[PATCH][TUPLES] Two simple bug fixes.

[Diego Novillo]

On Sun, Apr 13, 2008 at 05:04, Richard Guenther
<richard.guenther@gmail.com> wrote:

>  Can we put an assert guarded by #if ENABLE_CHECKING for the case you ran
>  into in gsi_last?  I also hit similar issues if you split a bb when
>  there are active
>  tsis in that bb (they get invalid bbs and strange things occur).

We can't really assert anything in gsi_last.  gsi_last_bb(bb) is
not semantically equivalent to gsi_last(bb_seq(bb)).

In gsi_last() a NULL sequence will simply return an iterator to
a NULL statement and associated to no basic block.  We cannot
really abort because there are sequences that are created and
manipulated before they are associated to a basic block.

OTOH, gsi_last_bb() will return an iterator to a NULL statement,
but associated to the basic block requested in the argument.
That's what makes it useful for empty blocks.

Doug, could you add a note explaining this difference in
gsi_last()/gsi_first()?


Thanks.  Diego.