To read this README offline in a sanely formatted way, grip it, or Pandoc it into your preferred file format.
ECMAScript, formerly known as JavaScript, is still the only language supported by web browsers, so for the frontend of a web app it is a requirement. Things are slowly changing (Brython, WebAssembly), but as of early 2020, ES remains the dominant language for the frontend, as well as the implementation language under the hood of the alternatives.
Here's a quick overview for those coming from Python, to hit the ground running. Some experience with the Lisp family helps, but is not a requirement.
ECMAScript is no longer the Turing tarpit it used to be. Modern ECMAScript (ES6 and beyond) is essentially like Python, but with a lot more semicolons and curly braces, and with (more than?) half of the standard library missing. As a pythonista, you'll almost feel right at home.
If you're hardcore, and have copious free time, read the standard. Also all of MDN, and for good measure, w3schools.com.
Just like Python, ES is a duck-typed, dynamic language. It's an interpreted language in the same modern sense as Python: ES compiles to bytecode, which then runs on the ES virtual machine. There are several implementations (engines), for example SpiderMonkey in Firefox, and V8 in Chrome and in Node.js.
Truthiness semantics are similar to Python's, with similar short-circuiting logic. The logic operators are spelled as in C, &&
(and), ||
(or), !
(not). The constants true
and false
are spelled in lowercase. The constant null
is ES's equivalent of Python's None
. There is also the constant undefined
, which ES itself uses to indicate e.g. that a variable has been declared, but not initialized. As an argument to a function call, undefined
means use the default value for the parameter.
Full list of ES operators on MDN.
Parentheses around the check in if
are mandatory. The spelling for chaining if
s is else if
, and the bare else
is available as usual. Like in C, curly braces around the body (a.k.a. suite) are optional if there is only one statement in the body. Unlike in Python, for multiple statements the braces are required; in ES indentation is for humans only.
Basic containers are Array
(like Python's list
), Set
(like Python's set
), and... the object. There's no separate dict
type, Object
itself acts like one. Attributes (ES calls them properties) can be accessed with obj.x
as well as obj['x']
; the latter is handy if the name is stored in a variable. There's no generic __getattr__
override; unlike Python, ES doesn't go to superhuman lengths to expose its moving parts (Evelyn Woods). You can define Python-style properties, though, triggering computation on access to a specific attribute - this feature is known as accessor properties, a.k.a. getters and setters.
Strings are quoted using "..."
or '...'
. Just like in Python, they do the same thing, and avoid the need to escape the other type of quote inside the string. Unlike in Python, there is no """..."""
.
The equivalent of Python's f-strings are template literals. They are quoted using backticks `...`
, and ${...}
inside a template literal interpolates an expression. For example, const s = `Hello, ${name}!`;
interpolates the value of the variable name
into the string. (The exclamation mark is just part of the string; only the ${...}
is treated specially.)
ES6 introduced a pythonic module system. Beware though, if you don't use modules, the global top-level scope is shared, unlike in Python. If you want to try modules, read up on import
and export
, and in the HTML <script>
tag, use type="module"
instead of type="text/javascript"
. Some sources recommend, as a convention, the use of the file extension .mjs
instead of the traditional .js
for module-based code.
The classical, pre-ES6 solution that you'll still often see in the wild, is the module pattern: to emulate modules using an IIFE (immediately invoked function expression). The IIFE approach takes advantage of the closure property to hide internal details from the shared top-level scope. Such an old-style "module" typically contains code at the end of the function to explicitly inject an object into the top-level scope. This object is, by convention, named after the "module", and contains its exports.
Unlike Python, ES has a proper lambda (for some reason, known as function
; no nod to history?), that can contain several statements and use all features of the language. Even in Python, the single-expression limitation of lambda
is really no limitation, see unpythonic
for two different ways to place multiple expressions there. Instead, what bites us in Python is that a Python lambda
only accepts expressions, and many language features of Python are only available as statements. ES does not go the extra mile to shoot itself in the foot like Python does. ES supports both function statements (for defining a named function) and function expressions (for use as a λ). Both use the same keyword, function
, and whatever you can put in the body of a function statement, you can put in the body of a function expression.
