Re: ia64 unwind thoughts

[Richard Henderson <rth at redhat dot com>]

On Fri, Feb 23, 2001 at 07:06:41PM +0000, Jason Merrill wrote:
> If there are no cleanups, we should give a null landing pad offset as well,
> since we don't need to return to the function.

So (@Start == @LPStart && LPad == 0) means keep unwinding?
How abort when @Start != @LPStart?  How can a null landing
pad be represented there?

One could say that LPad must always have a non-zero value,
but I bet that's not what HP had in mind.  Not that it
matters that much, given the other tweeks we'll be making.

> Except that would produce a call to unexpected(), which isn't right either.

Oh, that's right, no-throw is supposed to be terminate(), isn't it.

> BTW, weren't you planning to stick with PC ranges rather than specific call
> sites, anyway?

Yeah.  It's too hard to genericly figure out where to put
(or fudge) the label.  Particularly with targets like MIPS
where the jsr is a macro that expands to a whole series of
insns.

> >> The cleanups don't became handlers; call sites within the inlined function
> >> gain action records that refer to the handlers for the enclosing try
> >> blocks.
> 
> > If there are no enclosing try blocks?  Say if f1 got inlined
> > in the above example?
> 
> Then they still have no action records.

They've got to have something.  To clarify:

	struct A { ~A(); };
	int x;
	void f1()
	{
	  A a;
	  f2 ();
	}
	void f0()
	{
	  f1 ();
	  try {
	    f2 ();
	  } catch(...) {
	    x++;
	  }
	}

after inlining we have 

	  f2();			// L1
	  a.~A();
	  f2();			// L2
	R2:
	  return;
	L1:
	  a.~A();
	  _Unwind_Resume;
	L2:
	  x++;
	  goto R2;
	
So L1 is cleanup only, and L2 is a handler.  As I read the
document, however, both get an action filter of 0.

There are four cases that we want to clearly distinguish:

	(0) Can't throw.
	(1) Can throw, no cleanup or handler this function.
	(2) Can throw, only cleanup this function.
	(3) Can throw, (...) handler.


r~