[Python-Dev] Are property descriptors intended to be immutable?

[Guido van Rossum]

On Sat, Feb 7, 2009 at 9:41 AM, Curt Hagenlocher <curt at hagenlocher.org> wrote:
> On Fri, Feb 6, 2009 at 4:04 PM, Guido van Rossum <guido at python.org> wrote:
>> On Fri, Feb 6, 2009 at 2:44 PM, Curt Hagenlocher <curt at hagenlocher.org> wrote:
>>> ...because they're not quite :).  Should I file this as a bug report?
>>
>> No, this is just how it works. I hope they aren't documented as immuable?
>
> Not that I know of :).  But the individual properties of the
> descriptor are all read-only and that the implementations of setter,
> getter and deleter return new objects instead of mutating and
> returning the old descriptor.  So it seemed a little odd that there
> was just one way remaining in which the object could be mutated.

You're basically using an "API" that wasn't meant to be invoked
directly. I don't particularly care that properties can be mutated,
but I don't really care that they should be 100% immutable either. IMO
it's enough that the APIs that exist to be used explicitly don't
modify the property in place; getter and friends return a new object
so that you can override or extend a property in a subclass, and there
obviously you don't want to mutate the base class.

In the balance I'd say that the "backdoor" you found is probably
harmless, and perhaps someone can find a good use for it, so I'm
disinclined to "fix" this.

-- 
--Guido van Rossum (home page: http://www.python.org/~guido/)