The practical advantage is that a proper λ is often useful for achieving a top-down presentation order, especially when callbacks are involved. You can first say what your goal is, and then right there in the argument list of the higher-order function you're calling, write the callback λ that handles the details - instead of having to define a named function up front, before the overly rigid shape of the language even allows you to state what you're writing about. If λ being anonymous is a problem for human-readable tracebacks, then the language should allow naming function expressions, not cripple the λ construct, dammit.
ES gets this right: you can name a function expression. See Named function expressions demystified. As a bonus, the name will be bound inside the function (but not outside it!), which is useful for recursion. Also, if you write a declaration of the form const f = function (...) {...};
, ES infers (like Racket) that the intent is to define a function named f
, so it'll set the name for tracebacks, even though the function expression itself is anonymous.
On the other hand, a very nice feature of Python, which ES simply does not have, is passing arguments by name. You can approximately emulate that if you try hard enough, but that's not very idiomatic. So be very, very careful about the ordering of arguments in function calls. Dynamic typechecking and no named arguments is a silly and dangerous combination, but that's ES for you.
DANGER: In ES, providing the wrong number of arguments to a function call fails silently. No helpful TypeError
, unlike in Python! If too few arguments were given, any leftover parameters are simply set to undefined
. If too many arguments are given, the rest are just dropped on the floor. So be careful - the language won't help you. (This is probably how haskellers feel about Python.)
As a pythonista, you'll definitely want strict mode ('use strict';
), which e.g. requires declaring variables (to explicitly scope them and to ensure there are no typos). It's the default for modules, but not for classical scripts.
Just like Python, ES does not have tail call optimization, so you have to be careful about the maximum recursion depth. If arbitrarily deep tail-recursion is important to what you're doing, you could always define your own trampoline, like Clojure does. See unpythonic.tco
for how to do that in Python. The same technique applies trivially to ES.
The terminology in ES is a bit different from Python. For example, what Python calls attributes, ES calls properties. What Python calls properties, ES calls accessor properties. What Python calls unpacking, ES calls destructuring. (Dictionary unpacking is object destructuring.) What Python calls lambda
(in the practical sense of a short, anonymous snippet, usually to be fed into a higher-order function such as map
, filter
or reduce
), ES calls arrow functions.
Language versions of modern ES are numbered starting from ES6, a.k.a. ES2015, based on the year it was released. As the preface of the standard says, ES6 was a huge 15-year standardization effort. Since then, there has been one release of the language standard per year, each time adding 1 to the version number. So e.g. ES7 is ES2016, and ES10 is ES2019. These annual releases often contain a small number of updates, compared to the major overhaul that was ES6.
Lexical, just like in Python.
A nice feature in ES, that Python doesn't have, is fine-grained control of the scope and constness of variables (names, really, in the same sense as in Python):
var
: scoped to the nearest enclosing function scope, like Python's locals,let
: scoped to the nearest enclosing curly braces, andconst
: likelet
, but with the language-level guarantee that aconst
variable cannot be re-assigned to point to something else.
Of these, let
and const
(declared at the appropriate level) are the most useful, as they can be scoped to exactly where you want, avoiding pollution of a function's top-level scope by temporaries local to e.g. an if
branch.
Unfortunately, due to historical reasons, the design is slightly botched. The var
kind may shoot you in the foot:
- At the global top level (when not inside any function),
var
will happily pollute the shared global scope. (Good luck not trampling over anyone else's definitions by accident.) - A
var
variable always gets initialized toundefined
. It will not throw aReferenceError
if you try to access it in the declared-but-not-yet-defined state. This may smite you much later.
In contrast, let
and const
are safe:
- At the global top level,
let
andconst
do not pollute the shared global scope. They remain local to the file in which they are declared. - Accessing a
let
orconst
value, inside the scope where it's declared, but before the assignment that sets its value has been executed, will throw a helpfulReferenceError
(just like Python'sUnboundLocalError
in the analogous situation).
Example:
function f () {
console.log(x); // ReferenceError!
let x = 42;
}
The name x
will be in scope immediately when f
is entered, but it is only usable after the x = 42
has been executed - as expected, if you're used to Python. The let x
part instructs the ES compiler to allocate the variable x
in the lexical scope where the let
appears, whereas the x = 42
takes place at runtime, when execution hits that line. The analogous situation in Python is:
def f():
print(x) # UnboundLocalError!
x = 42
This is completely analogous, because in the ES example above, the let
appears in the top-level scope of the function f
. Keep in mind that in ES, we could also let
inside e.g. an if
branch, which has no direct analogue in Python, because in Python the smallest unit of scope is the function. (In Python, fine-grained scoping is not available; with the technical detail that a list
/set
/dict
comprehension is shorthand for an immediately called function that does a specific task.)
