Re: [Sbcl-devel] exhaust.impure

[Nikodemus S. <nik...@ra...>]

On 4 April 2011 19:00, Jim Wise <jw...@dr...> wrote:

>  ::: Running (TRACE-RECURSIVE :ENCAPSULATE NIL)
>  fatal error encountered in SBCL pid 27960:
>  fixup vector 4a1c66a8 has a bad widetag: 38
>
> Any pointers at where to look first are welcome -- I'd guess this is
> similar to the image-corruption failure mode which led to this test and
> the previous being disabled on darwin, but don't know for sure.

If Solaris used SIGILL traps like Darwin, than would almost certainly
be the same issue: the runtime support for SIGILL traps isn't quite
complete (I don't remember the exact issue offhand.)

However, UD2-BREAKPOINTS feature seems to be used only on Darwin, so
it the problem is likely different.

Callchains leading to arch_*_breakpoint and arch_do_displaced_inst and
friends in *-arch.c are probably of interest.

It's also worth checking that the rewritten instructions don't contain
addresses.

Cheers,

 -- nikodemus