Hacker News Reader

Comment by davidjfelix | Hacker News Reader

IMHO prefix commas is fixing one mistake with another mistake. Enforce trailing commas.

I absolutely love prefixing commas. Super readable and good for diffs.

Sure trailing commas also solve the same problems but you need language level support for them. Plus they add one additional character compared to the prefix solution.

I wish I could use prefix comma formatting in other languages but not chance to convince coworkers to use that not to mention getting code formatting support.

Why would you think they are a "mistake"? What is your hang-up?

The only solution to this war is no commas!

To roughly translate the first example in the article (I shortened the value of `d`):

  svg [ 
    x=[0px]
    y=[0px]
    viewBox=[0 0 21 21]

    path [
      fill=[#FFF]
      stroke=[#146AFF]
      strokeWidth=[2]
      d=[M11.1,1.4l2.4,...]
    ]
  ]

You can remove all whitespace and it's still unambiguous.

It's a perfectly viable lightweight syntax for SVG.

Another example would be Lisp. And yeah generally, fair point.

In Elm though it would cause some ambiguity because of the function call syntax not requiring parentheses. For example:

    [foo bar baz]

Is this a List with three elements or is it one element whose value is a function call on foo with the arguments bar and baz? Who knows. Sure you could have implicit commas on line break but that would be weird.

Yes, in Elm or Lisp this would be problematic.

In this syntax though it isn't. Anything between brackets like this:

  [foo bar baz]

that does not contain any nested brackets is a simple tree which contains "foo bar baz" as its sole value (called a suffix).

In this syntax whitespace is not a separator.

So you can store a single piece of text in such a simple tree.

Now if you want to have 3 pieces of text you do this:

  [foo][bar][baz]

3 simple trees together -- they are subtrees of a top-level tree.

If you want to have 3 of these you do this:

  [[foo][bar][baz]]
  [[foo][bar][baz]]
  [[foo][bar][baz]]

The linebreaks don't change the tree structure, they are inserted purely for readability.

There are no implicit commas or anything.

Each subtree is also potentially a name-value pair. If you insert names before the opening brackets:

  [x [foo] y [bar] z [baz]]

You have 3 pairs.

The names themselves are also pieces of text (whitespace is not a separator), so you can have names like:

  [my x [foo] your y [bar] their z [baz]]

Still 3 pairs.

Now to turn trees like this into something useful, for example XML/SVG, you'd do a little bit of processing on them, e.g. you could transform the names like so:

  my x [foo] your y [bar] their z [baz]

into:

  <myX>foo</myX><yourY>bar</yourY><theirZ>bar</theirZ>

in this manner you could transform:

  svg [x=[0px] y=[0px] view box=[0 0 21 21] path [
    fill=[#FFF] stroke=[#146AFF] stroke width=[2] d=[M11.1,1.4l2.4,...]
  ]]

into:

  <svg x="0px" y="0px" viewBox="0 0 21 21"><path
    fill="#FFF" stroke="#146AFF" strokeWidth="2" d="M11.1,1.4l2.4,..."
  ></path></svg>

Notice that here I used a convention where names which end with "=" become XML attributes, whereas names which don't become children.

I have used the same convention here (except I don't bother with transforming names with spaces into camelCase): https://github.com/jevko/specifications/blob/master/easyjevk... to generate this HTML file: https://htmlpreview.github.io/?https://github.com/jevko/spec...

Now I intend to write specifications that codify conventions like this into different formats based on this fundamental syntax of square brackets.

It can be useful for all kinds of things. Its advantage is extreme simplicity and flexibility.

BTW, for clarity I have to say that the format that I used here: https://news.ycombinator.com/item?id=32995047 does a bit more transformations -- it actually sometimes treats whitespace as a separator (e.g. in `svg width[391]` space is a separator). That allows for extreme conciseness, but is not necessary and introduces complexity.

They're a mistake because correcting the language level issue with trailing commas to fix diffs was the correct solution. Instead everyone had to convince themselves that they prefer leading commas, which is not how commas are used, over whitespace. Have you ever considered maybe your coworkers are right?

> which is not how commas are used

How is that an argument? "I am right because I am right?"

Look around you and have some awareness. https://en.wikipedia.org/wiki/Comma there ya go. Do you see a whole lot of uses with it prefixing things? This is the common way that commas are used in grammar and also in programming languages. I'm not sure why everything has to be an argument, but that's it. It's ugly and I don't like it - find solace in letting me lament without having to reach consensus with your uncommon preferences.

Precisely, or even find a way not to require commas.

I don't think we've seen a language that is explicitly designed to be as ergonomic / accommodating as possible to developers.

Something that is the good protocol truism equivalent of "be liberal in accepted input and strict produced output".

Granted that flies in the face of another programming language design best practice of having 1 way to do a particular thing.

> Precisely, or even find a way not to require commas.

Here it is: https://news.ycombinator.com/item?id=33457176

Once I used a really REALLY good formatter like prettier for JS or elm-format for elm I got it. I realized that I don't care about the format, I care to see the jump from unformatted to formatted that means what I wrote is valid syntax. The leading commas (a convention but not actually necessary for elm) make diffs easy and make lists a bit easier to read. The syntax is almost the same in Haskell, but Ormolu puts the commas at the end and somehow it's harder to pick apart complicated list entries. Can't really explain why.

Do you have a good reason for enforcing trailing commas vs leading?

Leading is necessary to minimize diffs when you don't support trailing commas:

    Prelude> [1,]

    <interactive>:4:4: error: parse error on input ‘]’
    Prelude> [1]
    [1]

This means if you go multiline in the "usual" style:

any time you need to append a new item, you need to modify the existing last line as well:

Meanwhile:

If you support trailing commas, you can just use one on the last line for the same effect:

Incidentally, some crazy people put the closing symbol on the last line:

    [ 1
    , 2
    , 3 ]

That completely defeats the point:

     [ 1
     , 2
    -, 3 ]
    +, 3
    +, 4 ]

Do many/most feel this way? I always felt this argument was the diff tail wagging the code readability dog, so to speak

It does feel like a misplaced priority.

It's very important to me. I don't feel readability is impacted by having commas before or after. As long as it's consistent.

I did prefix my opinion with "IMHO". But no. Leading commas are, in my opinion, less common in life and a distracting, surprising style for the sake of fixing a syntactic language mistake. "Enforce" might have been an overly demanding word. Enforcing when there's a newline is probably more apt or simply allowing it to prevent giving leading comma folks ammunition.

something like `[hello, world]` should probably be allowed since `[hello, world,]` feels extraneous.

Reply on news.ycombinator.com