When inside a function, sometimes a strategically placed var
can be useful to "leak", on purpose, a variable to the nearest enclosing function scope. If you want to use the technique of reading the final value of a loop counter after a loop finishes, you'll need a var
, if you want to declare the counter variable in the for
loop header. Just be careful not to read the value of the counter before it is assigned. Alternatively, you can let
the name outside the loop beforehand, at the function scope, and then just set its value (not declare) it in the for
loop header. This is one more line of code, but free of pitfalls.
A for...of
loop is like Python's for...in
.
CAUTION: ES also has the syntax for...in
, but that's a different thing.
Like in C, curly braces around the loop body are optional if there is only one statement in the body. Unlike in Python, for multiple statements the braces are required; in ES indentation is for humans only.
This example:
inp = [1, 2, 3, 4, 5];
out = [];
for (let x of inp) {
out.push(function () {return 2*x;});
}
for (let f of out) {
console.log(f());
}
will print 2, 4, 6, 8, 10, as expected, because using let
in a for
loop will create a fresh x
for each iteration. (It can even be const x of inp
if you don't need to re-assign the name x
inside the loop body. It's a different x
each iteration, so nothing is getting re-assigned by the loop itself.) If you use var x of inp
, then it behaves like in Python. Then there is only one x
, which lives in the function scope, and you'll get 10, 10, 10, 10, 10. Similarly if you let x
at the function scope, and then for (x of inp)
.
For objects acting as dictionaries, the equivalent of d.keys()
is Object.keys(d)
, for d.values()
it's Object.values(d)
, and for d.items()
, Object.entries(d)
. You could use for...in
, instead of for...of
with Object.keys
, but there is a difference: for...in
also walks the prototype chain, whereas Object.keys
skips that and only gives the keys for the object's own properties.
ES has generators, and for...of
loops iterate over them. To define a generator, use function*
(note the star; this seems a nod to naming conventions in the Lisp family) instead of function
; you can then use yield
in the body.
There is no list
/set
/dict
comprehension syntax. For lists, see Array.prototype.map
(in plain English, that's the method map
of Array
objects) and Array.prototype.filter
. So for example to create a list of even squares, this Python:
evsq = [x**2 for x in range(10) if (x**2 % 2) == 0]
becomes this ES:
const evsq = range(10).map(x => x**2).filter(y => y % 2 === 0);
The code is presented in the same order as it executes. You can chain the map
and filter
operations in whatever order, depending on what makes the most sense for your use case. The exponentiation operator was added in ES7 (ES2016). In ES6 and below, Math.pow(base, exponent)
was used instead.
CAUTION: I'm glossing over the fact there's no range
in ES. But you can use the one we provide further below.
If you need a dictionary comprehension, one possibility is to define an analogue of map
for objects acting as dictionaries:
function dictmap (fkey, fval, olddict) {
let newdict = {};
for (const [key, val] of Object.entries(olddict))
newdict[fkey(key, val)] = fval(key, val);
return newdict;
}
const d1 = {a: 1, b: 2};
const d2 = dictmap((k, v) => k + k, // how to make new key
(k, v) => v**2, // how to make new value
d1);
}
I have chosen to include both k
and v
as parameters to both fkey
and fval
to emulate how Python scopes them in its dict
comprehension (both k
and v
are available anywhere in the body expression).
Converting the code to use the implicit this
instead of an olddict
parameter, you could even monkey-patch such a function onto Object.prototype
... but that's probably a bad idea.
For equality comparisons, you'll most often want to use ===
(three equals signs) and !==
. They are the strict equality and strict inequality operators, which skip some implicit type conversions.
For custom objects, all of ES's equality comparison operators behave like Python's is
:
foo = {a: 1};
bar = {a: 1};
console.log(foo === bar); // false
console.log(foo == bar); // false
console.log(Object.is(foo, bar)); // false
If you want to make a custom by-value equality comparison for your own class (like Python's __eq__
dunder override), just write a custom function and call that. No nice infix syntax for you.
