Re: mudflap problem

["Frank Ch. Eigler" <fche at redhat dot com>]

Hi -


On Fri, Jan 28, 2005 at 03:20:33PM -0500, Chris Scott wrote:
> [...]  A thought about this. Adding padding 
> would catch iterations, but it would not catch discrete references that 
> step outside the bounds of an object. We discovered the sourceforge 
> bounds checker does not rely on its padding to make this work. [...]
> This is because the sourceforge checker creates an 
> "out of bounds" pointer for i when i goes outside of b. [...]

That's right.  Like the other old bounded-pointers gcc patch (Hi Gary!),
the sourceforge-hosted bounds-checker patch uses "fat" pointers for
some local pointer variables.  This enables this particular check.

Mudflap could be extended transparently to make use of the technique,
and it might make a performance difference (no need to check the lookup
cache) the as well as catching a different (larger?) space of errors.
The necessary transformations could fit well into the tree-ssa framework.


- FChE

Attachment: pgp00000.pgp
Description: PGP signature