Champions: Sven Sauleau (@xtuc), Daniel Ehrenberg (@littledan), Myles Borins (@MylesBorins), Dan Clark (@dandclark), and Nicolò Ribaudo (@nicolo-ribaudo).
Status: Stage 3
⚠️ The specification in this proposal might be out of date. tc39/ecma262#3057 is the latest version.Some of the changes present in the current specification have not been presented to committee yet: #142
Please leave any feedback you have in the issues!
The Import Attributes proposal, formerly known as Import Assertions, adds an inline syntax for module import statements to pass on more information alongside the module specifier. The initial application for such attributes will be to support additional types of modules in a common way across JavaScript environments, starting with JSON modules.
The syntax will be as follows (shown here is the proposed method for importing a JSON module):
import json from "./foo.json" with { type: "json" };
import("foo.json", { with: { type: "json" } });The specification of JSON modules was originally part of this proposal, but it was resolved during the July 2020 meeting to split JSON modules out into a separate Stage 3 proposal.
Standards-track JSON ES modules were proposed to allow JavaScript modules to easily import JSON data files, similarly to how they are supported in many nonstandard JavaScript module systems. This idea quickly got broad support from web developers and browsers, and was merged into HTML, with an implementation for V8/Chromium created by Microsoft.
However, in an issue, Ryosuke Niwa (Apple) and Anne van Kesteren (Mozilla) proposed that security would be improved if some syntactic marker were required when importing JSON modules and similar module types which cannot execute code, to prevent a scenario where the responding server unexpectedly provides a different MIME type, causing code to be unexpectedly executed. The solution was to somehow indicate that a module was JSON, or in general, not to be executed, somewhere in addition to the MIME type.
Some developers have the intuition that the file extension could be used to determine the module type, as it is in many existing non-standard module systems. However, it's a deep web architectural principle that the suffix of the URL (which you might think of as the "file extension" outside of the web) does not lead to semantics of how the page is interpreted. In practice, on the web, there is a widespread mismatch between file extension and the HTTP Content Type header. All of this sums up to it being infeasible to depend on file extensions/suffixes included in the module specifier to be the basis for this checking.
There are other possible pieces of metadata which could be associated with modules, see #8 and tc39/proposal-import-reflection#18 for further discussion.
Proposed ES module types that are blocked by this security concern, in addition to JSON modules, include CSS modules and potentially HTML modules if the HTML module proposal is restricted to not allow script.
There are three places where this data could be provided:
- As part of the module specifier (e.g., as a pseudo-scheme)
- Separately, out of band (e.g., a separate resource file)
- Challenges: How to load that resource file; what should the format be; unergonomic to have to jump between files during development (further discussion: #13)
- In the JavaScript source text
- Challenges: Requires a change at the JavaScript language level (this proposal)
This proposal pursues the third option, as we expect it to lead to the best developer experience, and are hopeful that language design/standardization issues can be resolved.
Import attributes have to be made available in several different contexts. They use a key-value syntax is used preceded by the with keyword, with the key type used as an example indicating the module type. Such key-value syntax can be used in various different contexts.
The ImportDeclaration would allow any arbitrary attributes after the with keyword.
For example, the type attribute could be used to indicate a module type, for example importing a JSON module with the following syntax.
import json from "./foo.json" with { type: "json" };The with syntax in the ImportDeclaration statement uses curly braces, for the following reasons (as discussed in #5):
- JavaScript developers are already used to the Object literal syntax and since it allows a trailing comma copy/pasting attributes will be easy.
- it clearly indicates the end of the attributes list when splitting them across multiple lines.
Similar to import statements, the ExportDeclaration, when re-exporting from another module, would allow any arbitrary attributes after the with keyword.
export { val } from './foo.js' with { type: "javascript" };The import() pseudo-function would allow import attributes to be indicated in an options bag in the second argument.
import("foo.json", { with: { type: "json" } })The second parameter to import() is an options bag, with the only option currently defined to be with: the value here is an object containing the import attributes. There are other proposals for entries to put in the options bag: for example, the Module Source Imports proposal introduces a phase property.
