Awesome

Call-this operator for JavaScript

ECMAScript Stage-1 Proposal. J. S. Choi, 2021.

• Formal specification
• Babel plugin: Not yet

This proposal is a resurrection of the old Stage-0 bind-operator proposal. It is also an alternative, competing proposal to the Stage-1 extensions proposal. For more information, see § Related proposals.

Syntax

The syntax is being bikeshedded in issue #10.

receiver~>fn(arg0, arg1)
receiver~>ns.fn(arg0, arg1)
receiver~>(expr)(arg0, arg1)

A member expression, a call expression, an optional expression, a new expression with arguments, another call-this expression, or a parenthesized expression. The value to which this expression resolves will be bound to the right-hand side’s function object, as that function’s this receiver.

A variable that must resolve to a function object.

Instead of a single variable, the right-hand side may be a namespace object’s variable, followed by a chain of property identifiers. This chain must resolve to a function object.

An arbitrary expression within parentheses, which must resolve to a function object.

A series of argument expressions, which may include spread ... syntax.

Description

The syntax is being bikeshedded in issue #10.

(A formal specification is available.)

The call-this operator ~> is a left-associative binary operator. It calls its right-hand side (a function), binding its this value to its left-hand side (a receiver value), as well well as any given arguments – in the same manner as Function.prototype.call.

For example, receiver~>fn(arg0, arg1) would be equivalent to fn.call(receiver, arg0, arg1) (except that its behavior does not change if code elsewhere reassigns the global method Function.prototype.call).

Likewise, receiver~>(createFn())(arg0, arg1) would be roughly equivalent to createFn().call(receiver, arg0, arg1).

If the operator’s right-hand side does not evaluate to a function during runtime, then the program throws a TypeError.

The operator’s left side has equal precedence with member expressions, call expressions, new expressions with arguments, and optional expressions. Like those operators, the call-this operator also may be short-circuited by optional expressions in its left-hand side.

The operator’s right-hand side, as with decorators, may be one of the following:

• A single identifier or private field (like fn or #field).
• A chain of identifiers and/or private fields (like ns.fn or this.ns.#field).
• A parenthesized expression (like (createFn())).

For example, receiver~>ns.ns.ns.fn groups as receiver~>(ns.ns.ns.fn).

Similarly to the . and ?. operators, the call-this operator may be padded by whitespace or not.
For example, receiver~>fn
is equivalent to receiver~>fn,
and receiver~>(createFn())
is equivalent to receiver~>(createFn()).

The call-this operator may be optionally chained with ?. (i.e., ?.~>):
receiver~>fn will always result in a bound function, regardless of whether receiver is nullish.
receiver ?.~>fn will result in null or undefined if receiver is null or undefined.
receiver ?.~>fn(arg) also short-circuits as usual, before arg is evaluated, if receiver is nullish.

A new expression may not contain a call-this expression without parentheses. new x~>fn() is a SyntaxError. Otherwise, new x~>fn() would be visually ambiguous between
(new x)~>fn() and new (x~>fn()).

Why a call-this operator

In short:

1. .call is very useful and very common in JavaScript codebases.
2. But .call is clunky and unergonomic.

.call is very common

The dynamic this binding is a fundamental part of JavaScript design and practice today. Because of this, developers frequently need to change the this binding. .call is arguably one of the most commonly used functions in all of JavaScript.

We can estimate .call’s prevalences using Node Gzemnid. Although Gzemnid can be deceptive, we are only seeking rough estimations.

The following results are from the checked-in-Git source code of the top-1000 downloaded NPM packages.

These results suggest that usage of .call is comparable to usage of other frequently used standard functions. In this dataset, its usage exceeds even that of console.log.

Obviously, this methodology has many pitfalls, but we are only looking for roughly estimated orders of magnitude relative to other baseline functions. Gzemnid counts each library’s codebase only once; it does not double-count dependencies.

First, we download the 2019-06-04 pre-built Gzemnid dataset for the top-1000 downloaded NPM packages. We also need Gzemnid’s search.topcode.sh script in the same active directory, which in turn requires the lz4 command suite. search.topcode.sh will output lines of code from the top-1000 packages that match the given regular expression.

