Is behavior of += intentional for int?

Steven D'Aprano steve at REMOVE-THIS-cybersource.com.au
Sun Aug 30 12:34:17 CEST 2009


On Sun, 30 Aug 2009 02:33:05 -0500, Derek Martin wrote:

> On Sat, Aug 29, 2009 at 07:03:23PM +0000, Steven D'Aprano wrote:
>> On Sat, 29 Aug 2009 11:11:43 -0700, zaur wrote:
>> 
>> > I thought that int as object will stay the same object after += but
>> > with another integer value. My intuition said me that int object
>> > which represent integer value should behave this way.
>> 
>> If it did, then you would have this behaviour:
> 
> No, you wouldn't; the behavior you described is completely different
> from, and incompatible with, what zaur wrote.
> 
> He's saying that instead of thinking the integer value of 3 itself being
> the object, he expected Python's object model would behave as though the
> entity m is the object, and that object exists to contain an integer
> value.

What is "the entity m"?

Is it the name m, as follows?

>>> m = 3  # bind the object 3 to the name m

Or is it the literal "3" (without quotes)?

Or an object holding the value three?

Or something else?


> In that case, m is always m, 
> but it has whatever integer value
> it is told to hold at any point in time.  The self-referential addition
> would access the value of m, add the operand, and store the result back
> in the same object as the object's value.

Ah wait, I think I get it... is m a memory location? So when you say:

m = 3

the memory location that m represents is set to the value 3, and when you 
say:

m += 1

the memory location that m represents is set to the value 4?

That would be how Pascal and C (and presumably other languages) work, but 
not Python or Ruby or even VB (so I'm told) and similar languages. Java 
has a hybrid model, where a few data types (such as ints) are handled 
like C, and everything else is handled like Python. Consistency was never 
Java's strong suit.


> This is not the way Python
> works, but he's saying this is the intuitive behavior.

It isn't intuitive if you've never been exposed to Pascal- or C-like 
languages. If your only programming language was Haskell, the very idea 
of mutating values would be alien. So I guess when you say "the intuitive 
behaviour", what you actually mean is "familiar".


> I happen to
> agree, and argued at length with you and others about that very thing
> months ago, when some other third party posted with that exact same
> confusion.
> 
> By contrast, your description maintains the concept of numerical value
> as object that Python uses, and completely misses the point.  I did find
> the description you gave to be highly enlightening though...  It
> highlighted perfectly, I think, exactly why it is that Python's behavior
> regarding numerical values as objects is *not* intuitive.  Of course,
> intuition is highly subjective.

What exactly is it about Python's behaviour regarding numbers that is not 
intuitive? That you can't do this?

>>> anum = 2
>>> alist = [anum]
>>> anum += 1
>>> print alist  # this doesn't work
[3]

That won't work in any language that I know of -- as far as I am aware, 
the above is impossible in just about every common programming language. 
(My C and VB are nearly non-existent, so I may be wrong about them.) Here 
is Ruby's behaviour:

irb(main):001:0> anum = 2
=> 2
irb(main):002:0> alist = [anum]
=> [2]
irb(main):003:0> anum += 1
=> 3
irb(main):004:0> puts alist
2
=> nil


Just like Python.


> I believe it boils down to this:  People expect that objects they create
> are mutable.

Why would they expect that? Is there any evidence apart from the 
anecdotal complaints of a few people that they expect this? People 
complain equally when they use a mutable default value and it mutates, or 
that they can't use mutable objects as dict keys, so this suggests that 
people expect objects should be immutable and are surprised when they 
change.

If you're going to argue by analogy with the real world (as you do 
further on), I think it's fair to argue that some objects are mutable 
(pieces of rubber that expand into a balloon when you blow into them), 
and some are immutable unless you expend extraordinary effort (rocks). I 
would be gobsmacked if my desk turned pink or changed into an armchair, I 
expect it to be essentially unchanging and immutable. But I fully expect 
a banana to turn black, then squishy, and finally white and fuzzy if I 
leave it long enough.


