Re: Anyone got a tree or RTL corruption bug?

["Zack Weinberg" <zackw at stanford dot edu>]

On Mon, May 14, 2001 at 09:44:55AM -0700, Janis Johnson wrote:
...
> use a debugger to call these functions with the wrong
> kind of argument (I've done that a few times lately by accident).

On Mon, May 14, 2001 at 01:46:30PM -0400, DJ Delorie wrote:
> 
> Pass a tree to debug_rtl or an rtl to debug_tree.  I do that by
> accident in gdb all the time.

On Mon, May 14, 2001 at 05:12:10PM -0700, Mike Stump wrote:
> 
> Oh, oh, oh, I have a testcase.  Take a tree and then use pr on it.
> Then, take a rtx and do a pt on it.  I think it is a great testcase!

Unfortunately, getting *that* not to crash is going to take a lot more
work than the simple patch I had in mind ... and that pretty much
means that the simple patch is not going to help much.  I was thinking
I could get away with just checking each pointer, y'see, but the usual
effect of calling debug_tree() on RTL or debug_rtx() on trees is that
it tries to look up garbage as the _code number_.  Then it blows up
because there's no such entry in the table, or because it is reading a
garbage format, or ...

Basically I need an even more thorough set of checking macros than the
existing checking macros, and it's going to take a major overhaul of
print-rtl.c and print-tree.c.

I'm still interested in doing the patch, but don't expect it any time
soon.

-- 
zw    The master came to a yatai which was selling hot dogs. 'What do you
      want on your hot dog?' he was asked. 'Nothing,' he replied. Then the
      hot dog was enlightened.
      	-- a koan by Eric Hallstrom