On F# code readability

http://www.clear-lines.com/blog/post/On-F-code-readability.aspx

by gebe • 13 years ago

38 70 13 years ago

crntaylor• 13 years ago

I disagree with his example of 'unintelligible' code. He suggests replacing

    match list with
      | x::xs -> ...

with

    match list with
      | head::tail -> ...

to make it clear what the names are referring to. But in F#, something to the left of :: can never be anything other than the head of a list, and something to the right of :: can never be anything other than the tail of a list! Calling them 'head' and 'tail' doesn't add any extra information. It just adds more characters.

The names x and xs are entirely appropriate when you don't know anything about the values you're working with (not even their types). All you know is that x has a type 'a, and xs has the type 'a list. Giving them abstract names reflects the fact that you are writing an abstract function. If you are writing a more concrete function, use more concrete names (e.g. trade or blogPost or playerAction... but not head and tail!)

Also, I don't know any F#, but is he correct when he says that the order of declarations in a file matters, and that the order of files in a project matters? That sounds crazy to me. What is the benefit? Both Haskell and Ocaml get along fine without relying on declaration order to do type inference.

Other than that - good article, I agree!

sdevlin2• 13 years ago

> Also, I don't know any F#, but is he correct when he says that the order of declarations in a file matters, and that the order of files in a project matters?

He is correct. (At least, this was true a couple years ago. I haven't used F# since then, but I doubt this has changed.)

I don't have enough F# experience to speak to any potential benefits of this restriction.

thomasz4• 13 years ago

It's a pain in the ass, and the only benefit I can think of is that it made the compiler easier to implement.

watt2• 13 years ago

Oh but it does.... I don't know first thing about F#, so "x::xs" looks alien and very unapproachable (and also will make the rest of method body confusing).

In my mind it's that code readability is at last being discussed, signals that F# might be nearing mainstream.

sdevlin2• 13 years ago

Surely questions of readability should be decided with an eye towards experienced users of the language and not those who "don't know the first thing about F#".

I agree with the GP that x::xs is more appropriate.

abraininavat• 13 years ago

Is readability to people who don't know the first thing about F# at all relevant? If that's the most important metric then we should write out all our programs in plain English.

Readability to F# programmers is what matters, along with a host of other concerns, such as terseness.

LeonidasXIV• 13 years ago

> Oh but it does.... I don't know first thing about F#, so "x::xs" looks alien and very unapproachable (and also will make the rest of method body confusing).

I use OCaml and I think of it as idiomatic naming for pattern matching on a list, in a similar way to how iteration variables are often named `i`, although `index` might be more superficially clear, but in reality it is not really since one "knows" `i` is the interation variable, just like `x::xs` means `x` is an element and `xs` is a list of x'es, that is elementy of the same type as x. Pretty clear in my opinion.

chc2• 13 years ago

I do like the principle of thinking about code readability. But unless you're writing an introductory tutorial, your code should absolutely not be optimized for reading by people who know nothing about the language. You should try to make it readable for the people who will actually be working with the code.

Optimizing code for people who don't know anything about the language basically means sticking as close to the C or Python way of doing things as possible. If your language is significantly different from C, this is obviously not going to be a global maximum of readability.

papsosouid8• 13 years ago

>so "x::xs" looks alien and very unapproachable

And how is head::tail any better? That's like insisting for loops use "counter" instead of "i".

kvb3• 13 years ago

I agree that using head::tail is usually not more readable, though I think that sometimes it is better to be a bit more descriptive than just x::xs (e.g. key::keys). It really depends on the context as well as the experience level of the people who will be reading the code.

It is true that the order of files matters. Presumably this is for the same reason that order matters within a file.

mich414• 13 years ago

> Also, I don't know any F#, but is he correct when he says that the order of declarations in a file matters, and that the order of files in a project matters? That sounds crazy to me. What is the benefit? Both Haskell and Ocaml get along fine without relying on declaration order to do type inference.

IIRC in OCaml you have to compile modules in topological order wrt dependencies.

sitharus3• 13 years ago

Yes, order of files in a project matters. If Things.fs contains type Foo and Main.fs uses it, then the project must look like

    MyProj
    |- Things.fs
    |- Main.fs

Visual Studio has a few functions to manage this order.