> At least, unless they specify otherwise.  It is so in some
> other programming languages which people may be likely to be familiar
> with (if they are not taking their first forray into the world of
> computing by learning Python), and even "real world" objects are
> essentially always mutable.
[snip example of a 2002 Buick LeSabre]

Be careful bringing real-world examples into this. People have been 
arguing about identity in the real-world for millennia. See, for example, 
the paradox of my great-grandfather's axe. My great-grandfather's axe is 
still in my family after 80 years, as good as new, although the handle 
has been replaced four times and the head twice. But it's still the same 
axe. An even older example is the paradox of the Ship of Theseus.


> Numbers are fundamentally different from objects.  The number 3 is a
> symbol of the idea of the existence of three countable objects.  It can
> not be changed 

Doesn't this contradict your claim that people expect to be able to 
mutate numbers? That you should be able to do this?

>>> x = 3
>>> id(x)
123456
>>> x += 1
>>> assert x == 4
>>> id(x)
123456


You can't have it both ways -- if people think of objects as mutable, and 
think of numbers as not-objects and unchanging, then why oh why would 
they find Python's numeric behaviour to be unintuitive?


> (though it can be renamed, if you so choose -- just don't
> expect most people to know what you're talking about).  It is
> unintuitive that 3 is an object; 

Says you. People have considered numbers to be eternal, unchanging, 
immutable entities going back to at least Plato. If people are 
comfortable thinking that there are Platonic ideal numbers, why wouldn't 
they think of them represented in computers as immutable objects?

What I think is that some people, such as you and Zaur, have *learned* 
from C-like languages that numbers are mutable not-objects, and you've 
learned it so well that you've forgotten that you ever needed to learn 
it. Consequently what you actually mean when you say Python is 
unintuitive is that Python is not like some other language (and is like 
yet other languages).


> it is rather what we use to describe
> objects -- the value of the object.  It is an abstract concept, and as
> such it is not an object at all.  You cannot hear 3, taste 3, nor smell
> 3.  You can neither see nor touch 3, though you can certainly see 3
> *objects* if they are present, and you can certainly see the symbol '3'
> that we use to represent that idea... but you can not see three itself,
> because there is no such object.

Human beings are excellent at reifying abstract things into (imaginary) 
objects. We talk about letting in the cold (we actually let out the 
heat); we talk about souls and life-force as if they are things; we treat 
justice and love and mercy and hate as palpable things instead of 
emotions and abstract entities. Plato imagined that everything has a 
pure, abstract form, an ideal, and people find that intuitive: we expect 
that there is an archetypal "chair" that embodies everything that is 
chair-like and nothing which is not, if only we could find it.

Numbers are no different. Reifying them into objects comes easy to people 
who haven't learned differently.


> The only way to see three is to
> envision 3 of some object.  The number 3 does not have a value; it IS a
> value (it is the symbolic representation of the value of three).  To say
> that 3 is an object that has a value is a bit like saying the length of
> a car is an object that itself has a length. 

No, the length of a car is an object which *is* a length, it doesn't 
*have* a length.


> It just doesn't compute.

None of this explains why you would expect to be able to mutate the value 
three and turn it into four.

I don't think your argument makes sense -- I think your explanation is 
tangled up in knots. Everything you say about numbers being unchangeable, 
immutable entities supports Python's behaviour to make numbers immutable, 
and yet you argue that since people expect numbers to be unchanging, they 
therefore expect Python ints to be mutable! This makes no sense.


> THAT is why Python's behavior with regard to numerical objects is not
> intuitive, and frankly bizzare to me, and I dare say to others who find
> it so.
> 
> Yes, that's right.  BIZZARE.

I think you have confused yourself. If the number three cannot change, 
and you have a label "m" associated with three, then if you add one to 
it, you *can't* have the three mutate into a four, because numbers cannot 
change. Obviously the label has to detach itself off the three and onto 
the four -- precisely the behaviour Python uses. Plato would surely have 
found this commonsensical, and I would say that it's only those who have 
become used to Pascal-like languages that don't.


-- 
Steven



More information about the Python-list mailing list