Hacker News Reader

Comment by jarcane | Hacker News Reader

Python is deliberately bad at functional programming by design. Guido has decreed it.

It seems to be pretty good at map, filter, any, all, reversed, reduce, zip, list comprehensions, dict comprehensions, lambdas, and closures.

Granted, it would be unusual to see all, or even most, of those in one place, but they see use enough that calling python bad a functional programming is a little odd. I'd say it's pretty good at functional work in the small while somewhat ill suited in the large.

Just because it has those functions doesn't make it good at them. Have you ever seen a Pythonic program converted into functional style? It looks horrendous.

I would imagine it depend on what extreme you take it to. You don't have pattern matching or tail recursion, which make certain styles impossible.

There's a much better discussion over here

http://stackoverflow.com/questions/1017621/why-isnt-python-v...

If you define "functional style" as using immutable data structures, higher-order functions, avoiding side-effects and using referentially transparent variables, I think Python programs look fine, and I try to write all the code I can this way (although practicality beats purity).

If you define "functional style" as writing Lisp or Haskell programs using Python syntax, I agree they look horrendous.

Or you can use Hy (http://hylang.org) and write Python in Lisp syntax.

Where did you get immutable data structures from in Python, and what do you do when every third party library you interact with uses mutable ones?

Besides tuples, I mostly treat the standard data structures as if they were immutable, without relying on them actually being :) As for third-party libraries, very few of them actually mutate structures passed by my code, in my experience; if they do, I simply make a copy before passing it.

> Where did you get immutable data structures from in Python

Here's my favored source (in the Standard Library) https://docs.python.org/2/library/collections.html#collectio...

> and what do you do when every third party library you interact with uses mutable ones?

That's situational; which library are you having a problem with? I haven't really had that problem (okay, sometimes the tooling around web frameworks do stupid things, but that's not common).

Note that just because something is mutable doesn't mean you have to mutate it. Yes, Python doesn't as widely enforce non-mutation as Haskell, OCaml, etc., but that doesn't mean that suddenly all the Python programmers lost their minds and started mutating all the things. We're not savages.

> Note that just because something is mutable doesn't mean you have to mutate it

I agree, and I don't mutate it, but my third party libraries and my collegues functions will. Not all of them, but enough to introduce annoying bugs.

This is why I asked, "Which library are you having a problem with?"

You've claimed that third party libraries and your colleague's functions will mutate the data structures you pass them, but this has very rarely been my experience.

You must have much better colleagues than me!

[EDIT: I shouldn't impugn my coworkers: I had trouble resisting the siren call of side effecting functions!]

So don't do that!

Python list/dict comprehensions are awesome. The only other language I've seen that has comparably awesome comprehensions is Haskell which (opinion ahead) offers easier access to more power at the cost of some readability upfront.

Sadly I've seen functional operations like map/filter/reduce etc discouraged in (commercial) projects I've worked on, though I guess there's a bit of an argument there for debugability (make it a word?) since hunting down side effects may be more onerous?

I think the logic was that comprehensions should replace map and filter for most cases.

    [fn(a) for a in seq]

for example or

    [a for a in seq if pred(a)]

and so map and filter are 'old'-style and deprecated. But I have to agree, I think that's not a clear benefit. For example

    map(fn, seq)

and

    filter(pred, seq)

are clearer. Once you get more complex, however, the comprehension syntax does shine.

Comprehensions don't help at all with reduce, of course, which has been tidied out of __builtins__ now.

I think it's a matter of taste or what you're used to. I find the first two examples clearer than the map/filter ones.

Are map and filter more concise? Yeah but only in this case. If you would need to do some processing, they would get messy fast because of a lambda.

In my experience, the more complicated processing ends up being composed up of smaller units (that are testable), and then map/filter remain the most clear way of representing the iteration. Using lambdas for non-trivial process is just as bad as throwing that same complicated processing into a list comp.

I work with a lot of people who have never drunk the functional koolaid, professionally. They have to do an extra mental step with map or filter - "oh map, that's where the first argument is applied to each item in the second - I think - perhaps the second argument is applied to the first, right, Stack Exchange". More explicit is clearer, then. 'Clear' like beauty, is subjective. But productivity isn't. And whatever my personal feelings, the productivity of the team wins.

On the second point. I agree. Python's lambdas are a code-smell, for me. Which I don't say lightly, I'm a Scheme/Lisp guy deep down. But they fundamentally work against the aesthetics of the Python language, imho. A local function declaration is much less fragile.

But I would say that as code complexity increases, in my experience, comprehensions don't last long either. They've a narrow range of applicability before you get a big block of spaghetti code. Then it is much better to define a local function and, essentially, 'map' it (even if that 'map' is done as comprehension). Of course, YMMV.

100% agree that lambdas fight python's aesthetics. On the expression complexity front I can definitely say it's easy to slowly increase their complexity until they're no longer readable. As a side note, comprehensions in Python (2 at least, not sure about 3) leak scope a bit which can lead to some really funky bugs if you're in a loop and absentmindedly use a loop variable in a comprehension.