./search.topcode.sh '\.call\b' | head
grep -aE  "\.call\b"
177726827	debug@4.1.1/src/common.js:101:					match = formatter.call(self, val);
177726827	debug@4.1.1/src/common.js:111:			createDebug.formatArgs.call(self, args);
154772106	kind-of@6.0.2/index.js:54:  type = toString.call(val);
139612972	readable-stream@3.4.0/errors-browser.js:26:      return _Base.call(this, getMessage(arg1, arg2, arg3)) || this;
139612972	readable-stream@3.4.0/lib/_stream_duplex.js:60:  Readable.call(this, options);
139612972	readable-stream@3.4.0/lib/_stream_duplex.js:61:  Writable.call(this, options);
139612972	readable-stream@3.4.0/lib/_stream_passthrough.js:34:  Transform.call(this, options);
139612972	readable-stream@3.4.0/lib/_stream_readable.js:183:  Stream.call(this);
139612972	readable-stream@3.4.0/lib/_stream_readable.js:786:  var res = Stream.prototype.on.call(this, ev, fn);

We use awk to count those matching lines of code and compare their numbers for call and several other frequently used functions.

> ls
search.topcode.sh
slim.topcode.1000.txt.lz4
> ./search.topcode.sh '\.call\b' | grep -E --invert-match '//.*\.call|/\*.+\.call|[^a-zA-Z][A-Z][a-zA-Z0-9_$]*\.call\( *this|_super\.call|_super\.prototype\.|_getPrototypeOf|_possibleConstructorReturn|__super__|WEBPACK VAR INJECTION|_objectWithoutProperties|\.hasOwnProperty\.call' | awk 'END { print NR }'
315922
> ./search.topcode.sh '\.bind\b' | awk 'END { print NR }'
170248
> ./search.topcode.sh '\b.map\b' | awk 'END { print NR }'
1016503
> ./search.topcode.sh '\bconsole.log\b' | awk 'END { print NR }'
271915
> ./search.topcode.sh '\.slice\b' | awk 'END { print NR }'
182292
> ./search.topcode.sh '\.set\b' | awk 'END { print NR }'
168872
> ./search.topcode.sh '\.push\b' | awk 'END { print NR }'
70116

Note that, for .call, we use grep to exclude several irrelevant occurrences of .call either within comments or from transpiled code. We err on the side of false exclusions.

These excluded patterns were determined by an independent investigator (Scott Jamison), after manually reviewing the first 10,000 occurrences of .call in the dataset for why each occurrence occurred.

The excluded [^a-zA-Z][A-Z][a-zA-Z0-9_$]*\.call\( *this pattern deserves a special note. This pattern matches any capitalized identifier followed by .call(this. We exclude any such occurrences because any capitalized identifier likely refers to a constructor, and using .call on a constructor is a pattern that has largely been obviated by class and super syntax. It is likely that this pattern erroneously excludes many legitimate uses of .call from our count, but this bias against .call is acceptable for the purposes of rough comparison.

There are a variety of reasons why developers use .call. These include:

Wrapping a receiver’s method before calling it:

// Wrapping a receiver’s method before calling it:
assertFunction(obj.f).call(obj, f);

// From bluebird@3.5.5.
tryCatch(item).call(boundTo, e);

Conditionally switching a call between two methods:

const method = obj.f ?? obj.g;
method.call(obj, arg0, arg1);

// From debug@4.1.1.
// createDebug is an object either for Node or for web browsers.
createDebug.formatArgs.call(self, args);

Reusing an original method on a monkey-patched object:

// From graceful-fs@4.1.15.
return fs$read.call(fs, fd, /*…*/)

Protecting a method call from prototype pollution:

// From lodash@4.17.11.
// Object.prototype.toString was cached as nativeObjectToString.
nativeObjectToString.call(value);

…and other reasons. Developers do all of this using .call, and the sum of these uses propels .call to being one of the most used operations in the entire language.

