Learning Javascript by hacking JSHint

About ten days ago, I decided I should learn Javascript. I used to have a terrible opinion about that language, mostly because all I knew about it was the very bad parts… All globals, weak typing years of irritating approach for web programing.

Go for javascript

But then, at FOSDEM 2013, I let me convinced by my Mozillian friends that it is actually not such a bad language, and it is possible to make clean and neat code while ignoring those bad parts. Then I thought that even the holy C has its share of bad stuff, so why not give it a try?

So, I decided to buy and read Crockford’s book about the “Good parts” of Javascript (pretty appealing, when we only know about the bad ones), and updated my knowledge of the bad stuff. And that let me discover more deeply hidden bad stuff.

But there’s hope, and that’s for the next real version of Javascript, being currently drafted by the Javascript committee. That’s how I discovered the specification of the let keyword that introduces real block scoping to Javascript, destructuring expressions and a few other features that I would have missed having spent a few years in the Python world.

I also went over the Javascript pages of mozilla and discover they implemented a lot of really neat little features before they were considered or being considered stable in the draft.

Getting dirty: working on the JSHint linter

While I was testing scripts and language features, I could get easily lost by all the different ways to write a code and how a simple typo can make a code break in weird ways. That’s why I installed syntastic plugin for vim that uses jshint as backend lint for Javascript.

But whenever I was using a new or non-standard syntax, the lint would consider my code as buggy and that was really unpleasant, as I was still learning and couldn’t tell whether the lint or my code is wrong. That’s how I considered getting into the source of JSHint and update its features.

I have already worked on three compilers in my life, and thought that the algorithmic principle would remain the same, so the language’s features and syntactic sugar would become more obvious to me.

JSHint is based on JSLint, written by Douglas Crockford himself (and for the anecdote, when I looked at a git blame on the parser source, only blank lines were still from his code ;-) ). Basically it implements a Pratt Parser, using the principle of top-down operator precedence, associating semantic rules with the first token on the line, instead of using a grammar rule.

On the good side, it makes the code really easy to understand, as most of the time to understand how a language feature is implemented, you shall begin with the first token, and read the code to get where you need. On the other side, it makes semantic rules harder to implement and look like a dirty hack when there’s ambiguity on the first token.

Two problematic cases were the ambiguity between destructuring expressions, array comprehension and JSON syntax, and the second one was all the let syntaxes implemented by mozilla. The solution is always to look ahead what kind of tokens will be used (which mimics the purpose of a grammar based set of rules), to decide what kind of construct is being used.

What did I learn?

Javascript is a really easy language, though not really simple, and has some functional programing features that can make it almost as elegant as a real functional language. I could cite many of the chapters from the Crockford’s book, “recursion”, “Scope”, “Closure”, “Callbacks”, “Cascade”, “Curry”, “Memoization”… I’ve met most of those features and patterns in the parser, that made it a really cool real world exercise.

Real world use of good practices

The heart of the JSHint parser is the expression() function. It will decide what can of expression it is (as formalized by the Pratt Parser algorithm) and call functions associated with the first token’s identity.

Then, every token will get an associated function by the use of handlers functions (like prefix, infix or stmt) that takes a token’s name as first argument, and the applied function as second argument. That “handler” function will bind the actual parsing function to the given token type, using the Pratt Parser formalism. That’s a cool example of the use of a mix of the language features exposed by Crockford.

New features

But I’ve also got deeply into the new syntaxes ES6 will offer, or Mozilla implemented for their javascript engine.

  • let token: block scoping… and more

I have implemented the new blockscoping feature, with the let token that declares a variable in that scope. That feature is pretty neat, as it can help use variables more closely to were they are being used.

But what is even better, imho, is the mozilla extensions to the let keyword, that defines a new let expression, so you can declare and define a bound variable local to an expression:

var y=2; let(x=1) print(x, y); print(let(x=1) x, y);

  • destructuring expressions

That’s something that should be in every high level language. It helps declare a bunch of variables out of an array or an object. It becomes really important in Javascript for the require mechanism, so you can easily select what you want to import. New Mozilla APIs are using it a lot:

const {foo, bar} = require(“foobar”);

instead of

var foobar = require(“foobar”); var foo = foobar.foo; var bar = foobar.bar;

  • array comprehensions

That’s also some syntactic sugar that is really helpful by times, but can make a code really hard to read some other times. The first implementation of such a feature is in current Mozilla’s JS implementations, though the EcmaScripct committee is deprecating it in its current version. The first implementation is harder to implement in a parser and unnatural to read (you use first what you will define after), but is very similar to the python syntax.

let foo = [x for (x of [1, 2, 3, 4, 5, 6]) if (x % 2 === 0)]

Whereas the new drafted syntax is simpler to implement and read but will be confusing for people from the python world:

let foo = [for (x of [1, 2, 3, 4, 5, 6]) if (x % 2 === 0) x]

the latter syntax is still not implemented in JSHint as of today, but it will be as soon as the esnext/moz syntax features are correctly decorralated.

  • function generators

The introduction of the yield keyword to create generators so one can iterate over the results of a function while the function is iterating is really a great new feature. The only perk, is that the ES6 committee introduced the function* token to identify generator functions, whereas Mozilla already implemented it using the standard function token. Both syntaxes will be supported in either the esnext or moz mode.

  • Mozilla specific features

And finally, I implemented a few really neat mozilla specific language features, such as mutiple catch block support, as well as conditional expressions in catch. Function expression closure, for each support (that is now deprecated towards for of)… Most of them being explained over here or on MDN.

To conclude

I have discovered a new language and the approach that goes with it. Sadly the web is full of bad answers for bad questions about javascript, but that’s because it is a language being used by a lot of people. Java has still the same problem (and that’s maybe their second common point after their name :-) ).

I’m a bit sad to discover that the language evolution is still being teared between implementors and comittees that can’t get something as standard as it would be needed for the Great Interoperability of the Web. But I hope that the comitte will make things right, and that we won’t have to wait 2018 to have JS as correct as it would be needed today.

I still got some trouble apprehending the different contexts of the variable this in function calls, and I think it is one of the worst things about Javascript. I like behaviours to be homogeneous and simple, and this is not the case. But I have hope that the => syntax that is being considered stable in most recent drafts will improve the situation, by not setting this up anymore, but doing it explicitely.