Per Bothner writes:
> David Daney wrote:
> > I keep threatening to do this for gcj, but was thinking that the best
> > place to do it may be the front end. Once in tree-ssa form it is more
> > difficult to figure out that a call to _Jv_AllocObject followed by a
> > call to the constructor is the equivalent of the 'new' operator.
>
> Why is that relevant to escape analysis? A constructor is just
> another method, in this context.
Indeed: the only thing that has to be recognized as "special" is a
malloc-like function.