.call is clunky

In spite of its frequency, .call is clunky and poorly readable. It separates the function from its receiver and arguments with boilerplate, and it flips the “natural” word order, resulting in a verb.call–subject–object word order:

fn.call(rec, arg0).

JavaScript developers are used to using methods in a subject–verb–object word order that resembles English and other SVO human languages. This pattern is ubiquitous in JavaScript as dot method calls:

rec.method(arg0).

Consider the following real-life code using .call, and compare them to versions that use the call-this operator. The difference is especially evident when you read them aloud.

// kind-of@6.0.2/index.js
type = toString.call(val);
type = val~>toString();

// debug@4.1.1/src/common.js
match = formatter.call(self, val);
match = self~>formatter(val);

createDebug.formatArgs.call(self, args);
self~>createDebug.formatArgs(args);

// rxjs@6.5.2/src/internal/operators/every.ts
result = this.predicate.call(this.thisArg, value, this.index++, this.source);
result = this.thisArg~>this.predicate(value, this.index++, this.source);

// bluebird@3.5.5/js/release/synchronous_inspection.js
return isPending.call(this._target());
return this._target()~>isPending();

var matchesPredicate = tryCatch(item).call(boundTo, e);
var matchesPredicate = boundTo~>(tryCatch(item))(e);

// async@3.0.1/internal/initialParams.js
var callback = args.pop(); return fn.call(this, args, callback);
var callback = args.pop(); return this~>fn(args, callback);

// ajv@6.10.0/lib/ajv.js
validate = macro.call(self, schema, parentSchema, it);
validate = self~>macro(schema, parentSchema, it);

// graceful-fs@4.1.15/polyfills.js
return fs$read.call(fs, fd, buffer, offset, length, position, callback)
return fs~>fs$read(fd, buffer, offset, length, position, callback)

In short:

Very common × Very clunky = Worth improving with syntax.

Concerns about ecosystem schism

The answer to whether multiple ways or syntaxes of doing something are harmful critically depends on the duplication’s effect on APIs and how viral it is.

Suppose we’re considering having two syntaxes 𝘟 and 𝘠 to use APIs. If module or person 𝘈 uses syntax 𝘟 which interoperates better with syntax 𝘟 than syntax 𝘠 and that pressures module or person 𝘉 to use syntax 𝘟 in their new APIs to interoperate with person 𝘈’s APIs, that virality encourages ecosystem forking and API wars. Introducing multiple such ways into the language is bad.

“On the other hand, if person 𝘈’s choice of syntax [i.e., 𝘟] has no effect on person 𝘉[’s choice of syntax, 𝘠,] and they can interoperate without any hassles, then that’s generally benign.”

From the 2022-01-27 dataflow meeting.

• 𝘟: Some APIs (like “functional” APIs) use non-this-based ƒs.
• 𝘠: Some APIs (like “object-oriented” APIs) use this-based ƒs.

This schism between 𝘟 APIs and 𝘠 APIs is already is built into the language. The schism is such that prominent APIs like the Firebase JS SDK have switched from 𝘠 to 𝘟 (e.g., for module splitting).

But the call-this operator, together with the pipe operator |>, would make interoperability between 𝘟 and 𝘠 more fluid – and it would make the choice between 𝘟 and 𝘠 less viral – bridging the schism:

import { x0, x1 } from '𝘟';
import { y0, y1 } from '𝘠';
input |> x0(@)~>y0() |> x1(@)~>y1();

Non-goals

A goal of this proposal is simplicity. Therefore, this proposal purposefully does not address the following use cases:

Function binding and method extraction are not a goal of this proposal.

Changing the this receiver of functions is more common than function binding, as evidenced by the preceding statistics. Some TC39 representatives have expressed concern that function binding may be redundant with proposals such as PFA (partial function application) syntax. Therefore, we will defer these two features to future proposals.

Extracting property accessors (i.e., getters and setters) is also not a goal of this proposal. Get/set accessors are not like methods. Methods are properties (which happen to be functions). Accessors themselves are not properties; they are functions that activate when getting or setting properties.

