Re: non-local exits with signal and condition-case

[Nic Ferrier <nferrier@ferrier.me.uk>]

Stefan Monnier <monnier@iro.umontreal.ca> writes:

> Could you explain in more detail what you do that doesn't work well
> with catch?

I'll use a specific example (but it really is indicative, I find this
happens to me more often that not doing non-local exits).

This is a heavily simplified version of something I just wrote for the
new elnode based emacswiki:

 (defun make-page (page-name)
   (let ((buf (find-file-noselect page-name)))
     (with-current-buffer buf
       (when (re-search-forward "^REDIRECT \\(.*\\)" nil t)
         (throw :redirect (match-string 1))))
     ;; Otherwise we need to send the page
     (start-page buf)
     (send-page (buf->html buf))))

 (defun serve-page (http)
   (let ((page-name (get-page http)))
     (let ((value 
            (catch :redirect
              (make-page page-name))))
       (when value
         (send-redirect http value)))))


Note how the throw is being used to communicate place and value but in
the catch I only need to know that there was an exceptional condition.

This is the essence of the problem I think. catch does not support, of
itself, exceptional conditions. You have to overlay support for them
with extra typing.

In this example, I could be in trouble is make-page ever returns
anything. So I could throw a cons with a type indicator (say, :redirect)
and check for the type in the when. But that is a lot more work.

I've played with making macros around catch but in the end they seem
little better than having a signal. Is there a technical reason why a
signal is bad compared to catch/throw?



> BTW, you might like the `cl-return' macro as well
> (to use within `cl-block's).

I strongly dislike sprinkling cl- namespace through my code. It makes it
much less readable to me.


>>  (defmacro defsignal (err-symbol inherits-list message)
>>    (let ((errv (make-symbol "err-v")))
>>      `(let ((,errv ,err-symbol))
>>         (put ,errv
>>              'error-conditions
>>              (quote ,inherits-list))
>
> You probably want to cons `errv' in front of inherits-list.

I thought about that before I sent it - I think you are probably
right. But my understanding is that you don't need the symbol to be
present; you can disassociate the symbol used to send the signal from
the symbol(s) used to capture it.

I think (as you suggest) a define-signal form should probably not support that
directly because it seems quite counter intuitive.


> FWIW, I wouldn't mind introducing such a macro.  Not for the use of
> non-local exits, but to make it easier to define new error conditions.
>
> I would also welcome changes to make the "error-message" part more
> flexible (e.g. be able to use a function, maybe).
>
>> I'm not sure where to put this.
>
> We could add it to subr.el (tho I'd call it define-signal).

Shall I send patches then?



Nic