[Numpy-discussion] augmented assignment and in-place operations

Robert Kern robert.kern@gmail....
Wed Oct 29 23:18:26 CDT 2008


On Wed, Oct 29, 2008 at 22:37, Alan G Isaac <aisaac@american.edu> wrote:
> On 10/29/2008 3:43 PM Robert Kern wrote:
>> The defining characteristic is
>> that "x <op>= y" should be equivalent to "x = x <op> y" except
>> possibly for *optional* in-place semantics.
>
> This gets at a bit of the Language Reference that I've
> never understood.
>
>         when possible, the actual operation is performed
>         in-place, meaning that rather than creating a new
>         object and assigning that to the target, the old
>         object is modified instead.
>
> What does that mean?  I assume "when possible" means in part
> "when mutable", ruling out e.g. Python integers or floats.

In part, yes. But also "when possible" excludes in-place matrix
multiplication on ndarrays since we don't, as numpy policy, resize
ndarrays implicitly.

> But the rest does not really seem part of the language, but
> rather seems to be normative?  That is, I could define
> __iadd__ anyway I want.  Is the above saying that I "should"
> define __iadd__ to do an in-place operation "if possible".
> If so, why is such a normative statement part of the
> language reference?  Or is it a statement about the language
> that I'm just not getting?

It is a guide to the intended usage of the feature. You can see other
such guides in the documentation for __repr__, for example. The use of
the word "should" is not rigidly consistent throughout the document.
In other places, it is closer to the meaning of MUST in RFC 2119:

  http://www.faqs.org/rfcs/rfc2119.html

The reason such guides are in the reference manual is because ...
well, where else are you going to put them? Describing the use case of
a feature is, in this case I think, essential to describing the
feature. You have to explain why one would bother defining __iadd__ if
one already had __add__. Similarly, you have to explain why one would
use __repr__ over __str__ or vice versa.

-- 
Robert Kern

"I have come to believe that the whole world is an enigma, a harmless
enigma that is made terrible by our own mad attempt to interpret it as
though it had an underlying truth."
  -- Umberto Eco


More information about the Numpy-discussion mailing list