Re: [tree-ssa] Small changes to tree-cfg

[Diego Novillo <dnovillo at redhat dot com>]

On Sat, 2003-09-13 at 14:53, Zdenek Dvorak wrote:

> > define ptbbn
> > set $BLOCK=$arg0
> > set $BB=(( basic_block_info )->data.bb[$BLOCK])
> > set debug_tree_bb ($BB)
> > end
> 
> You really consider this to be argument for "Not really necessary"?
>
Yup.  It's the same wrapper, but instead of adding to cc1's image size,
we do it as a gdb macro.  But, as I said, I don't think it hurts having
the function.  So the change is OK.

> By this argumentation no debugging functions are neccesary, as you
> may inspect the structures directly.
> 
Come on.  That's silly :)

> > Parts of this patch were hardly obvious.  Please revert and address the
> > issues.
> 
> What exactly do you consider unobvious?
> 
The change to FOR_EACH_BB_REVERSE.  Since you implemented it originally,
I wanted you to explain why/how it could fail.  I see now that
remove_unreachable_block() ultimately calls expunge_block(), which
removes the block from the pool.  Therefore, accessing bb->prev_bb after
remove_unreachable_block() implies a reference to freed storage.

That's the kind of analysis I was looking for.

The patch is OK then.


Thanks.  Diego.