Proposal for adding a mechanism for enabling a more extensive standard library in JavaScript via a collection of built in modules. With this infrastructure in place it will be possible to start iterating on standard library features as additional modules.
Table of Contents
The goal of this proposal is to define a mechanism for enabling a more extensive standard library in JavaScript than what is available now. Currently it is the case if new properties are added to the language, they are added someplace in the global object.
This proposal would not change the behavior of any existing code or add any new syntax, except possibly syntax needed for importing the standard library code. There likely will be new properties and methods added to enable this feature.
The contents of the standard library is tangential to this proposal, and would be built and expanded upon in later efforts. Such a library would only cover features which would be useful in JavaScript in general, not things which are tied to the web platform. Host environments built on JavaScript could provide their own library components. (A good heuristic: if something would make sense on a web browser but not in node or on embedded devices or robots, it probably isn't in scope.) See #16 for discussion of the extent and contents of the library.
Most programming languages have core language features (syntax, operators, primitives etc.) along with a standard library of commonly used functionality. Developers are able to use this library in their programs immediately because the library is bundled with the runtime.
The JavaScript language does not have a standard library. As a result new functionality is added to the global object, or developers find and adopt libraries that they bundle with the rest of their application. Because these libraries need to be included in every program instead of being provided by the runtime, the code size of applications grow, with users paying the cost of downloading and parsing the common library components. This can increase downloading times and makes it nearly impossible for JavaScript engines to cache the common code.
JavaScript programs run on a variety of clients with different runtime implementations and versions. It is quite likely that the bundled library code needs to account for different host environments, thus increasing code size. Library code bundled with the JavaScript engine could be optimizied for the target host environment.
To enable developers to access common functionality provided by the JavaScript engine we propose creating modules built into the host. This will allow library code bundled with the program to shift towards being available on the host JavaScript implementations.
Disclaimer: This proposal covers adding a mechanism for enabling a standard library in JavaScript implementations, it does not describe its contents. This is considered a tangential effort that would be built and expanded upon in the future when a mechanism for accessing it is in place.
Having a standard library readily available at runtime means programs won't have to include the functionality available in the standard library, reducing the download and startup cost of the program. The functionality will also be standardized across implementations giving developer consistency in quality, behavior and speed. For some JavaScript engines, built in modules will allow them to reduce application memory footprint
Although JavaScript engines already has the notion of a common library through the global object,
the standard library components can be accessed using the current asynchronous import() API.
Module scripts can also access built in library components using the import syntax.
Traditional script code will be able to access these same standard library components through
a new synchronous API, importNow().
Most of these mechanism should already be familiar to developers, allowing them to opt-in to
this built in library functionality with little effort.
Modules for the standard library should be able to be written in plain JavaScript for the most part but for hot pieces of code the engine could provide a native implementation.
To import modules from the standard library the engine has to be able to distinguish between standard library
modules and other (user defined) modules. To allow the engine to do this standard library modules will use a
prefix in the module identifier string. This is prefered over other alternatives because it is does not
introduce new syntax for loading standard library modules and stays close to the import statement developers
should already be familiar with.
The js: prefix will be reserved as the namespace for the JavaScript language only and will be governed by
TC39, making the standard library a true JavaScript standard library. This will allow the committee to work
safely within this namespace when designing and developing the standard library over time.
Alternatives for distinguishing standard modules are documented in Appendix A.
By creating a namespaces specifically for the JavaScript standard library, developers will know what to expect
when importing from using the js: prefix across different implementations and can be assured the same
modules are available across these implementations (not considering implementation constraints, vendor
timelines or version differences).
It is completely feasable that more namespaces are introduced which are goverened by other standards bodies or organizations. However it is important that these namespaces stay independent of each other to avoid conflicts, hamper development within namespaces due to outside pollution or time constraints due to dependencies on other organizations.
The namespace used for the JavaScript standard library will be registered with IANA to prevent collisions in the future. Any other organization introducing a namespace prefix would be encouraged to do the same.
All exported objects and classes from the standard library will have their prototype frozen. This will prevent prototypes from imported objects to be modified outside of the module causing prototype pollution.
In the past the committee had to make concessions to maintain web compatibility when adding new functionality
to built-in objects. By freezing the prototype of standard library exports, it will no longer be possible for
third party code to modify or extending library code in a possibly incompatible way. This will allow for more
flexibility when designing and developing the standard library. Extending standard library classes and objects
can still be done using extend or Object.create.
We can start off by conventionally enforcing Object.freeze on exported Objects from standard library modules.
If this turns out to be hard to check and enforce a separate proposal can be created to describe automatically
freezing prototypes at the module boudary for standard library modules.
Note, the current direction is to freeze module contents thus requiring shimming code to wrap module classes and objects. Given some discussion around this point, it may be dropped from the proposal before built in modulesa re added to the standard.
To allow the JavaScript engine to participate in module resolution and loading the HostResolveModuleIdentifier Abstract Operation should be changed to allow more than one resolver. The resolvers are arranged in a chain and consulted by the engine one by one to resolve a requested ModuleIdentifier. This mechanism is heavily inspired by Pythons importing system for modules.
The implementation of the chain will be hidden inside the engine and will provide the same guarantees HostResolveModuleIdentifier currently does:
- It must return an instance of ModuleRecord
- It must throw an error if a ModuleRecord cannot be created
- It must be idempotent for pair of (ModuleIdentifier, ReferencingModule)
Importing is done in two phases: resolution and loading. Resolution is done through objects implementing the ResolveModuleIdentifier operation and loading is done through objects implementing the LoadModuleIdentifier operation. Objects implementing both operations are referred to as importers.
