Stephen -

It looks to me not like a bug, but rather that you got bit by one of the
"subtleties" of the ANSI spec.  Specifically, the handler blocks in your
test methods do not terminate in explicit handler actions (e.g., resume,
return, resignal, etc.).  In that case, the action taken is return -
i.e., terminate evaluation of the protected block and return control to
its sender.  When I ran your tests, it looks like that's what's
happening, and that's correct.

Oddly though, when I ran the tests (updated 2.7 image, minus the most
recent exception fix), the output printed in my transcript was what you
said you expected but did *not* get.  This discrepency feels a little
creepy to me, especially around inherently weird places like exceptions,
so I'd appreciate it if you could rerun your tests some time and drop me
a line if you still see the trouble.

Also, shall we take this offline for a bit, so's not to bore the
newsgroup with the details as we try to see what's up here?

Thanks for beating on it a bit ... exceptions (imho) really need to be
100% bug-free, and so far they probably ain't.

Paul
(ps - in case anyone out there is feeling particularly Terse, there's
still too much code in Exception and its friends ...)

Stephen Pair wrote:
> 
> > Stephen:
> >
> > This bug would have left you in the hands of the wrong exception
> > handler, which has nothing to do with method returns per se, so I think
> > the answer is "no".  You could have run into a hole in the
> > ensured-execution machinery, though - I don't specifically know of any
> > bugs in there, but it is one of the more twisted areas in the code, so I
> > wouldn't be surprised to find one.
> >
> > I would appreciate hearing about any remaining problems after you
> > install the fix - test cases that expose bugs are not always easy to
> > think of!
> >
> > Cheers,
> > Paul
> 
> Ok, I built a little test case to see if the problem has been corrected.  I
> ran the test to verify that it reproduces the bug, then I filed in your fix,
> and the bug appears to still be there.
> 
> I've attached a change set that reproduces the bug.  Some instructions are
> in the preamble.  Load it and execute "ExTest test."
> 
> The method #test calls #test2.  After the inner handler handles the
> exception, an inspection of the call stack reveals that the caller of #test2
> is: "[] in MenuItemMorph>>mouseUp:"!  I've put a commented line in #test2
> that you can uncomment to get the debugger that shows this scenario.
> 
> Thus, #test2 never returns execution to #test (and the rest of #test never
> executes).
> 
> I should have reported this issue a few weeks ago, but didn't get around to
> it (sorry).
> 
> Thanks,
> Stephen
> 
>   ------------------------------------------------------------------------
>                        Name: ExceptionTest.cs
>    ExceptionTest.cs    Type: unspecified type (application/octet-stream)
>                    Encoding: quoted-printable