Re: r2533 - in /1.2: errors.py prompt/interpreter.py relax -- August 11, 2006 - 15:38

> > > There are a couple of reasons I didn't do it this way. One is that
> > > __str__ is a 'special' function as far as Python is concerned, and it
> > > exists for a specific purpose (same with all the double-underscore
> > > functions). Although it's difficult to imagine how it might cause
> > > problems in this case, I'm inclined to avoid messing with it on
> > > principle.
> >
> > Too late, I've already destroyed it ;)  I replaced it so that it
> > returns the specific RelaxError message generated and stored in
> > 'self.text'.  I think the only problem with this special '__str__()'
> > function is that if it generates an error, you're in big trouble.
>
> You havn't destroyed it. The special purpose of self.__str__() is to
> return a string representation of self (ie. the current instance of the
> class). This is exactly what __str__() does, as it stands. You have
> overridden the __str__() inherited from the parent class (Exception) to
> give a more meaningful string representation of the various RelaxErrors.
> This is exactly the intended use of this special function. The mistake
> would be to add additional behaviour to this function which is not
> related to the string representation. To give an example of how it might
> cause confusion, imagine a case where we wanted to downgrade a
> particular RelaxError to a RelaxWarning at run time. To do this, we
> might insert the following:
>
> try:
>     do_something_bugy()
> except RelaxError:
>     err = sys.exc_info()
>     msg = err[1].__str__()
>     warn(RelaxWarning(msg))
> finally:
>     del err
>
> This gets the exception instance from sys.exc_info, and explicitly gets
> the string representation to use it as a warning message, so we get the
> relevant message, even though we havn't raised the exception. Of course
> we dont expect this code to cause a state.save(), as it would if that
> code was included in __str__()

I know, I was just kidding about the destroying bit.  I've done quite
a lot of interesting things with that __str__() function - it's very
temperamental.  I wonder if it's possible to change the RelaxError
behaviour prior to the error being raised?  The raise statements used
to call them probably prevents this so the 'try:' is necessary (then
again I don't know too much about how 'raise' works).  It would be
useful to have an 'except:\n    raise' statement before the 'finally:'
statement to allow all non RelaxErrors to be raised.  Also creating
something like the RelaxWarnError class which prints 'RelaxError:'
rather than 'RelaxWarning:' would be less confusing.

Edward