Be aware that builtin types come in two flavors, and this language implementation detail is not papered over as meticulously as in Python. For example, in ES, string literals (primitive strings) and String
objects are considered different. To test whether a value is a string, use (typeof obj === "string") || (obj instanceof String)
. Neither test alone will do.
typeof obj
identifies the primitive type ofobj
. For anything that is not a primitive, the result is"object"
.obj instanceof Foo
tests whether the object was produced by the constructor functionFoo
, or anything that inherits fromFoo
.
(Maybe this is a good time to point out that in ES, strings internally use an UCS-2 encoding with surrogates. More about Unicode in ES.)
Much has been said on the internet about the ECMAScript object model being foreign and weird to many developers - based on prototypes, not classes - but actually that's pretty much exactly how Python works, too.
Especially, the delegation mechanism is essentially the same. Recall that in Python, if an attribute is not found directly on the object, it is next looked up on the class, which is also just an object (essentially a prototype!), and if still not found, then up the MRO. This is exactly what ES does, with the small difference that unlike Python, ES only supports single inheritance. ES looks up a property on the object itself first, and if not found, then walks up the prototype chain.
Consider that in Python, you want to have your methods defined on the class, not on the instance. Python's class
syntax conveniently automatically does this when you define a method in the class
body. Similarly, in ES, you want your methods defined on the prototype, not on the instance. Similarly, this is conveniently automatically done when you use ES's class
syntax, introduced in ES6. See [1], [2] for an overview.
The pre-ES6, classical way to define objects in JavaScript is to represent the object by a constructor function that returns the new instance. This seems a nod to the Lisp family, using the closure-object-duality. In the wild, you'll see functions that assign methods to this.prototype.xxx
, so that's what it means. But for new code, it's perhaps clearer to use the class
syntactic sugar. C++ and Java programmers beware; class
is just sugar, the ES object model remains prototype-based regardless of which syntax is used.
The naming convention is that just like in Python, a class name starts with a capital letter. Similarly, if the object is defined using the constructor function syntax, the name of the function (by convention) starts with a capital letter.
Sometimes object literals are useful. An object literal looks almost exactly like a Python dictionary:
obj = {a: 1, b: 2};
If a key is a valid identifier name, it is not necessary to quote it. An identifier name is either a valid identifier (i.e. anything that can be used as a variable name) or a reserved word of the language. See the link for more details.
This is easy to remember by recognizing that the keys are property names. To read the "a
" property of obj
, one would write obj['a']
, or equivalently, obj.a
. So the object literal syntax just mirrors this.
Unlike Python, where the feature was removed in 3.0, ES supports destructuring also in a function parameter list, both in regular function definitions and in arrow functions.
Square brackets pull elements from the beginning of an array:
function f ([a, b]) {
console.log(a);
console.log(b);
}
arr = [1, 2];
f(arr);
g = ([a, b]) => a;
console.log(g(arr)); // 1
Here both f
and g
take a single argument, which must be an array.
Curly braces pull properties by name from an object:
function f ({a, b}) {
console.log(a);
console.log(b);
}
obj = {a: 17, b: 42};
f(obj);
g = ({a, b}) => a;
console.log(g(obj)); // 17
Here both f
and g
take a single argument, which must be an object.
Just like Python's unpacking, ES's destructuring can be used to perform a pythonic swap:
let a = 1;
let b = 2;
[a, b] = [b, a];
console.log(a); // 2
console.log(b); // 1
Keep in mind that as a general rule, I/O in ES is async. This applies also to the loading of scripts, and of any assets. Node.js and libraries for it (e.g. lockfile) make some exceptions here, but generally it is better to write async code, because that eliminates the I/O wait time, allowing the program to do something else while waiting for data to arrive (often, over the network!).
In ES, Web workers provide genuine OS threads, and they can be used for true parallelism (unlike genuine OS threads in Python, where the GIL limits performance). They are usually defined in a separate .js
file, but if that feels overly bureaucratic in a small program, it's possible to embed the code into the same file, by generating an object URL (see the MDN link). Web workers are best only used for computationally heavy background tasks that must not hang the UI.
