Why doesn't dbus-handle-event catch all errors?

Why doesn't dbus-handle-event catch all errors?

[Daniel Colascione]

dbus-handle-event uses a condition-case that catches dbus-error, but 
errors in general. Why don't we send all lisp errors from handler 
functions back to callers as dbus errors?

Re: Why doesn't dbus-handle-event catch all errors?

[Michael Albinus]

Daniel Colascione <dancol@dancol.org> writes:

> dbus-handle-event uses a condition-case that catches dbus-error, but
> errors in general. Why don't we send all lisp errors from handler
> functions back to callers as dbus errors?

Because they aren't dbus errors. If your handler has some stupid code
like `(car 42)', the resulting lisp error `(wrong-type-argument listp 42)'
shouldn't be shown as *dbus* error.

You're free to write a handler which returns any error as dbus error, if
your program logic requires this.

Best regards, Michael.

Re: Why doesn't dbus-handle-event catch all errors?

[Daniel Colascione]

On 02/20/2014 11:22 PM, Michael Albinus wrote:
> Daniel Colascione <dancol@dancol.org> writes:
>
>> dbus-handle-event uses a condition-case that catches dbus-error, but
>> errors in general. Why don't we send all lisp errors from handler
>> functions back to callers as dbus errors?
>
> Because they aren't dbus errors. If your handler has some stupid code
> like `(car 42)', the resulting lisp error `(wrong-type-argument listp 42)'
> shouldn't be shown as *dbus* error.

Any nonlocal exit of a dbus handler message handler ought to be a dbus 
error: that nonlocal exit results in a program that expects a dbus reply 
not getting one at all.

> You're free to write a handler which returns any error as dbus error, if
> your program logic requires this.

The only reasonable implementation of a dbus handler is one that 
transforms lisp errors to dbus errors, so why not do that automatically 
in dbus.el? Silently swallowing lisp errors is the less common case, and 
if a program needs to do that, it can wrap its logic in ignore-errors 
inside the handler.

Re: Why doesn't dbus-handle-event catch all errors?

[Michael Albinus]

Daniel Colascione <dancol@dancol.org> writes:

>> You're free to write a handler which returns any error as dbus error, if
>> your program logic requires this.
>
> The only reasonable implementation of a dbus handler is one that
> transforms lisp errors to dbus errors, so why not do that
> automatically in dbus.el? Silently swallowing lisp errors is the less
> common case, and if a program needs to do that, it can wrap its logic
> in ignore-errors inside the handler.

There is the special error type `dbus-error'. Any handler could catch
lisp errors and transform them into something which is understandable to
another process. You do not want to send "(wrong-type-argument listp
42)" or something like this to another process, which isn't aware of Lisp.

Your handler shall include something like

(defun my-dbus-method-handler (&rest args)
  (condition-case err
      (...)
    (error (signal 'dbus-error (list "Something went wrong, please try it again")))))

Best regards, Michael.

Re: Why doesn't dbus-handle-event catch all errors?

[Daniel Colascione]

On 02/24/2014 12:37 AM, Michael Albinus wrote:
> Daniel Colascione <dancol@dancol.org> writes:
>
>>> You're free to write a handler which returns any error as dbus error, if
>>> your program logic requires this.
>>
>> The only reasonable implementation of a dbus handler is one that
>> transforms lisp errors to dbus errors, so why not do that
>> automatically in dbus.el? Silently swallowing lisp errors is the less
>> common case, and if a program needs to do that, it can wrap its logic
>> in ignore-errors inside the handler.
>
> There is the special error type `dbus-error'. Any handler could catch
> lisp errors and transform them into something which is understandable to
> another process. You do not want to send "(wrong-type-argument listp
> 42)" or something like this to another process, which isn't aware of Lisp.

Why not? It's more informative than "something went wrong".

>
> Your handler shall include something like
>
> (defun my-dbus-method-handler (&rest args)
>    (condition-case err
>        (...)
>      (error (signal 'dbus-error (list "Something went wrong, please try it again")))))
>

Sure. My point is that the framework should provide *something*, even if 
it's just the above, instead of silently dropping lisp errors and not 
telling callers that anything went wrong. Handlers that want to swallow 
errors can do so explicitly.