xscale-elf testsuite results

[Andrew Haley]

Tom Tromey writes:
 > >>>>> "Andrew" == Andrew Haley <aph@redhat.com> writes:
 > 
 > Anthony> I had no problem building an xscale-elf toolchain with this
 > Anthony> flag.  Did you try that?
 > 
 > Andrew> No.
 > 
 > It only worked for AG due to the spec file build bug.  It definitely
 > won't work in general, since the bug is not platform-dependent.
 > 
 > I came up with this patch.  It does work.  I probably need at least a
 > comment explaining what is going on.
 > 
 > I think this solves the problem because the `final' errors can only
 > occur when modifying a field of `this'.  And this patch works by
 > disabling checking if we're doing that.  This is based on the theory
 > that we never need a test to see if `this' is non-null.  But is that
 > theory accurate?  What if we're in a final non-static method?
 > Actually, in that case I think the caller checks.  At least according
 > to parse.y.  So I think this is safe.
 > 
 > Comments?  Anthony, could you try a full xscale-elf rebuild with this?
 > Andrew, Alex, do you think there is a better way to approach this
 > problem?

Oh, I get it now.

I have been persuaded that `this' can never be null, because every
non-static method is always called via an object reference.  To begin
with I couldn't prove that this would never be null, and so I put the
check in the compiler.

I agree that this patch is safe.

Andrew.