The benefit is faster compilation, because F# requires recursive types to be declared together, such as

    type Foo = A of Bar | B
    and Bar = {Baz: Foo}

this doesn't really come up that much.

lysium2• 13 years ago

> this doesn't really come up that much.

Unless, of course, if you don't know or expect that 'rule' and you are wondering, why your type is not visible in that other module for over an hour...

gtani• 13 years ago

If you were writing for haskell/scala/ML'ers, x::xs, for C#ers, head::tail

I haven't been tracking closely but no Forward declarations was one of common complaints, along with organizing projects in subdirectoris (I think they have fixes for that now) and VS support in general

http://stackoverflow.com/questions/1378575/f-forward-type-de...

__________

Also Xamarin shd be releasing... soon

http://news.ycombinator.com/item?id=5251413

Associat0r• 13 years ago

> Also, I don't know any F#, but is he correct when he says that the order of declarations in a file matters, and that the order of files in a project matters?

> That sounds crazy to me. What is the benefit? Both Haskell and Ocaml get along fine without relying on declaration order to do type inference.

Here is great explanation on the benefits of this. http://cs.hubfs.net/topic/None/59219#comment-70220

mokus• 13 years ago

I've long felt that very-short names like "x" and "i" should be treated like pronouns in English. They should only be used when the scope is fairly limited and there shouldn't be too many in use at once. There are probably other sensible guidelines I've missed here.

When used well they often make code more readable than "descriptive" names, just like most English text is more readable with pronouns than without. In any case, categorically claiming they are bad makes about as much sense to me as banning pronouns from English.

jamesaguilar• 13 years ago

That's a really great rule of thumb. I think I might use that. x::xs might not be so nice if the body of the match has twenty lines in it, especially if y::ys also exists. That'd be the pseudo-equivalent of saying, "Joe and Sam were getting food and he wanted a burger but he wanted steak so he finally argued that they should go to Bob's Grill which had both and he agreed." (Not really, because the sentence is actually totally ambiguous whereas with variables it just involves an extra lookup for the reader.) But if the body is one or two lines, there's not much purpose in the extra characters.

flaie2• 13 years ago

You're right about the fact that on the left on `::` can never be anything other than the head of a list.

But I'd like to add that throughout almost all OCaml code I've written and used to read, that type of matching was written:

    match list with 
        | h::t ->

Wich reads easily `h` for `head` and `t` for `tail`. I'm surprised it hasn't been reproduced in the F# world.

`h::t` seems even better to me than `head`/`tail` or `x`/`xs` which for the second one, means nothing really to me.

mich414• 13 years ago

Maybe your code isn't famous enough. I've seen lots of x::xs's everywhere, but never a single h::t.

Personally, I always use x::xs in one-liners and something meaningful (no, neither head nor h is meaningful) in longer matches.

waxjar• 13 years ago

I don't know any F# either, but I know that in Ruby the order of files does matter. When you define a class that inherits from a class that lives in another file, you'll have to require that one first.

pash2• 13 years ago

> ... in F#, something to the left of :: can never be anything other than the head of a list, and something to the right of :: can never be anything other than the tail of a list!

Yes, but the names you assign in a pattern get used outside the pattern. Presumably there will be an x/head or an xs/tail on the right side of the arrow in your example, and it's there that meaningful names help.

That said, I'm quite fond of the x:xs naming convention. It's particularly useful when you're doing something with two or more lists in the same scope, and you pattern-match to produce multiple heads and tails. For instance (Haskell syntax, as I don't know F#):

  zip :: [a] -> [b] -> [(a, b)]
  zip (x:xs) (y:ys) = x : y : zip xs ys
  zip _ _ = []

This also illustrates when you might well disregard the trope of using "meaningful" names. The code above is polymorphic and will work for any types a and b, and for any values of those types, no matter what they represent. Functional programming really encourages this sort of generality. Yes, it's very much the case that, as the article's author wrote, "functional code tends to focus on applying generic transformations to 'things', and not that much on what the 'thing' might be."

