[randomguy1254]

Closures are functions that can capture values/references visible in the scope where the function is defined/instantiated. They are basically a short hand way of creating function objects, functions which have data members.

[Chris2048]

I believe all functions are object in python, so you can add attributes to them, but you'd be better of creating a class with a __call__ method. Relevant scope can be captured in the class __init__.

Classes themselves can hold multiple functions/methods, and if you want a function that carries state, maybe a generator would be useful?

[randomguy1254]

I think this is the point. You can do all these things, but they are all extra hoops you have to jump through when all you want/need is an anonymous on-the-spot closure.

[Chris2048]

I'll grant you that a class isn't an anonymous function, arguably it's the same with lambdas that aren't as powerful as proper function defs.

But I'm not sure this is a problem. In fact, in some ways it may be advantageous e.g. static analysis.

I'd need a use-case for the anonymous closure, but a generator is a basic concept in python, so I son't think it's any trickier than a js function.

I know how in js there is a degree of 'binding' scope and hiding functionality in closures - I think in python (for better or worse) things would by convention be implemented differently.

As an aside, would you say js scope/closure approach is a bit similar to the approach in R?

[randomguy1254]

"lambdas that aren't as powerful as proper function defs" I don't really follow this. I think we should use the right tool for the job. I don't need something that is "more powerful" but comes with extra verbosity and requires me to jump around the source code if I'm not using the extra power.

"In fact, in some ways it may be advantageous e.g. static analysis." Curious why static analysis can't handle closures? Also, I want to use expressive constructs as a programmer, not be arbitrarily limited by what some static analyzer can handle. There may be cases where you are tied to a certain language and analyzer, but in the general case, I don't follow this argument.

"I'd need a use-case for the anonymous closure." I don't want to delve too much into the usefulness of closures, but they definitely exist for a reason. They are useful as parameters to metafunctions for example, which are used extensively in certain programming styles (ie, more functional styles). Closures are basically syntactic sugar. You don't need them, but they can really make things cleaner/simpler in some cases, and are natural constructs to use in certain paradigms. They can also be used in factory functions as a more succinct way to create an object (point might be lost if you are not familiar with this style).

"I know how in js there is a degree of 'binding' scope and hiding functionality in closures." I'm not really talking about this, or JS in particular. I'm really just talking about the concept of a closure which you can find in many languages. Python already has closures (lambda), but why they can only be one expression instead of multiple lines seems somewhat arbitrary, besides maybe ease of implementing the language/problems specific to Python implementation (can you explain this one?). From a purely programming perspective, I shouldn't care about what a language has trouble with, I care about using high level and expressive constructs, I don't want to limit my expressiveness because of some technical issue in a language. Many languages implement closures efficiently.

"As an aside, would you say js scope/closure approach is a bit similar to the approach in R?" Haven't really used R, but in the languages I have used with closures, they all are basically the same. Really the wikipedia article on closures captures this.

[Chris2048]

> jump around the source code

A named function should be easier to track due to it being named, and properly defined in one place.

I believe debugging can be harder when you see some random anonymous function and have no idea where it is defined (because, I think, lambdas don't hold their definition line-nos?)

Since we're talking about multi-line functions, the added verbosity wouldn't be a lot. Additionally, you would be forced not to nest functions too much.

> why static analysis can't handle closures

Would it require tracking run-time state/modifications to the closure?

> I want to use expressive constructs as a programmer, not be arbitrarily limited by what some static analyzer can handle.

You could argue the same for any language with types - why should I be limited by the constraints of the type system?

Because it's easier to verify the code, and it doesn't actually limit what's possible to develop, only how it must be developed (in a more explicit style).

I see it as putting an extra burden on the developer, which makes it harder for sure, much as writing tests puts an extra burden on the developer.

> why they can only be one expression

http://www.artima.com/weblogs/viewpost.jsp?thread=147358

Seems like it is mainly a technical decision.

> I don't want to limit my expressiveness because of some technical issue in a language

I feel this is a problem with js (versus, say, Java) - faster to write, harder to analyse. What's so important about expressiveness that you wouldn't want the code easier to maintain?

> Many languages implement closures efficiently

I can't really complain about this wrt Python when I'm not sure what it would add. Even if lambdas could be multi-line, I would not want to use them for sake of their parameter-binding semantics - in fact, I consider it a python gotcha. I