Multiple importers can be registered with the engine and will be appended to the chain. The engine will always register the internal JavaScript Standard Library importer first to make sure it is at the front of the chain.
During the resolution step (phase 1) importers perform the ResolveModuleIdentifier operation in order of registration. This operation results in a ModuleResolutionRecord that is passed along the chain and can be used by subsequent importers.
After the resolution step the loading step (phase 2) is invoked. During this step the importers are visited in reverse order to perform the LoadModuleIdentifier operation to try and load the requested ModuleIdentifier. The operation is passed the ModuleResolutionRecords from the resolution step. When the ModuleIdentifier can be loaded a full ModuleRecord is returned and the chain is exited immediately. When no ModuleRecord is produced the next importer is consulted until the chain is exhausted, resulting in a ModuleNotFound error.
During the loading step (phase 2) the chain is traversed in reverse order to allow for higher ranked (e.g. registered later) importers to possibly override lower ranked importers. This will allow the host for example to override standard library modules and achieve shimming / polyfilling.
Alternatively, provisions are made for code to import a built-in module, shim it and update the host module table to point at the shimmed version. This can be do repeatively so that shims / polyfills can be layored on top of each other.
There are three use cases that a polyfilling solution for the standard library should support:
- Add missing parts of the standard library
- Update incomplete implementations
- Redact parts of a library component.
Polyfilling is intended to cover these three use cases only.
For the web platform polyfilling could be done using the Import Maps Proposal. A resolver registered by the embedder (a web browser in this case) could check the import map to see if a standard library ModuleIdentifier should be redirected to another implementation.
This subject has been talked about in the past and has related efforts:
Coming Soon™
One of the requirements for the standard library is being able to distinguish when a module is user defined and when it should be loaded from the standard library. There are a lot of approaches to doing this and this section contains alternatives considered to the recommendation earlier in the proposal.
Important: The code samples and modules used in the examples are purely for illustrative purposes
Identifier Based
When importing a module the ModuleSpecifier is required to be a string. Interpreting the contents of the string is currently left to the embedder effectively always making it a user defined module.
To circumnavigate this and preserving the current behavior a different ModuleSpecifier form using a special Identifier could be used instead of a string literal:
import { ... } from std.SomeStandardModule;The std identifier is used to change the signature in a way that makes it possible for engines to detect
this is a module that should be loaded from the standard library. The downside of using std.________ is that
it start to look like a global object that is also available in other contexts.
It could also be possible to use specialized tokens in place of Identifier prefix, for example similar to C/C++:
import { ... } from <SomeStandardModule>;While this makes importing standard library modules distinctly different from user defined modules this is also one of the downsides. The syntax is different from the import syntax developers should already be familiar with and a dynamic variant would be difficult (gramar wise).
Not using a prefix also has the downside of requiring everything from the standard library to live in the same space creating a new “global” namespace and preventing the use of multiple namespaces for different contexts.
Reference: tc39/ecma262#395-comment-371865432
Separate Keyword
When importing a module the ModuleSpecifier is required to be a string. Interpreting the contents of the string is currently left to the embedder effectively always making it a user defined module.
To circumnavigate this and preserving the current behavior a different keyword from import could be
introduced specifically for importing modules from the standard library:
include { ... } from <SomeStandardModule>;Important: The
includekeyword does not exist and is used for illustrative purposes only
The seperate keyword is emphasised by combining it with an Identifier for the ModuleSpecifier. The separate keyword already distingishes the module enough so we can also safely use a string literal for the ModuleSpecifier:
include { ... } from "SomeStandardModule";While this makes importing standard library modules very distinctly different from user defined modules this
is also one of its downsides. The syntax is different from the import syntax developers should already be
familiar with and a dynamic variant of the keyword would also have to be created.
Not using a prefix also has the downside of requiring everything from the standard library to live in the same space creating a new “global” namespace and preventing the use of multiple namespaces for different contexts.
Reference: tc39/ecma262#395-comment-196917747
URL Based
There is already support for importing modules from a URL in browsers like Safari using import
statements inside of a <script type="module" ... />. The host portion of the URL could be used as a prefix
for distinguishing standard library modules and fits very natural within the concept of URLs:
import { ... } from "https://www.ecma-international.com/ecmascript/SomeStandardModule";Using a URL with a domain for importing modules does not always make sense outside of a web context, for example in Node.js or embedded devices. It also requires ownership of the domain over a long period of time. A transfer of the domain might make the prefix useless or exclude a number of modules from being imported.
Reference: tc39/ecma262#395-comment-328528910
NPM Style
The Node Package Manager (NPM for short) has an established format for importing packages published to a central registry. The format allows packages to be grouped under organizations and this could be leveraged as a prefix for the JavaScript standard library:
import { ... } from "@std/SomeStandardModule";While this format makes sense in environments where NPM is already used it is not as universal as a URL. The NPM style format also suffers some of the same downsides as a URL Based approach.
This format would require ownership over the organization on NPM (https://npmjs.com) and keeping multiple namespaces in mind would require all to do the same. But NPM is only the most commonly used registry, there might be other registries, possibly private ones that would be impossible to get ownership over a namespace. Using this format could exclude modules from being imported.
The biggest downside utilizing this format is standard library modules looking the same as user defined modules which might cause confusion, for example where modules come from in a Node.js/NPM context. The two modules behave different and the same assumptions can't be applied to both.
Reference: tc39/proposal-javascript-standard-library#16-comment-
- none yet