Re: [tree-ssa] Correct the initialization of an edge flag

[Andrew MacLeod <amacleod at redhat dot com>]

On Mon, 2003-04-14 at 12:50, Diego Novillo wrote:
> On Sun, 2003-04-13 at 19:29, Pop Sébastian wrote:
> > Hi,
> > 
> > This patch marks the following edges as fallthru:
> > 
> >                LOOP_EXPR
> > 	           |
> >                    v
> >              LOOP_EXPR_BODY 
> > 
> > Bootstrapped on i686.  Ok?
> > 
> What are you trying to fix?  Andrew talked about needing fallthru edges
> for one reason or another, but I can't remember.  Andrew what was the
> problem you had run into?
> 
I think one of them was a requirement if we ever decided to delink basic
blocks from the tree chain.. ie, have each basic block stand alone. Then
you needed them to emit RTL properly...  But thats not an issue right
now, nor maybe ever.

The other time was when/if we try to get rid of that NEXT_BLOCK_LINK
stuff of yours in the cfg builder... I think fallthru edges might have
been necessary then when buidling the cfg.  For the moment, we don't use
them either...

All we were looking at making the FALLTHRU edges represent in the tree
structure were tree edges which control falls through with an explicit
transfer of control, for want of a better definition.

so no block has more than one fallthru successor edge, but it can have
more than one predecessor fallthru. ie, the 2 branches of a COND_EXPR
can both fallthru into a common successsor block...

I do not think there is anything in the compiler that is currently using
this, that I can think of on a omnday morning... :-)
Andrew