Patch: assume-compiled fix

[Andrew Haley]

Tom Tromey writes:
 > This is another -fassume-compiled bug fix.
 > 
 > Suppose some ancestor of the current class is in a no-assume-compiled
 > package.  When generating the vtable for the current class, we
 > shouldn't generate a reference to any symbol in that ancestor class.
 > 
 > This patch fixes the problem by not generating a vtable if any
 > ancestor is not compiled.  Instead the vtable is built at runtime.  We
 > could be a bit more precise and only avoid vtable generation if this
 > class actually inherits a method from an uncompiled ancestor, but that
 > seemed like more work than was useful to me.
 > 
 > Tested on x86 Red Hat Linux 7.3.  Ok for the trunk?

OK.

I assume by "tested" you mean "no regressions".  I suppose we'll need
test cases for -fassume-compiled.

Andrew.