Re: [tree-ssa] Make tsi insert iterators handle chains of COMPOUND_EXPRs

[Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>]

Hello,

> On Thu, Aug 28, 2003 at 09:53:23PM +0200, Zdenek Dvorak wrote:
> >     if (mode == TSI_SAME_STMT)
> > !     i->tp = &(TREE_OPERAND (*last, 1));
> > !   else if (mode == TSI_NEW_STMT)
> > !     i->tp = last != &t ? last : i->tp;
> [...]
> >         if (mode == TSI_NEW_STMT)
> >   	i->tp = &(TREE_OPERAND (*(i->tp), 1));
> > +       else if (mode == TSI_CHAIN_END)
> > + 	i->tp = last != &t ? last : &(TREE_OPERAND (*(i->tp), 1));
> 
> Correct me if I'm wrong.  It appears as if in the first case
> TSI_NEW_STMT leaves the pointer at the end of the chain that
> we inserted, and the second case leaves the pointer at the
> beginning.

the behavior is this:

for tsi_link_after:
  TSI_NEW_STMT -- at start of the chain
  TSI_CHAIN_END -- at the end of the chain

for tsi_link_before
  TSI_NEW_STMT -- at the end of the chain
  TSI_CHAIN_END -- at the start of the chain

while this may seem strange, it is intentional; with this behavior,
tsi_link_xxx (tsi, a, TSI_CHAIN_END);
tsi_link_xxx (tsi, b, TSI_CHAIN_END);

will link chain b to the correct position wrto chain a, regardless of
what xxx is.

(the behavior of TSI_NEXT_STMT is then set to the only senseful choice
that makes it to behave as now when we insert only single statement).

Zdenek