  >> x = 5
  >> [x for x in range(10)]
  >> x # => 9

Yes, that was a pain. Particularly because it was only true of list comprehensions, not dict comprehensions or generators:

    Python 2.7.10 ...
    >>> a = 3
    >>> [a for a in range(10)]
    [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
    >>> a
    9
    >>> a = 3
    >>> list(a for a in range(10))
    [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
    >>> a
    3

Guido called it Python's 'dirty little secret'. It is fixed in Py3 though.

Some lisps have similar comprehensions. Common Lisp has `iterate` (listcomp) and arguably `loop` (listcomp). Clojure has `for` (genexp) and `(doall (for ...))` (listcomp). A very common clojure library (Plumbing) implements dictcomps via for-map, although you could do python 2.6-style dictcomps yourself via

    (defmacro dict-comp 
      [bindings & body] 
      `(into {} (for ~bindings (do ~@body))))

It's really not good. Python is an imperative language and if you dig a little, you can find why [0]. The reduce function has been moved to functools module [1]. If you try to write in purely functional style in Python, you will have a bad time.

[0]: http://python-history.blogspot.hu/2009/04/origins-of-pythons...

[1]: https://docs.python.org/3.2/library/functools.html#functools...

This is silly. You're pointing out some historical reasons why some functional features have been removed, nerfed, or moved out of __builtins__, but none of this is a discussion of problems you've come across with actually writing Python in a functional style.

I've written tens of thousands of lines of code in a functional style in Python, and while it's different from Scheme and takes a little getting used to, Python is a very capable programming language for functional programming.

It does lack tail call optimization, but I think you'd find that even in Scheme, most times where you would write a recursive function, you should be writing something with one of the higher-order functions. 99.9% of recursive functions contain a buggy hand-rolled implementation of map, reduce, or filter. Scheme is a more academic language, so it makes sense for them to include this feature, which gives language users the same power as the language creators. But Python is intended for professional development and as such, it makes sense to force users into a better style.

Yes, Erlang is such a toy language[1], it's never used for "professional development" at all[2].

[1] http://learnyousomeerlang.com/recursion [2] http://www.fastcompany.com/3026758/inside-erlang-the-rare-pr...

> Yes, Erlang is such a toy language[1], it's never used for "professional development" at all[2].

From your sarcastic tone it seems like you take issue with something you think I've said, but since your post is unrelated to anything I actually said, I can only guess that you've inferred some meaning that isn't there.

The point is, the idea that it's somehow "unprofessional" to write a recursive function is an absurd assertion. Many, many functional languages beyond just "academic" ones use recursion as a standard idiom, and using that to dismiss Scheme is itself an inflammatory statement especially on a site that is literally running on top of a Scheme derivative.

Sometimes, it's just the clearer way to get the job done. I realize it is these days considered "un-Pythonic" to consider there to be more than one way to do something, but that is in fact one of the reasons why I no longer use Python. The attitude of its creator and its community have shifted against creative solutions and I have no use for that attitude in a technology field.

I tried to gently inform you that you had misinterpreted what I said. That was your opportunity to pause and think, and realize that I did not say what you thought I said, but you missed it.

> The point is, the idea that it's somehow "unprofessional" to write a recursive function is an absurd assertion.

...which is why I didn't say that. What I said was, "[E]ven in Scheme, most times where you would write a recursive function, you should be writing something with one of the higher-order functions. 99.9% of recursive functions contain a buggy hand-rolled implementation of map, reduce, or filter."

There are certainly cases where writing a recursive function is cleaner, and I would never say it's "unprofessional" to write a recursive function. However, I've debugged enough Scheme to know that if you're writing a recursive function you're probably reimplementing one of the core higher-order functions, and in doing so you're increasing your chance of bugs. You're also making your code less composable. If you're not experience enough with Scheme to have recognized this fact yet, you're probably not experienced enough to recognize exceptions to the rule.

> Many, many functional languages beyond just "academic" ones use recursion as a standard idiom, and using that to dismiss Scheme is itself an inflammatory statement especially on a site that is literally running on top of a Scheme derivative.

Scheme is my favored language for personal projects. I write more Python because there's more work available for Python, but when it has been my choice I've chosen Scheme. The reason I've written so much Python in a functional style was because I learned that style from Scheme. So if you think I'm dismissing Scheme, you're hilariously wrong. My statement isn't inflammatory; you became inflamed all on your own.

> I realize it is these days considered "un-Pythonic" to consider there to be more than one way to do something, but that is in fact one of the reasons why I no longer use Python. The attitude of its creator and its community have shifted against creative solutions and I have no use for that attitude in a technology field.

Cool. I thought I was the one being inflammatory.

It seems that you saw me say, "Scheme is a more academic language" and decided to take offense to that. Well, news flash, Scheme was designed at the MIT AI Lab. It's an academic language. That doesn't mean it's a "toy language" or that it's only an academic language; I didn't say either of those things. It's also ridiculous that you needed to bring Erlang into the conversation to defend Scheme: Scheme has plenty of history of professional use at both HP and Sun and is perfectly well respected in its own right.

Please try to learn something from this experience.

Why does it matter if reduce is no longer a keyword?

Actually it wasn't a keyword, it was just something that was "imported" by default.

reduce was not a keyword, it was a builtin function. So:

    reduce = 0

is valid. But really it doesn't matter.

Are you sure about lambdas..? They are kind of crippled in comparison to almost any other language which has them.

Reply on news.ycombinator.com