Most ES code is just fine single-threaded, with the help of async I/O. In ES, async comes in many flavors. Traditionally callbacks were used, leading to nested callback hell (with the accompanying runaway indentation) when several async operations needed to be chained. The next evolution were Promises. But nowadays you'll want the modern syntactic sugar: async function
and await
, like in Python 3.5+. Unlike in Python, there is always an event loop, so an async function can be called normally from anywhere.
In ES, async functions are actually built on Promises, so understanding those first helps. See JS for impatient programmers. When called normally, an async function will return a Promise
. So essentially async function
is just (highly convenient) syntactic sugar for defining a function that returns a Promise
.
An ES promise has then
and catch
methods, which can be used to schedule code to run upon success and upon error. (Promise
is essentially a Maybe monad. An error causes Promise
to skip any success callbacks until it finds the next error handler.)
When lexically inside an async function, just like in Python, you can await
another async function. This takes care of arranging things so that the code after the await
runs after the result is in. That code essentially implements the then
method for the Promise
. For error handling, you can surround the await
in a try
/catch
; the exception handler will then act as the catch
method of the Promise
. Just like in Python, the async
/await
syntactic sugar makes the code look like regular synchronous code, while it actually runs asynchronously.
If you find yourself in legacy callback hell, you may be interested in promisification (i.e. converting callbacks to promises). At the backend, Node.js already has util.promisify
, but using the above information you can easily code your own for the frontend.
For loading static assets, the simplest way is to use an <object>
tag in the HTML document. Then you can just grab the object from the DOM after the document has finished loading. You'll want to set window.onload
, or do
something similar.
When this is not an option (e.g. when the filename must be computed at runtime), you can use XMLHttpRequest (a.k.a. XHR) to load assets. There is usually a same-origin policy in effect, unless you enable Cross-origin resource sharing (CORS) in your HTTP server that's serving the app. In practice this means that using XHR, you can load assets from the directory your .js
file lives in, as well as any of its subdirectories.
Even for local testing, you need an HTTP server. Just opening the .html
locally in a web browser fails if your app uses XHR. For example, the contentDocument
property of an <object>
asset might never become available, when the .html
is viewed directly from a file, because then the origin is null
, and hence without CORS, XHR is not allowed. When the same files are served by an HTTP server, the origin is set, and it works fine.
For a quick HTTP server for testing, python3 -m http.server
does the trick. Run it in the directory that has your index.html
, and then browse to http://localhost:8000
.
The frontend does not support writing files. There are three solutions, depending on what you need:
-
Let the user download the data as a file, which they can then save locally.
- This is convenient for exporting data, when there is no need to load that data back later, such as saving copies of plotted figures.
-
Cookies, for storing small amounts of data on the client side. Best for unimportant temporary data.
- Note any applicable privacy laws, and that cookies typically expire after a given deadline.
- If your app is later moved to a different domain, typically it can no longer access the cookies from the old domain.
- To allow the user to transfer such data manually, you could zip and base64 encode it, then allow the user to enter such a base64 encoded string to load the data back in.
-
A custom server-side app, which can accept e.g. a
multipart/form-data
HTTP POST request, and save the data into a file at the backend.- See Resumable.js for a frontend library that can send data in that format. For a matching backend for Node.js, see resumable-node.
Proper error handling mostly isn't a thing in the ES culture, but the language does support try {...} catch (e) {...} finally {...}
. The parameter to catch
is mandatory, until very recent versions of the language.
The curly braces are always mandatory, unlike in some other constructs such as for
and if
.
Catching a specific exception type isn't supported by the standard, but you can catch, test (type or duck), and re-throw when no match:
try {
some_failing_operation();
}
catch (e) {
if (e instanceof MyError) {
console.log("yay");
}
else {
throw(e);
}
}
There are very few builtin exception types, but much of the time, even in Python, TypeError
(ES has that too) and ValueError
(ES: RangeError
) are the most useful. If you want to make your own exception type to be able to catch that specifically:
class MyError extends Error {
constructor (...args) { super(...args); }
}
Unfortunately that's a bit more boilerplate than Python's class MyError(Exception): pass
, but that's ES for you. The super
call is really necessary even in simple cases like this.
