextract ends up doing that.

I think what jextract *should* do is to reuse the interface for
`callback_t` whenever possible (which should be declared at the
toplevel).

In other words, it seems like jextract is re-generating a new interface
every time it encounters a reference to a function pointer typedef,
which seems a bug.

Maurizio



On Thu, 2021-01-28 at 18:48 +0100, Filip Krakowski wrote:
> Hi,
> 
> I just noticed that jextract generates duplicate code (interfaces)
> for 
> function pointers within each struct instead at the top-level.
> Each interface's name seems to have a suffix based on the occurrence
> of 
> the declaration in the source file.
> 
>     typedef void (*callback_t)(void *arg);
> 
>     typedef struct first_handler {
>          callback_t cb;
>     } first_handler_t;
> 
>     typedef struct second_handler {
>          callback_t cb;
>     } second_handler_t;
> 
> This code generates the following java source code.
> 
>     @C("struct first_handler")public static class first_handler {
>     ...
>          public static interface cb {
> 
>                  void apply(jdk.incubator.foreign.MemoryAddress x0);
> 
>                  public static  @C("callback_t") MemorySegment
>     allocate(cb fi) {
>                      return RuntimeHelper.upcallStub(cb.class, fi,
>     ucx_h_constants_0.cb$FUNC(),
>     "(Ljdk/incubator/foreign/MemoryAddress;)V");
>                  }
> 
>                  public static  @C("callback_t") MemorySegment
>     allocate(cb fi, NativeScope scope) {
>                      return allocate(fi).handoff(scope);
>                  }
>              }
>     ...
>     }
> 
>     @C("struct second_handler")public static class second_handler {
>     ...
>          public static interface cb$0 {
> 
>                  void apply(jdk.incubator.foreign.MemoryAddress x0);
> 
>                  public static  @C("callback_t") MemorySegment
>     allocate(cb$0 fi) {
>                      return RuntimeHelper.upcallStub(cb$0.class, fi,
>     ucx_h_constants_0.cb$0$FUNC(),
>     "(Ljdk/incubator/foreign/MemoryAddress;)V");
>                  }
> 
>                  public static  @C("callback_t") MemorySegment
>     allocate(cb$0 fi, NativeScope scope) {
>                      return allocate(fi).handoff(scope);
>                  }
>              }
>     ...
>     }
> 
>  From my point of view both interfaces are identical. Is there a
> reason 
> for placing them inside the struct class?
> 
> 
> Best regards
> Filip
> 
>