But, if it helps to clarify what you're doing at the domain level, you can (and should) assign meaning through types and by renaming and adding restrictive type signatures to polymorphic code:

  data Child = ...
  data Chore = ...
  type Assignment = (Child, Chore)

  assign :: [Child] -> [Chore] -> [Assignment]
  assign = zip

This not only helps readers understand what's going on, it also enlists the type-checker in helping to ensure you've, say, assigned chores only to your kids and not to your pets.

pash2• 13 years ago

Whoops, I changed my example midway through and ended up with something screwy. The first case of the definition of zip above should of course read:

  zip (x:xs) (y:ys) = (x, y) : zip xs ys

MichaelGG3• 13 years ago

Preferring head::tail over x::xs sorta indicates they have read little functional code. Using x and xs (plural) is pretty common and straightforward. head::tail doesn't help at all.

Edit: Also, even in C#, the whole "one class per file" is often unwise. So many times I come across projects that have a bunch of bloody files to define interfaces with a handful of members. And then a bunch of implementations that have 2-line implementations. It's sorta ridiculous. I think it's a holdover from Java's idiotic "tie the filesystem directly into the compilation output" approach.

NateDad• 13 years ago

Actually, if you're browsing the code in Visual Studio, it's a hell of a lot easier to jump around the code if you have one class per file. All the main navigation UI is geared to the file level - the tree in solution explorer is all files. The tabs in the editor are lists of files. ctrl-tab brings up a list of files. You can bring up a new tab group and see two files side by side.

So, yes, one thing per file is actually a pretty good idea. Sometimes it's fine to put multiple things into a single file, if there's a lot of small implementations... but for anything even mediumly big, it's good to give it its own file. It helps a lot with navigation.

And yes, I know you can do "go to definition" etc, but if you want to go back and forth between code, separate files is a lot easier. It also makes source control a lot easier, because you can see a file changed and know immediately what it will and will not affect.

Which is not to say that I don't think java got it wrong, it did, it's just that the problem is not with one class per file, but rather that the file heirarchy matters.

MichaelGG3• 13 years ago

In C#, VS has a class view that lets you jump around by class name. You can split the window and compare within the same file just fine. And if anything, you're complaining about VS's lack of capabilities.

I'm not arguing that you never want a separate file, just that having a forced "one class per file" rule just sprays files all over and means you have to jump around more. Java's hierarchy+class=file is just silly; there's nothing stopping people from enforcing that themselves, if they so choose.

sitharus3• 13 years ago

That depends. I tend to leave small support classes in the same file, but anything public goes to file per type.

Why? Because I have to work with people who don't have resharper installed, and without resharper it's near impossible to find those types without doing a full project search.

MichaelGG3• 13 years ago

Is a search that hard? Go To Definition? Class View?

tikhonj• 13 years ago

The x::xs example is bad because it's a widely used idiom, similar to using i as a loop counter.

Also, I think that short names are completely justified if the variable is used on a single line. In fact, in those cases, the shorter name is often clearer because it doesn't obscure the structure of the line.

So, my rule of thumb: top-level bindings, function arguments for complicated functions and local bindings that are either used in more than one place or away from where they are defined all get descriptive names. (Although I do try to keep even those brief--more than two words is probably too much, and even two is suspicious.) Function arguments to small functions and local variables that are used immediately (usually from pattern matches) get short names--one or two characters, usually.

Of course, this really is just a rule of thumb. If deviating from it seems to improve my code, I deviate. But it does give a good idea of the shape of my code.

gnaritas4• 13 years ago

> The x::xs example is bad because it's a widely used idiom, similar to using i as a loop counter.

And IMHO bad for the same reason. I'd rather see "index" than "i" and I'd rather see "head::tail" than "x::xs", abbreviations for variable names are terrible; the English language is not running out of words and readability is more important than brevity. Of course, I'm a Smalltalker and we like to actually have readable code.

mich414• 13 years ago

Everybody who has ever seen any C code has

  #define i index

programmed deeply in his brain's CPP.

