[C++ PATCH] Don't create an INTEGER_CST for aggregates (empty structs)

[Paolo Carlini]

Hi everyone, and thanks for the interesting messages!

I'm jumping in only now, sorry...

>>(snip)...
>>whereby you change the function type and cause link failures.
>>
>true, this fix is no good for so6, but if a middle-end fix can't be found,
>the problem can be mostly removed in so-7 (and perhaps should be fixed
>anyway, dummy parameters aren't necessary to do dispatch).
>  
>
Indeed, I think there is a rather general consensus, that we should lean 
toward avoiding dummy parameters also for other reasons (see, e.g., the 
comment at the beginning of cpp_type_traits.h) but I had not considered 
before that this could be also useful from the point of view of the 
quality of the final code! Very interesting stuff! I encourage you to 
work out something concrete, maybe on top of your last patch, which I 
keep installed in my local v7 tree.

In the meanwhile I will try to summarize Gaby's observations in the 
audit trail of 20408.

In my opinion, if we can avoid the (small) code pessimization caused by 
the empty classes in most of the cases *and* we keep the pressure on 
20408, it's safe to proceed with further refactoring in the algorithm area.

Thanks again, for now,
Paolo.