Re: More miscellaneous patches

[Jason Merrill <jason at cygnus dot com>]

I tracked down the bug.  Here's a short testcase that crashes with
-gdwarf-2 -O.  Making g() inline removes the crash, but the block
structure is still wrong; the block containing g's locals is within
the block for the inlined copy of f.  This happens because:

retrofit_block inserts the inlined block ahead of the locals block,
because it fails to find an insn for the locals block.  This happens
because when we call retrofit_block we're inside a sequence for the
current STMT_EXPR, and we can't see the rest of the function.

identify_blocks therefore assigns the wrong blocks to the various
block notes, and reorder_blocks dutifully fixes things up to reflect
the nesting indicated by the notes.

Your ball.

Jason

struct A { };

inline A f () { return A(); }

template <class T> void g()
{
  A a = f();	
  while (1) { }
}

int 
main()
{
  g<int> ();
}