>When interpreting the above warning it looks to me like turning on the optimizer 
>masks the fact that some variables are not referenced. I have seen the 
>inverse (ie. variables becoming unused when the code got optimized) but this 
>looks suspicious to me.

No, this is normal.  GCC's unused/uninitialized warnings depend on
dataflow information computed by the optimization pass.

Until last month sometime, there *was* one erroneous warning that
showed up when compiling elink3.o with -O instead of -O2. (the
variable was actually initialized, but GCC couldn't prove it.)

I've since commited a workaround for that.

I've had great success with CGD's mbuf.h patch.  Until the compiler
bug is found and fixed, why don't we commmit that and turn on
_LOSING_COMPILER in the i386 types.h?  I mean, we _know_ the i386
kernel has bugs without that..

--Jonathan