Host environments (e.g., the Web platform, Node.js) often provide various different ways of loading modules. The analogous string could be passed through these ways of loading other kinds of modules.
new Worker("foo.wasm", { type: "module", with: { type: "webassembly" } });Sidebar about WebAssembly module types and the web: it's still uncertain whether importing WebAssembly modules would need to be marked specially, or would be imported just like JavaScript. Further discussion in #19.
Although changes to HTML won't be specified by TC39, an idea here would be that each import attribute, preceded by with, becomes an HTML attribute which could be used in script tags.
<script src="foo.wasm" type="module" withtype="webassembly"></script>(See the caveat about WebAssembly above.)
In the context of the WebAssembly/ESM integration proposal: for imports of other module types from within a WebAssembly module, this proposal would introduce a new custom section (named importattributes) that will annotate with attributes each imported module (which is listed in the import section).
This proposal does not specify behavior for any particular attribute key or value. The JSON modules proposal will specify that type: "json" must be interpreted as a JSON module, and will specify common semantics for doing so. It is expected the type attribute will be leveraged to support additional module types in future TC39 proposals as well as by hosts. HTML and CSS modules are under consideration, and these may use similar explicit type syntax when imported.
Attributes in addition than type may also be introduced for purposes not yet foreseen.
JavaScript implementations are encouraged to reject attributes and type values which are not implemented in their environment (rather than ignoring them). This is to allow for maximal flexibility in the design space in the future--in particular, it enables new import attributes to be defined which change the interpretation of a module, without breaking backwards-compatibility.
Why not both? The champions of this proposal think that exploring both an in- and out of band solutions to various kinds of metadata. While we prefer in-band metadata for module types, we are happy to see the development of various out-of-band manifests of modules being proposed and implemented in certain JS environments:
- import maps to map module specifiers to URLs/paths
- Node.js policy files, e.g., to perform integrity checks on modules
This proposal does not exclude out-of-band metadata being used for module types. And it definitely doesn't argue that all metadata should be in-band. For example, integrity hashes simply don't work in-band, both because module circularities make them impossible to calculate, and because of the need for a "cascading" update when a deep dependency changes.
Out-of-band solutions face certain downsides; these are not necessarily fatal, but are interesting to take into account when considering the solution space and making tradeoffs:
- By-hand authoring experience: While an in-band solution is somewhat verbose, it is also more straightforward for developers to adopt when writing code without much tooling. For smaller projects developers do not need to create an extra file by hand.
- Tooling complexity for large projects: For large project with many dependencies, developers will not have to worry about creating a large manifest by compiling the metadata of all of their dependencies. Module authors will also not have to worry about shipping a manifest in order for consumers to be able to run their modules.
- Performance tradeoffs: The experience in Node.js's experimental, out-of-band policy files is that they can carry significant startup cost, due to certain aspects of loading and parsing.
A central goal of this proposal is to share as much syntax and behavior across JavaScript environments as possible. To the same end, we also propose a standardization of JSON modules to the extent that this is possible (omitting just the contents of the redundant type check, which necessarily differs between environments, in addition to the pre-existing host-defined parts such as interpreting the module specifier and fetching the module).
However, at the same time, behavior of modules in general, and the set of module types specifically, is expected to differ across JavaScript environments. For example, WebAssembly, HTML and CSS modules may not make sense in certain minimal embedded JavaScript environments. We hope that environments can experiment and collaborate where it makes sense for them.
We see the management of compatibility issues across environments as similar, independent of whether metadata is held in-band or out-of-band. An out of band solution would also suffer from the risk of inconsistent implementation or support across host environments if some kind of coordination does not occur.
The topic of attribute divergence is further discussed in #34.
