Path: news.daimi.aau.dk!news.uni-c.dk!newsfeed.sunet.se!news00.sunet.se!sunic!news.sprintlink.net!news.up.net!news.mtu.edu!msunews!agate!howland.reston.ans.net!tank.news.pipex.net!pipex!dispatch.news.demon.net!demon!sunsite.doc.ic.ac.uk!yama.mcc.ac.uk!news.salford.ac.uk!aber!not-for-mail
From: pcg@aber.ac.uk (Piercarlo Grandi)
Newsgroups: comp.object,comp.lang.eiffel,comp.lang.c++,comp.lang.beta,comp.lang.java,comp.lang.sather
Subject: Re: What Should An Exception Handling Do? -- Clarification of rules
Date: 09 Apr 1996 19:34:02 +0100
Organization: Prifysgol Cymru, Aberystwyth
Lines: 150
Sender: pcg@osfb.aber.ac.uk
Message-ID: <vwjk9zpckmd.fsf@osfb.aber.ac.uk>
References: <4irn11$7ln@mimas.brunel.ac.uk> <4j03p4$fbt@hoho.quake.net>
	<Doq3sv.MzA@research.att.com> <vwjg2ao6hqp.fsf@osfb.aber.ac.uk>
	<4jqjvg$d9@glympton.airtime.co.uk>
Reply-To: pcg@aber.ac.uk (Piercarlo Grandi)
NNTP-Posting-Host: osfb.aber.ac.uk
In-reply-to: wysiwyg@glympton.airtime.co.uk's message of 2 Apr 1996 08:11:44 +0100
X-Newsreader: Gnus v5.0.15
Xref: news.daimi.aau.dk comp.object:53965 comp.lang.eiffel:22721 comp.lang.c++:177781 comp.lang.beta:10719 comp.lang.java:32793 comp.lang.sather:12432

>>> On 2 Apr 1996 08:11:44 +0100, wysiwyg@glympton.airtime.co.uk (Adam L
>>> Rice) said:

pcg> The exception facility of e.g. C++ is merely a (rather ad-hoc and
pcg> opaque) way of allowing the definition of what are in effect
pcg> dynamically scoped procedure names, forcibly coupled with non-local
pcg> control transfer.

wysiwyg> So, let me see if I've understood. What you're basically saying
wysiwyg> is that instead of doing (in Java):

wysiwyg> try {
wysiwyg>   sock = new Socket("wallawalla.mit.edu", 3456);
wysiwyg> } catch (IOException e) {
wysiwyg>   ... // handle the exception
wysiwyg> }
wysiwyg> ... // use the socket

wysiwyg> we should do something like:

wysiwyg> if (Socket.ConnectWillWork("wallawalla.mit.edu", 3456) {
wysiwyg>   sock = new Socket("wallawalla.mit.edu", 3456);
wysiwyg>   ... // use the socket
wysiwyg> } else {
wysiwyg>   ... // handle the exception
wysiwyg> }

wysiwyg> Hmmmm... that's a lovely code style! I think we can call this techique
wysiwyg> 'APIx2', for obvious reasons.

wysiwyg> But I'm not entirely clear, it might be you were saying we should do:

wysiwyg> sock = new Socket("wallawalla.mit.edu", 3456, new CantConnectHandler());
wysiwyg> if (!CantConnectHandler.ConnectFailed) {
wysiwyg>   ... // use the socket
wysiwyg> }

You are making all this up, and stupidly too, for neither of the
examples you provide are anywhere similar to the ones I have given (for
the ones I have given are examples of how things would look like at the
place where the exception is *raised*, while your examples are about the
place where it is *detected*), and neither exhibits dynamically scoped
procedure names or non local control transfers, which are part of my
definition. Amazing ability to misread! (and yes, I apologize if what I
wrote was opaque, but I think I was darn well unambiguous).

Perhaps you should reread more carefully what I have written, and
perhaps refresh your memory on the notions of dynamic scoping and non
local control transfer and how they look like.

To help you see the light, here is your example:

wysiwyg> try {
wysiwyg>   sock = new Socket("wallawalla.mit.edu", 3456);
wysiwyg> } catch (IOException e) {
wysiwyg>   ... // handle the exception
wysiwyg> }
wysiwyg> ... // use the socket

rewritten using something less ``ad hoc'' than 'try', and with dynamic
scoping and non local control transfer as I had written:

  jmp_buf bailOutTo;

  static void localHandler(IOException e)
  {
    .... // handle the ``exception''
    longjmp(bailOutTo);
  }

  void try()
  {
    dynamic void (*IOExceptionHandler)(IOException) = localHandler;

    sock = new Socket("wallawalla.mit.edu", 3456);
    .... // use the socket
  }

where presumably 'bailOutTo' is initialized to some recovery point
somewhere.

'dynamic' would be new storage class, much like 'extern', but with
dynamic scoping within blocks.

The 'Socket' constructor would contain some bit of code like:

  dynamic void (*IOExceptionHandler)(IOException) = DefaultHandler;

  Socket::Socket(String s,int p)
  {
    ....
    if (nogood)
    {
      // we call the user's routine, equivalent to a 'raise'
      (*IOExceptionHandler)(new IOException(....));
      abort();
    }
    ....
  }

Naturally one would write it a bit better, but I have written it this
way to show the parallel between the 'try' construction and the more
linear one based on dynamic scoping and non local control transfers.

BTW, this style of ``exception handling'' has been rather the norm in
Lisp/Scheme for the past few dozen years, and could be added to
C++/Java/whatever in an afternoon (indeed 'dynamic' can be _simulated_
tolerably well by a few lines of macros in C++).

Dynamic scoping is useful BTW not just for providing ``exception
handling'', but for a number of other uses, all those like ``exception
handling'' where code semantics must depend on the dynamic context and
not the static context, for example when there are runtime
parameters. Currently this must be inelegantly simulated with idioms
like:

  SomeType GlobalParameter = defaultValue;

    ....

    {
      SomeType savedGlobal = GlobalParameter;
      GlobalParameter = localValue;

      ....

      GlobalParameter = savedGlobal;
    }

which if of course nothing but a manual (and hazardous) implementation
of dynamic scoping for GlobalParameter, and that is seen rather often in
C and C++ programs, instead of:

  SomeType Parameter = defaultValue;
    ....

    {
      dynamic SomeType Parameter = localValue;

      ....
    }

BTW, I really hate ``exception handling'', for in cases like say 0/0
there is not exception, jsut the inability to prescribe a particular
semantics statically. I think that something like ``deferred semantics''
would be a much more descriptive title.

PS instead of 'dynamic' I think that 'extern auto' would be OK, and
rather descriptive too, too bad that a C/C++ declaration can only have
one storage class specifier.