Using i for loop counter never causes confusion and using index makes the for statement 12 characters longer for no reason.

sseveran• 13 years ago

x::xs or x:xs in haskell is a very common naming pattern for pattern matching on lists. head and tail are functions that operate over lists so using them as a variable would be considered name shadowing. i is good to use as a loop counter in C for the same reason, its widely recognized and people will know what it means.

ajanuary• 13 years ago

w.r.t. x::xs vs. head::tail

One of the nice conceptual things about pattern matching is the pattern on the left hand side describes which bits of the structure you're interested in, while the stuff on the right hand side describes how to manipulate that data.

When you're looking at the rhs it shouldn't care about where data came from, it should just care about what the data is. I've got some arbitrary thing `x` and a list of arbitrary things `xs`. I've got some `person` and a list of `people`.

Using `head` and `tail` makes the rhs concerned with the structure of the incoming data. This becomes more of a problem if you have more complicated data structures on the lhs.

One of the key elements of readable code is how easy it is to map your conceptual model onto source code onto executable code. Using head::tail wouldn't help new people grok the separation of structure on the lhs and manipulation on the rhs.

All that said, it's such a minor point as to probably not really have any impact in this case. Just a bit of silly purist hand-waving.

ufo• 13 years ago

I really like this analogy. I think a good example of it is the following code for red-back trees:

    balance :: Color -> RBTree a -> a -> RBTree a -> RBTree a
    balance B (Fork R (Fork R a x b) y c) z d = Fork R (Fork B a x b) y (Fork B c z d)
    balance B (Fork R a x (Fork R b y c)) z d = Fork R (Fork B a x b) y (Fork B c z d)
    balance B a x (Fork R b y (Fork R c z d)) = Fork R (Fork B a x b) y (Fork B c z d)
    balance B a x (Fork R (Fork R b y c) z d) = Fork R (Fork B a x b) y (Fork B c z d)
    balance k a x b = Fork k a x b

The lhs does the dirty work of identifying the relevant bits in the tree and in the rhs things get rebalanced in a similar form.

(Another thing I find amusing in this example is that its one of the few examples of Haskell code where you can't get around the "copypasting" by using `let` or `where`)

thomasz4• 13 years ago

Speaking about F# and readability:

Reading F# (and Ocaml) without an editor sucks:

- Nobody bothers to include type or visibility annotations in function prototypes. Signature files are not local and thus are not included in diffs etc. If you think that doesn't matter, go to https://github.com/fsharp/fsharp/blob/master/src/fsharp/FSha... and tell me the type of generateWhileSome.

- The order of compilation matters, and is defined in the Build script:

> Essentially, I either start from the first line of the first file, and read forward, or the last line of the last file, working my way back."

That only works in the IDE. On github, the workflow goes like this: Open the build file, look for the first file. Find that file in the directory structure. Read that file and go back to the build script, rinse, repeat ad nausea.

kvb3• 13 years ago

There's no reason that this needs to be the case; see Tomas Petricek's tools for producing tooltips for F# code on the web [1][2]

[1] http://tomasp.net/blog/fswebsnippets-intro.aspx [2] http://tomasp.net/blog/fsharp-literate-programming.aspx

thomasz4• 13 years ago

Seems like the only thing left to do is getting patches into cat and diff. Seriously, adding support for inline signatures would solve the problem in a general way. Why not do this?

    val map : f:('a -> 'b) -> list:'a list -> 'b list
    let map f list = (*...*)

lysium2• 13 years ago

I'm all with succinctness. I'm wondering how much this has to do with F# per se than with imperative vs. functional programming language.

I think the same applies to Java vs. Scala / Clojure. (For example, in German, http://funktionale-programmierung.de/2013/02/26/scala-java-a... -- Google Translate: http://goo.gl/CF1q9)

rcoh• 13 years ago

503....Whoops.

Jabbles• 13 years ago

Can you really claim that it's a "one-liner" just because you can format it so that it has no newlines?

You can do the same thing with C, but I wouldn't let anyone do it in my codebase. Perhaps the F# community feels differently?

On F# code readability

Related Stories