[OpenAFS-devel] Aliasing bugs

[Benjamin Kaduk]

On Tue, 12 Jul 2011, Garrett Wollman wrote:

> Has anyone taken a look at the large number of pointer-aliasing bugs
> in the current codebase?  GCC emits a warning for these (if you
> configure OpenAFS with --enable-warnings), and it does not appear that
> anything is compiled with -fno-strict-aliasing to disable the
> optimizations that causes these bugs to be noticeable, unless
> compiling for Linux.

I've run into a few while working up my bsd.kmod.mk branch for the freebsd 
kernel module (which uses -Werror); staged fixes are in my working tree 
but have not been submitted.

A few of the errors are more structural in nature, e.g. the "unique 
identifier" for the vop_advlock routines that always has a 32-bit field 
but sometimes tries to squash a 64-bit value into it.

I'm sure there are a bunch more that I haven't explicitly run into, yet.

-Ben