Attributes are part of the module cache key and can affect how a module is loaded: the cache key is extended from (referrer, specifier) to (referrer, specifier, attributes).
Why not use more terse syntax to indicate module types, like import json from "./foo.json" as "json"?
Another option considered and not selected has been to use a single string as the attribute, indicating the type. This option is not selected due to its implication that any particular attribute is special; even though this proposal only specifies the type attribute, the intention is to be open to more attributes in the future. (discussion in #12).
We could permit import attributes to have more complex values than simply strings, for example:
import value from "module" with { attr: { key1: "value1", key2: [1, 2, 3] } };This would allow import attributes to scale to support a larger variety of metadata.
We propose to omit this generalization in the initial proposal, as a key/value list of strings already affords significant flexibility to start, but we're open to a follow-on proposal providing this kind of generalization.
We are planning to make descisions and reach consensus during specific stages of this proposal. Here's our plan.
Original plan before Stage 2 and Stage 3
We have achieved consensus on the following core decisions as part of Stage 2, including:
- The attribute form; key-value or single string (#12)
// Not selected
import value from "module" as "json";
// Not selected
import value from "module" with type: "json";
// Proposal that was approved from Stage 2 to Stage 3 the first time
import value from "module" assert { type: "json" };After Stage 2 and before Stage 3, we're open to settling on some less core details, such as:
- Considering alternatives for the
with/if/assertkeywords (#3)
import value from "module" when { type: 'json' };
import value from "module" given { type: 'json' };- How dynamic import would accept import attributes:
import("foo.wasm", { with: { type: "webassembly" } });For consistency the assert key is used for both dynamic and static imports.
An alternative would be to remove the assert nesting in the object:
import("foo.wasm", { type: "webassembly" });However, that's not possible with the Worker API since it already uses an object with a type key as the second parameter. Which would make the APIs inconsistent.
- The integration of import attributes into various host environments.
- For example, in the Web Platform, how import attributes would be enabled when launching a worker (if that is supported in the initial version to be shipped on the Web) or included in a
<script>tag.
- For example, in the Web Platform, how import attributes would be enabled when launching a worker (if that is supported in the initial version to be shipped on the Web) or included in a
new Worker("foo.wasm", { type: "module", with: { type: "webassembly" } });Standardization here would consist of building consensus not just in TC39 but also in WHATWG HTML as well as the Node.js ESM effort and a general audit of semantic requirements across various host environments (#10, #24 and #25).
- 2019-12: The proposal, named module attributes is approved for Stage 1 (notes part 1, notes part 2, slides) to explore metadata for module imports, and to explore guarantees about modules with no code execution.
- 2020-06: Module attributes advances to Stage 2 (notes part 1, notes part 2, slides), with consensus based on the restriction that import attributes cannot be part of the cache key in the modules map. The proposed syntax is
import { x } from "./mod" with type: "json", something: "else";. - 2020-09: The proposal, renamed to import assertions, advances to Stage 3 (notes, slides). The rename better describes the agreed assert-only semantics, and the keyword changes from
withtoassert. However, the proposal relaxes the caching restriction so that HTML can still include the module type as part of the cache key, while still respecting the "spirit" of the proposal. - 2021-05–2022-02: The proposal, with the
import { x } from "./mod" assert { type: "json" };syntax, is implemented and shipped in Chrome, Node.js and Deno. They all support the JSON modules proposal. - 2023-01: Due to incommpatibility with the semantics needed by HTML for non-JavaScript modules, specifically regarding HTTP fetching and CSPs, the proposal is demoted back to Stage 2 (notes part 1, notes part 2, slides) to investigate a solution to the web platform's needs.
- 2023-03: The proposal is renamed to Import attributes and moves back to Stage 3 (TODO: notes, slides). The restriction on the cache key is completely removed, and the keyword changes back from
asserttowith:import { x } from "./mod" with { type: "json" };. For compatibility with existing implementations, theassertkeyword will still be supported until it's safe to remove it, if it will ever be.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent