Re: implementing escape analysis

[David Daney <ddaney at avtrex dot com>]

Per Bothner wrote:
> David Daney wrote:
>
> > An object allocated by _Jv_AllocObject can not be stack allocated as 
> > it has a finalizer, but one allocated with _Jv_AllocObjectNoFinalizer 
> > can be, so there has to be some differentiation between all these 
> > different cases.
>
>
> Stack allocating objects with finalizers is straight-forward.
> C++ has done it for decades.

Well there you go!

There does have to be quite a bit of communication with the front-end so 
that you know which case is which and how to finalize.

The thing that concerns me is that in an ideal world you might do this 
in the middle-end, but if we wait for this to happen, it could be a long 
wait.

On the other hand if we hacked a limited version of it into the gcj 
front-end, it could be something that is possible for 4.2.

I think all the discussion is great.  But I think the added complexity 
of doing it at the tree-ssa level might be so great that it will never 
happen.

David Daney.