Here the ...
is literal; it's the splat operator, like *
in Python. In a parameter list, ...
defines a rest parameter, which must be the last one in the parameter list. It collects all remaining arguments into an array. In a function call, ...
unpacks an array into individual arguments. So, as a pythonista, you can mostly think of ES's ...
as the same as Python's unpacking *
.
ES as a language has many pitfalls. (That's not to say Python doesn't have its share, such as mutable default arguments; the loop variable scoping thing that affects closures created in a loop; and the various import traps for the unwary).
Examples:
-
Syntax for function expressions vs. function statements. If the context allows a statement, that takes precedence.
- Sometimes this means you must notice the parentheses around the construct vs. the lack thereof, or get smited.
-
Implicit semicolons: automatic semicolon insertion (ASI). It may do what you mean, except sometimes not.
- Unlike in Python, in ES newlines and indentation don't matter (except to humans); semicolons and curly braces do.
- ASI is an attempt to paper over this, but the ES language wasn't originally designed to be written without semicolons, so the rules may seem arbitrary and be difficult to remember.
-
The implicit
this
parameter, whose value depends on how the function was invoked.- In the wild, you may sometimes see the pattern
var that = this;
. There's nothing special aboutthat
; the intent is to store thethis
value from that scope for use in inner functions (which have their own implicitthis
, shadowing the outer one).
- In the wild, you may sometimes see the pattern
-
In strict mode, at the top level
this
isundefined
.- Before ES6, there was no standard portable name for the global object. In the frontend, you had to explicitly use
window
, and in the Node.js backend,global
. - Fortunately, ES6 added the silly-named
globalThis
, which should work uniformly in any ES6 environment. - OTOH, the main use of the global object - without caring about what it is - is for the pre-ES6 module pattern, to be able to inject exports, and then to access them from other scripts. If you use ES6 modules, this use case vanishes.
- Before ES6, there was no standard portable name for the global object. In the frontend, you had to explicitly use
-
Hoisting may cause the execution order to differ from the presentation order of the source code.
- Particularly, a function definition may be available earlier than it appears in the source code.
- See the temporal dead zone (TDZ) concerning differences between
var
vs.let
andconst
variables in how they hoist.
-
For nested function definitions, function statements are only allowed at the top level of the surrounding function, but variables can be defined anywhere.
- So you can just define a variable and assign a function expression to it to define a function anywhere.
- Function statements are hoisted to the beginning of the surrounding function scope (or top-level scope when outside any function). Function expressions are not hoisted.
- Just like in Python, nesting function definitions creates closures.
-
Creating an object instance requires saying
new Foo(...)
, not justFoo(...)
, unlike in Python; using thenew
operator will set the rightthis
for you.-
Calling the constructor function normally (without the
new
) fails to setthis
to the new object instance, which may smite you much later. -
The ES6
class
syntax helps. A constructor of aclass
cannot be called withoutnew
; ES will emit a helpfulTypeError
when attempting to call aclass
constructor without thenew
. -
For classical function-based constructors (no
class
) you could:function ctor (f) { // pythonic decorator: make f into a constructor function autonew (...args) { return new f(...args); } return autonew; } var MyObject = ctor(function (...) {...}); var mo = MyObject(); // now ok, MyObject is a ctor
but it's not idiomatic (and this barebones gist fails to set the name for tracebacks correctly).
-
-
Arrow functions don't have their own
this
. This lack of uniformity is supposedly a feature.- You can just use a function expression instead of an arrow function, but for the reasons in the link, an arrow function is sometimes a better fit in practice. Particularly, because an arrow function doesn't have its own
this
, using an arrow function makes it easy to refer to the surrounding context'sthis
, without defining athat
. - Somehow this brings to mind how the schizophrenic handling of indentation vs. parentheses in statements vs. expressions in Python is supposedly a feature. That's just nonsense, even though it comes from the BDFL. As SRFI-110: Sweet expressions shows, a consistent uniform approach is possible, with indentation and parentheses acting as alternative notation for structure regardless of context.
- You can just use a function expression instead of an arrow function, but for the reasons in the link, an arrow function is sometimes a better fit in practice. Particularly, because an arrow function doesn't have its own
Yes, literally. Even some of the Python builtins, though the language features to support them exist.
