Stefan Behnel wrote:
> Lisandro Dalcin, 12.03.2010 22:33:
>> See the testcase below. If the body of the except clause is empty, 'e'
>> never gets assigned to the exception value, ie...
>>
>> """
>>>>> repr(e)
>> 'ValueError()'
>> """
>>
>> e = None
>> try:
>> raise ValueError
>> except ValueError, e:
>> #a = 1
>> pass
>>
>> Could any of you fix this in order I can properly implement and test
>> "except A as B" ?? The fix should be done in "Nodes.py", at
>> "generate_handling_code()" ... The obvious patch is the one below, but
>> perhaps you can imagine a better way?
>>
>> diff -r 3932a0be3b6b Cython/Compiler/Nodes.py
>> --- a/Cython/Compiler/Nodes.py Thu Mar 11 18:57:06 2010 -0300
>> +++ b/Cython/Compiler/Nodes.py Fri Mar 12 18:30:43 2010 -0300
>> @@ -4497,14 +4497,6 @@
>> else:
>> code.putln("/*except:*/ {")
>>
>> - if not getattr(self.body, 'stats', True):
>> - # most simple case: no exception variable, empty body (pass)
>> - # => reset the exception state, done
>> - code.putln("PyErr_Restore(0,0,0);")
>> - code.put_goto(end_label)
>> - code.putln("}")
>> - return
>> -
>> exc_vars = [code.funcstate.allocate_temp(py_object_type,
>> manage_ref=True)
>> for i in xrange(3)]
>
> Funny. I just tried this with Py3 and it gives me this:
>
>
> Python 3.1.1+ (r311:74480, Nov 2 2009, 15:45:00)
> [GCC 4.4.1] on linux2
> Type "help", "copyright", "credits" or "license" for more information.
> >>> e = None
> >>> try: raise ValueError
> ... except ValueError as e: pass
> ...
> >>> e
> Traceback (most recent call last):
> File "<stdin>", line 1, in <module>
> NameError: name 'e' is not defined
>
>
> So you could argue that Cython is better behaved than Py3 here. ;)
>
> I files a bug report on Py3 for now.
>
> http://bugs.python.org/issue8130
>
> Anyway, given that keeping the variable inside of the except block is the
> expected behaviour in Py3, I think this should not be 'fixed'.
>
>
> > PS: Is this an instance of "premature optimization is the root of all
> > evil" ??
>
> I actually implemented it at the time because Python 3 was supposed to
> behave like this, and the reason for the change in Py3 was that exceptions
> (and their tracebacks, i.e. all impacted frames!) should never be kept
> alive any longer than strictly necessary. The official way to get the
> behaviour you want in Py3 is this:
>
> e = None
> try: raise ValueError
> except ValueError as err:
> e = err
>
> with 'err' not being defined outside of the except block.
Hmm. Reading the response to that bug, I'm worried as to how Cython can
behave as to not mess up either Python 2 or Python 3 code.
Proposal: We simply don't allow variable names in an "as" clause (or
","-claus, obviously) which are also found elsewhere in the local scope
(other as-clauses is OK, obviously).
This makes sure that Cython users write code which complies both with
Python 2 and 3, and can always be backwards-compatibly lifted when
Python 2 is irrelevant.
Any Cython code which breaks because of this would break with Python 3
semantics as well, so I guess that breakage is already made.
Not sure how easy it is to implement though...
--
Dag Sverre
_______________________________________________
Cython-dev mailing list
[email protected]
http://codespeak.net/mailman/listinfo/cython-dev
