G2 and valgrind

[Chris Shoemaker]

On Sun, Feb 19, 2006 at 02:01:08PM -0500, Phil Longstaff wrote:
> On February 19, 2006 06:43 pm, Derek Atkins wrote:
> > Phil Longstaff <plongstaff at newearth.org> writes:
> > > The attached patch replaces the const 16 with GUID_DATA_SIZE, only
> > > compares the guid data, not the other field in the struct, and removes
> > > the use of the dirent structure in md5 calculation.
> >
> > I've applied everything but the part to remove the dirent/md5 call.
> > We shouldn't reduce our entropy.  We should just add a valgrind
> > suppression for that particular issue.
> >
> > -derek
> 
> No problem.  I wasn't happy with it anyway.  I'm more concerned about the read 
> after the end of a malloc'ed area than about the uninitialized memory.  I'll 
> figure out a patch for the valgrind suppressions and send in a patch.

Been there, done that.  The problem with using valgrind suppressions
for this is that you'd need a supression for every codepath that jumps
on a guid read, like every instance of a ghashtable that uses guid
keys and every place we store a guid and later compare with it.  

I think it's better just to use the valid portion of the dirent.
Something like:

md5_process_bytes(de, strlen(de->d_name), &guid_context);
total += strlen(de->d_name);

-chris