To get you started, here are custom ES implementations of pythonic all
, any
, range
, zip
and enumerate
:
function all (predicate, iterable) { // Same semantics as in Python.
for (const x of iterable)
if (!predicate(x))
return false;
return true;
}
function any (predicate, iterable) { // Same semantics as in Python.
for (const x of iterable)
if (predicate(x))
return true;
return false;
}
// Unlike Python, we allow Infinity, so this also provides the equivalent of `itertools.count`.
// function* range(stop, {start=0, step=1} = {}) { // if you want to emulate kwargs.
function* range (...args) { // lazy, like in Python.
var start=0, stop=Infinity, step=+1;
if (args.length === 1)
[stop] = args;
else if (args.length === 2)
[start, stop] = args;
else if (args.length === 3)
[start, stop, step] = args;
else if (args.length > 3)
throw new TypeError("range: invalid call signature. Usage: range(), range(stop), range(start, stop) or range(start, stop, step).")
if (step === 0)
throw new RangeError("range: step must be nonzero.")
var cmp = (step > 0) ? ((a, b) => (a < b)) : ((a, b) => (a > b));
for (let k = start; cmp(k, stop); k += step)
yield k;
}
function iter (iterable) { // FP API, like in Python
const iterator = iterable[Symbol.iterator]();
return iterator;
}
function next (iterator) { // FP API, like in Python
return iterator.next();
}
function* zip (...xss) { // zip(as, bs, ...) --> [[a[0], b[0], ...], [a[1], b[1], ...], ...]
var its = xss.map(xs => iter(xs));
while (true) {
let vals = its.map(it => next(it));
// Unlike Python, which raises StopIteration when the iterable runs out,
// ES sets the `done` attribute of the object returned by `next`.
if (any((v => v.done), vals))
return;
yield vals.map((v => v.value), vals);
}
}
function* enumerate (iterable, start=0) {
let k = start;
for (let x of iterable) {
yield [k, x];
k++;
}
}
I have included iter
and next
for completeness, because defining a pythonic FP API for those makes the zip
implementation read more naturally for pythonistas. To build stuff like this, see e.g. the SO posts [1], [2].
Well, yes and no. The attentive reader may have noticed something called Symbol
being used in iter
.
The natural habitat of the symbol concept is the Lisp family, but as Paul Graham once famously said, Python effectively has a symbol data type but no syntax for it. The interpreter interns some short strings, so e.g. in CPython 3.6 "foo" is "foo"
evaluates to True
, despite the appearance that these are two different instances of the string "foo"
. The defining properties of symbols are that (1) a symbol is equal only to itself, and (2) to check whether two given symbols are equal, a pointer comparison (i.e. comparing object identity) is sufficient. This is achieved by interning the string representation. So it can be argued that in Python, "foo"
is a symbol. You just have to be aware of the rules which strings get interned; and since in Python this is considered a language implementation detail, you can't rely on it across versions or different interpreters (e.g. PyPy3).
In contrast, in Lisp, 'foo
is a symbol. Interning is guaranteed. It is different from the string "foo"
, for which no such guarantee is made.
Now, if you have some Lisp background, beware of the name. In ES, constructing a new Symbol('foo')
does the job of Lisp's (gensym 'foo)
; it creates a new unique symbol which is guaranteed not to conflict with any previous return value of Symbol
. To be able to refer to an existing symbol, the only way (just like with gensym
) is to stash the return value of the Symbol
constructor when you create that symbol. That return value is your symbol. By design, there's no other way to refer to a gensym (i.e., a symbol produced by gensym
; or in ES terminology, by the Symbol
constructor).
The ES standard predefines some symbols, and stashes them as properties on the Symbol
object. Particularly, the symbol Symbol.iterator
is effectively the name of the ES equivalent of Python's __iter__
method - without causing any name conflicts, no matter what names are used for any other properties of the object. Nothing except Symbol.iterator
itself compares equal to Symbol.iterator
, so no name can conflict with it.
When referring to properties named with a symbol, the square bracket notation comes in useful, because by design, a gensym has no textual representation. There is no string name you could place in the source code to refer to a gensym. A gensym may have a human-readable string representation that is printed when you print it, but that string is not its name; it cannot be used as an identifier to refer to it. (This is a feature; if it could, guaranteeing the absence of name conflicts between gensyms would be difficult or impossible.)
So in the iter
implementation above, iterable[Symbol.iterator]
retrieves the ES equivalent of the __iter__
method of the object iterable
, and then the ()
just calls the method.
