Re: recent regression

[Bryce McKinlay <mckinlay at redhat dot com>]

Ranjit Mathew wrote:

> This is because read_class() isn't quite equipped to handle source
> files that have been parsed already and it unnecessarily calls
> find_class() that emits this warning all over again.
>
> I had submitted this patch to resolve this issue a few days
> back:
>
>  http://gcc.gnu.org/ml/gcc-patches/2004-06/msg01589.html
>
> (For some reason this mail didn't get archived in the java-patches
> archives.)
>  

Weird. I never received it in java-patches either.

> Note that some time after the original code for parsing/laying-out
> was written, the meaning of CLASS_LOADED_P was changed:
>
>  http://gcc.gnu.org/ml/java/2001-04/msg00114.html
>
> and CLASS_PARSED_P was introduced. This meant that if you
> read code like the following (that's quite common throughout
> the front end):
>
>  if (! CLASS_LOADED_P (cls))
>    load_class (cls);
>
> and naively assumed that after this CLASS_LOADED_P (cls) would
> be true, you'd be wrong. :-/
>  

Yeah. Code along the lines of the following, which ought to guarantee 
that CLASS_LOADED_P gets set, is reasonably common:

  if (! CLASS_LOADED_P (class_decl))
   {
     if (CLASS_FROM_SOURCE_P (class_decl))
       safe_layout_class (class_decl);
     else
       load_class (class_decl, 1);
   }

I wonder if this functionality should just be encapsulated into 
load_class, or some other new function - it would simplify code at the 
call sites and also guarantee that CLASS_LOADED_P is true upon 
returning. Do we ever really want to load a class and not lay it out?

Regards

Bryce