Getters/setters have to be extracted using Object.getOwnPropertyDescriptor; they are not handled in a special way. This verbosity may be considered to be desirable syntactic salt: it makes the developer’s intention (to extract getters/setters – and not methods) more explicit.

const { get: $getSize } =
  Object.getOwnPropertyDescriptor(
    Set.prototype, 'size');

// The adversary’s code.
delete Set; delete Function;

// Our own trusted code, running later.
new Set([0, 1, 2])~>$getSize();

Function/expression application, in which deeply nested function calls and other expressions are untangled into linear pipelines, is important but not addressed by this proposal. Instead, it is addressed by the pipe operator, with which this proposal’s syntax works well.

Related proposals

Old bind operator

This proposal is a resurrection of the old Stage-0 bind-operator proposal. (A champion of the old proposal has recommended restarting with a new proposal instead of using the old proposal.)

The new proposal is basically the same as the old proposal. The only big difference is that there is no unary form for implicit binding of the receiver during method extraction. (See also non-goals.)

Extensions

The extensions system is an alternative, competing proposal to the Stage-1 extensions proposal.

An in-depth comparison is also available. The concrete differences briefly are:

1. Call-this has no special variable namespace.
2. Call-this has no implicit syntactic handling of property accessors.
3. Call-this has no polymorphic const ::{ … } from …; syntax.
4. Call-this has no polymorphic …::…:… syntax.
5. Call-this has no Symbol.extension metaprogramming system.

Pipe operator

The pipe operator is a complementary proposal that can be used to linearize deeply nested expressions like f(0, g([h()], 1), 2) into h() |> g(^, 1) |> f(0, ^, 2).

This is fundamentally different than the call-this operator’s purpose, which would be much closer to property access ..

It is true that property access ., call-this, and the pipe operator all may be used to linearize code. But this is a mere happy side effect for the first two operators:

• Property access is tightly coupled to object membership.
• Call-this is simply changes the this binding of a function call.

In contrast, the pipe operator is designed to generally linearize all other kinds of expressions.

|> does not improve .call’s clunkiness. Here is the clunky (and frequent) status quo again:

fn.call(rec, arg0)

Introducing the pipe operator |> fixes word order, but the result is even less readable. Excessive boilerplate separates the function from its receiver and arguments:

rec |> fn.call(@, arg0) // Less readable.

Only a separate operator can improve the word order without otherwise compromising readability:

rec~>fn(arg0)

The pipe champion group have been investigating whether it is possible to modify the pipe operator to address .call’s clunkiness while still addressing pipe’s other use cases (e.g., non-this-based, n-ary function calls; async function calls). It has still found none except a separate operator.

Just like how the pipe operator coexists with property access:

// Adapted from react@17.0.2/scripts/jest/jest-cli.js
Object.keys(envars)
  .map(envar => `${envar}=${envars[envar]}`)
  .join(' ')
  |> `$ ${^}`
  |> chalk.dim(^, 'node', args.join(' '))
  |> console.log(^);

…so too can it work together with call-this:

// Adapted from chalk@2.4.2/index.js
return this._styles
  |> (^ ? ^.concat(codes) : [codes])
  |> this~>build(^, this._empty, key);

PFA syntax

PFA (partial function application) syntax ~() would tersely create partially applied functions.

PFA syntax ~() and call-this ~> are also complementary and handle different use cases.

For example, obj.method~() would handle method extraction with implicit binding, which call-this does not address. In other words, when the receiver object itself contains the function to which we wish to bind, then we need to repeat the receiver once, with call-this. PFA syntax would allow us to avoid repeating the receiver.

n.on("click", v.reset.bind(v))
n.on("click", v.reset~())

In contrast, call-this changes the receiver of a function call. receiver~>fn(). (This unbound function might have already been extracted from another object.) PFA syntax does not address this use case.

// bluebird@3.5.5/js/release/synchronous_inspection.js
isPending.call(this._target())
this._target()~>isPending()