Just google for libraries. Plonk the library files where the browser can find them (in the context of your app), and tell your HTML to load them using <script>
tags.
Because in the browser, the theoretically simple becomes a legacy-driven quirk-hole (Jake Archibald), to get the load order of your .js
files right, see Deep dive into the murky waters of script loading, and the SO questions [1], [2]. Also a look at various onload events may be useful.
You can also use the ES6 module system and import the module in your .js
, if the libraries you're using support ES6 modules and your app uses them.
The backend engine Node.js has its own module system that predates ES6. NPM is the Node.js equivalent of PyPI. There's a package manager, npm
.
If you want to install Node.js-based tools (such as eslint
or tern
) globally (not just under a specific project), you'll want to set up a package.json
to appease the spirits, even though your tool installation directory is obviously not an NPM package. Use npm init
to run a questionnaire that will generate the file for you. It will produce something like this:
{
"name": "mydummypackage",
"version": "1.0.0",
"description": "Default package for npm installs",
"main": "index.js",
"dependencies": {
},
"devDependencies": {
},
"scripts": {
"test": "echo \"Error: no test specified\" && exit 1"
},
"author": "",
"license": "ISC"
}
You may also want to look into the npx
runner utility.
But if you don't need that many Node.js-based tools, the classic manual solution ln -s .../node_modules/sometool/bin/sometool ~/.local/bin/sometool
is also fine.
Things you may see in the wild.
Libraries often come in minified form (.min.js
), using one-character identifiers and no whitespace. It's still ES, just not human-readable. This optimizes download size; obfuscation is a side effect. There are several minifiers available.
Debuggers often support a map file that maps locations in the minified file to the original source code. Alternatively, when testing locally, you can just use the regular non-minified version of the library.
Beside minification, tree shaking, seen e.g. in Webpack and in rollup.js, is a commonly employed optimization technique for optimizing download size. It bundles only the code that a static analysis indicates can actually be reached (when starting from the initial entry point of the program).
Babel is an ES transpiler that takes in recent ("next-gen") ES, and produces equivalent older ES, compatible with older browsers.
Compatibility between different environments providing ECMAScript:
- Frontend: Kangax's compatibility table.
- Backend: Node.js compatibility table.
- Check an individual feature: caniuse.com.
As a rule of thumb for modern ES features in frontend work, Internet Explorer supports essentially nothing. All other browsers (even Edge), almost everything. Chrome and Opera are the best, with Firefox not too far behind. So for the frontend, to help yourself stay sane, just require your users to use anything except IE. Even corporate-minded environments should have at least Edge by now. (And everyone else already has Firefox or Chrome.)
If you ♡:
- FP: Ramda, a practical functional library.
- advanced math: LALOLib. E.g. solve eigenvalue problems using nothing but ES. Ideal for the frontend.
- data visualization (plotting): D3.js.
- syntactic macros: Sweet.js, the macro compiler for ES.
- Unlike MacroPy, Sweet.js requires a separate compile step, because ES has no load-time hook to plug a macro expander in to.
Decent development tools are essential:
- ESLint, the static analyzer. Catch potential bugs and style problems before you run.
- Tern.js, the smart autocompleter. It's a plugin supported by many editors/IDEs.
- A decent editor/IDE, such as Spacemacs, the beefed-up Emacs, with its
javascript
layer.- The layer uses
js2-mode
as the IDE. It will automatically use ESLint and Tern.js if available. - Plugins such as smartparens, rainbow-delimiters and highlight-parentheses really make a difference in a paren-heavy language (i.e. almost everything except Python).
- Emacs (in any variant, starting with version 24.4) also lets you prettify your view of the code, without actually changing the content of the source file you're editing. For example, if you find concise notation efficient, as in math, and want the ES keyword
function
to be displayed asλ
, that's possible. See prettify-symbols-mode. For reference, here's my config. - Obviously, there are other IDEs and I just happen to prefer a modern Emacs; your mileage may vary.
- The layer uses
- Some major web browsers come with a frontend debugger.
As for backend debuggers, you'll have to search for them yourself. I've debugged the backend the same way I debug Python, with copious amounts of console.log
(the ES equivalent of Python's print
).
That's all, folks!