Re: [tree-ssa] [PATCH]: Fix DCE not updating bb->end_tree_p

[Daniel Berlin <dberlin at dberlin dot org>]

On Tue, 15 Jul 2003, Andrew MacLeod wrote:

> On Tue, 2003-07-15 at 18:22, Daniel Berlin wrote:
> >
> >
> > On Tue, 15 Jul 2003, Diego Novillo wrote:
> >
> > > On Tue, 2003-07-15 at 18:15, Daniel Berlin wrote:
> > >
> > > > > > +      bb->end_tree_p = bsi_container (bsi);
> > > > > > +    }
> > > > > > +
> > > > > This should be bsi_stmt_ptr (bsi).
> > > > Nope, that won't work, that was my first try.
> > > >
> > > Oh, probably so that we can start reverse iterators at block's end.
> > Right.
> > > OK
> > > then.  Out of curiosity, remember what was the failure?
> > As you mentioned above.  if you point it at the statement, rather than the
> > compound_expr containing it, it can't go backwards.
> >
> > I think bsi_prev is actually broken, in that it doesn't do the opposite of
> > bsi_next/bsi_stmt in this regard (IE it should look in the first operand
> > of a compound_expr, like bsi_stmt would, when trying to find the prev)
> >
> heh?  bsi_prev() simply uses bsi_next() to iterate from bsi_start()
> until it points to the required statement, and then uses whatever the
> iterator was before we stepped into bsi_next() to get to the current
> stmt. So it should be the exact previous iterator...

But it doesn't work always.
Like I said.
Try replacing the loop with bsi_last (bb), and you'll see.