Andrew, thanks for putting together that hack. I will check it out.

Guido, you lost me at "hate odd punctuation ... maybe a different syntax can be invented". Do you mean introducing a new keyword or implementing this as a function? Or do you mean that some other punctuation might be less odd?

I'm willing to write a PEP, even if it's only purpose is to get shot down.

On Fri, Sep 18, 2015 at 3:45 PM, Guido van Rossum <> wrote:
FWIW, I generally hate odd punctuation like this (@ notwithstanding) but I'm not against the idea itself -- maybe a different syntax can be invented, or maybe I could be persuaded that it's okay.

On Fri, Sep 18, 2015 at 12:28 PM, Andrew Barnert via Python-ideas <> wrote:
On Sep 18, 2015, at 11:21, Trent Nelson <> wrote:
>> On Fri, Sep 18, 2015 at 10:42:59AM -0700, Mark Haase wrote:
>> StackOverflow has many questions
>> <> on the
>> topic of null coalescing operators in Python, but I can't find any
>> discussions of them on this list or in any of the PEPs. Has the
>> addition of null coalescing operators into Python ever been discussed
>> publicly?

I believe it was raised as a side issue during other discussions (conditional expressions, exception-handling expressions, one of the pattern-matching discussions), but I personally can't remember anyone ever writing a serious proposal. I think Armin from PyPy also has a blog post mentioning the idea somewhere, as a spinoff of his arguments against PEP 484 (which turned into a more general "what's wrong with Python's type system and what could be done to fix it). One last place to look, although it'll be harder to search for, is every time people discuss whether things like dict.get are a wart on the language (because there should be a fully general way to do the equivalent) or a feature (because it's actually only useful in a handful of cases, and it's better to mark them explicitly than to try to generalize).

But my guess is that the discussion hasn't actually been had in sufficient depth to avoid having it here. (Although even if I'm right, that doesn't mean more searching isn't worth doing--to find arguments and counter arguments you may have missed, draw parallels to successes and failures in other languages, etc.) And, even if Guido hates the idea out of hand, or someone comes up with a slam-dunk argument against it, this could turn into one of those cases where it's worth someone gathering all the info and shepherding the discussion just to write a PEP for Guido to reject explicitly.

Personally, for whatever my opinion is worth (not that much), I don't have a good opinion on how it would work in Python without seeing lots of serious examples or trying it out. But I think this would be relatively easy to hack in at the tokenizer level with a quick&dirty import hook. I'll attempt it some time this weekend, in hopes that people can play with the feature. Also, it might be possible to do it less hackily with MacroPy (or it might already be part of MacroPy--often Haoyi's time machine is as good as Guido's).

>> Python has an "or" operator that can be used to coalesce false-y
>> values, but it does not have an operator to coalesce "None"
>> exclusively.
> Hmmm, I use this NullObject class when I want to do stuff similar to what
> you've described:

This is a very Smalltalk-y solution, which isn't a bad thing. I think having a singleton instance of NullObject (like None is a singleton instance of NoneType) so you can use is-tests, etc. might make it better, but that's arguable.

The biggest problem is that you have to write (or wrap) every API to return NullObjects instead of None, and likewise to take NullObjects. (And, if you use a PEP 484 checker, it won't understand that an optional int can hold a NullObject.)

Also, there's no way for NullObject to ensure that spam(NullObject) returns NullObject for any function spam (or, more realistically, for any function except special cases, where it's hard to define what counts as a special case but easy to understand intuitively).

And finally, there's no obvious way to make NullObject raise when you want it to raise. With syntax for nil coalescing, this is easy: ?. returns None for None, while . raises AttributeError. With separate types instead, you're putting the distinction at the point (possibly far away) where the value is produced, rather than the point where it's used.

As a side note, my experience in both Smalltalk and C# is that at some point in a large program, I'm going to end up hackily using a distinction between [nil] and nil somewhere because I needed to distinguish between an optional optional spam that "failed" at the top level vs. one that did so at the bottom level. I like the fact that in Haskell or Swift I can actually distinguish "just nil" from "nil" when I need to but usually don't have to (and the code is briefer when I don't have to), but I don't know whether that's actually essential (the [nil]) hack almost always works, and isn't that hard to read if it's used sparsely, which it almost always is).

Python-ideas mailing list
Code of Conduct:

--Guido van Rossum (

Python-ideas mailing list
Code of Conduct:

Mark E. Haase