Exception confusion

[Guido van Rossum]

Martin Sjögren <martin at strakt.com> writes:

> On Wed, Aug 08, 2001 at 01:41:10PM +0200, Martin von Loewis wrote:
> > Martin Sjögren <martin at strakt.com> writes:
> 
> [snip]
> 
> > III. Raising an exception in the form klass,value is the same
> >      as raising it as klass(value)
> 
> On the contrary, it isn't!
> 
> You're saying that 'raise Exception, var' is the same as 'Exception(var)',
> but what about these two examples:
> 
>   raise Exception, 'foo'
> 
> This is most certainly equivalent to "Exception('foo')" in that this
> holds:
>   Exception('foo').args == ('foo',)
> 
> No worries there!
> 
>   raise Exception, ('foo', 'bar')
> 
> What happens now?  This is NOT equivalent to "Exception(('foo', 'bar'))"
> since:
>   Exception(('foo', 'bar')) == (('foo', 'bar'),)
> whereas "raise Exception, ('foo', 'bar') yields the tuple ('foo', 'bar')
> as exc.arg!
> 
> [snip more]
> 
> All I'm asking is WHY tuples are handled differently?  In my case I have
> an exception that may be raised either with a tuple or a string. Right now
> I have to do this:
> 
> except MyException, exc:
>     if len(exc.args) == 0:
>         arg = None
>     elif len(exc.args) == 1:
>         arg = exc.args[0]
>     else:
>         arg = exc.args
> 
> Why-oh-why do exceptions work like this? :(

(1) Why?  To make it easy to raise an exception with multiple
    arguments without using the E(a, b, c) notation, and also easy
    to raise an exception with a single argument.  This was important
    way back when we were encouraging folks to convert their
    string-based exceptions to class-based exceptions.

(2) Workaround?  When you want to raise an exception with a single
    argument that's a tuple, add a singleton tuple around it:

      raise E, (a,)

    Or, of course, you can write

      raise E(a)

--Guido van Rossum (home page: http://www.python.org/~guido/)