🚀 go-pugleaf

> From: furnish@dino.ph.utexas.edu (Geoffrey Furnish)
> Date: 07 Sep 1995 03:14:48 GMT
> Message-Id: <FURNISH.95Sep6221449@dino.ph.utexas.edu>
> 
> In article <0kHV_l3_3gX_03eF80@iassf.easams.com.au> rjl@iassf.easams.COM.AU (Rohan LENARD) writes:
> 
> > > The __unwind_function takes a pointer to the throw handler, and is
> > > expected to pop the stack frame that was built to call it, as well as
> > > the frame underneath and then jump to the throw handler.
> 
> In the case of HPPA, at least, these words are quite naive.

Not really, you might be missing the complexity involved in the simple
phrase `pop the stack frame'.  By this I mean, the general case of
poping any frame, that includes restoring saved registers and the
restoring all machines state necessary to resume the frame.

> At least, in my opinion, and in the opinion of others who have
> looked at the problem.  HP exception support, if anyone ever does
> get it working, will not look anything like the short inline asm you
> see for other architectures.

In fact, almost 100% of the unwinders will be lots more complicated.
They have to do all sorts of tricks.  Things like disassembling
instructions and looking for code sequences...  Really icky stuff.

> It will require C code, calls to poorly documented HP compiler
> support functions, an a serious amount of reverse engineering of
> various aspects of the HP stack unwinding mechanism.  It is /not/
> trivial, as this paragraph suggests.

I revised the time estimate to write one of these suckers.  :-)

Also, if anyone is interested in writing an unwinder, please get in
touch with me, as the way in which they are written is changing.  For
example, no more need to play games with the three registers.

The help group is a better place for general discussions, than the
bugs group.