inconsistency with += between different types ?

Andreas Leitgeb Andreas.Leitgeb at siemens.at
Wed Aug 7 22:06:42 CEST 2002


Christopher A. Craig <list-python at ccraig.org> wrote:
>> Immutable objects don't have to implement __iadd__ et al. at
>> all and in fact ints don't have __iadd__. Python falls back to calling
>> __add__ and normal assignment automatically.

> While this is true, "fixing" __iadd__ as he requested and retaining
> that behavior wouldn't solve the problem he is trying to address 

I suppose, I am the "he" addressed :-)
Meanwhile, I have (so I think) a clear idea about the interconnectedness 
  of += , __iadd__ and __add__  (and all the other ..=, __i...___,__...__)

currently:
  x+=y means:
    if  __iadd__ defined :
       call  x.__iadd__(y)  # (which is supposed to change x in-place), then
       re-assign the result to x 
        # (in the normal case, the final assignment does nothing, because
        #  "x" (outside) and the return'ed self (from inside) refer to the 
        #  same object.
        # otherwise it most likely causes problems (see: pitfall).)
    elif __add__ defined:
       call x.__add__(y) # (which is supposed NOT to change x in-place) and
       assign the result to x 
        # (possible free'ing the object that was previously referenced by x.)
    else: raise some error  # no addition defined

the suggestion:
  x+=y should mean:
    if  __iadd__ defined :
        just call x.__iadd__(y)  and let it modify x as it pleases
        # nothing else, return value is to be ignored.
        # or perhaps warned about, if it is neither None nor self,
        #   which would indicate a legacy dirty use.
    elif __add__ defined:
        ... # just like it is now.  no change in behaviour in this branch
    else: raise some error  # ...


I hope, it is clear now, what I had in mind with that suggestion, so
we can start discussing the point, rather than speculating about
wrong motivations (,which was generally not a bad thing, *initially*, btw.)

PS: the only (and very weak) counter-argument would be, if some
  class wanted to decide at runtime, whether to behave like immutable
  or not, but this effect could still be achieved by storing away the
  __i...__ methods into a local dict, and writing them back if the object
  should behave mutably again -- not a very likely scenario, anyway.

-- 
class number:
  def __add__(self,x):
    if self.val+x>4: return 'A bright suffusion of yellow'
    else return self.val+x



More information about the Python-list mailing list