Packages

webpack

5.112.2 · 23 Oct 2026MIT76.9M / weekRepositorynpmAtom feed

Your version

5.107.0

20 May 2026Where it was published

5.113.0 is the newest release of 5.x.

In short

1 removed14 new25 fixed

  • Removed

    Drop the __webpack_require__, __webpack_require__.d, and __webpack_require__.o runtime helpers from library: { type: "module" } bundles when the on-demand exports source they were emitted for ends up dropped (e.g. a single concatenated entry without an IIFE).

Release note

Minor Changes

  • Add module.generator.javascript.anonymousDefaultExportName option to control whether webpack sets .name to "default" for anonymous default export functions and classes per ES spec. Defaults to true for applications and false for libraries (when output.library is set) to avoid unnecessary bundle size overhead. Also extract anonymous default export .name fix-up into a shared runtime helper (__webpack_require__.dn), replacing repeated inline Object.defineProperty / Object.getOwnPropertyDescriptor calls with a single short call per module to reduce output size.

  • Support module concatenation (scope hoisting) for CSS modules with text, css-style-sheet, style, and link export types

  • The generator.exportsConvention function form for CSS modules now accepts string in addition to string.

  • Add linkInsert hook to CssLoadingRuntimeModule.getCompilationHooks(compilation) so plugin developers can control where stylesheet <link> elements are inserted into the document.

  • Add CssModulesPlugin.getCompilationHooks(compilation).orderModules hook.

  • Add a pure parser option for css/module and css/auto types matching postcss-modules-local-by-default's pure mode: every selector must contain at least one local class or id, otherwise webpack emits a build error.

  • Support CSS Modules @value identifiers as @import URLs and inside url() functions, e.g. @value path: "./other.css"; @import path; and @value bg: "./image.png"; .a { background: url(bg); }

  • Add experimental TypeScript support via experiments.typescript: true (auto-enabled by experiments.futureDefaults). Uses Node.js's built-in module.stripTypeScriptTypes (Node.js >= 22.6 with the stable mode: "strip" API, including Node.js 26) to transform .ts, .cts, .mts, data:text/typescript, and data:application/typescript modules — no type checking, only erasable TypeScript (types, generics, import type, casts). .tsx/JSX and non-erasable syntax (enum, namespace, parameter-property constructors, decorator metadata) are NOT supported; use a TSX-capable loader (e.g. ts-loader, swc-loader) for those.

  • Added an experiments.html flag that reserves the html module type for the first-class HTML entry-point support.

  • Preserve defer / source import phase keywords on external dependencies in ESM output, the same way import attributes are preserved.

  • Support the #__NO_SIDE_EFFECTS__ annotation to mark functions as pure for better tree-shaking.

  • Add module.generator.html.extract for HTML modules and the matching output.htmlFilename / output.htmlChunkFilename filename templates (defaults derived from output.filename / output.chunkFilename with .js swapped for .html, mirroring the CSS pipeline). When extraction is on, the parsed and URL-rewritten HTML is emitted as a standalone .html output file alongside the module's JavaScript export.

  • Add "module-sync" to default conditionNames for resolver defaults to align with Node.js, which exposes the module-sync community condition for synchronously-loadable ESM.

Patch Changes

  • Fix CSS modules composes so composes: foo from "./self.module.css" from inside self.module.css no longer creates a duplicate module instance. Fix CSS modules composes parsing so local() and global() function wrappers are tracked per class name. Fix CSS modules composes: ... from "<file>" so the composed files load in an order consistent with every rule's local composes order, instead of source first-appearance order.

  • Avoid emitting the __webpack_require__ runtime in CSS bundles when all imported CSS modules were concatenated into the same scope.

  • Recompute the CSS chunk's [contenthash] and the rendered CSS bytes when an asset referenced by url()/src()/string in CSS changes its hashed filename.

  • Embed an inline sourceMappingURL data URI inside the CSS when the parser.exportType option are text, style, or css-style-sheet. Also merge @imported CSS at build time for text and css-style-sheet exportTypes so the bundle ships a single accurate inline source map covering every contributing file. Map each generated CSS-module class export line in the JS bundle back to its selector position in the original CSS file (e.g. btn: "..." → .btn { ... }).

  • Fix CSS modules deduplication so a .module.<ext> file imported both directly (JS) and via icss (composes from / :import) becomes a single module instance.

  • Preserve @charset at-rule when CSS modules use exportType: "text".

  • Resolve [hash]/[fullhash] placeholders in output.publicPath when generating url() references for experiments.css.

  • Fix HMR for concatenated CSS modules with style exportType by using stable per-module identifiers for injected style elements and tracking inner module IDs of concatenated modules in HMR records

  • Fix CSS Modules @value resolution when the same local name is imported from multiple modules.

  • Fix typeof ns.default / ns.default instanceof X on a static import defer * as ns from "./mod" for default-only and default-with-named external modules under optimization.concatenateModules. The concatenated-module rewrite was collapsing ns.default to the deferred-namespace proxy itself instead of routing through the optimized .a getter (which lazily evaluates the module and returns its default value), so typeof ns.default observed "object" (the proxy) rather than the type of the default. The dynamic exportsType already used .a correctly; default-only and default-with-named now match.

  • Make import defer * as ns more spec-compliant: ns.x = value no longer triggers module evaluation (per the TC39 import-defer [[Set]] algorithm), and the deferred namespace is now a distinct object from the eager namespace, with the same Deferred Module Namespace Exotic Object shared across defer-import call sites for the same module.

  • Fixed spec deviations in the deferred namespace object returned by __webpack_require__.z (import defer * as ns / import.defer(...)).

  • Removed

    Drop the __webpack_require__, __webpack_require__.d, and __webpack_require__.o runtime helpers from library: { type: "module" } bundles when the on-demand exports source they were emitted for ends up dropped (e.g. a single concatenated entry without an IIFE).

  • Resolve the static specifier of a dynamic import() whose argument is a side-effect-free SequenceExpression, e.g. import((1, 0, "./mod.js")) is now treated the same as import("./mod.js") instead of being rejected as unresolvable.

  • Stable shared module ids and runtime-chunk emission order.

  • Fix snapshot validity check for context dependencies in watch mode by treating watchpack's existence-only entries ({}) as cache misses.

  • Support no-expression template literals in computed member access (e.g. import.meta[`url`]).

  • Improve tree-shaking in isPure: handle more expression types (ArrayExpression, ObjectExpression, NewExpression, ChainExpression, UnaryExpression (safe operators), MetaProperty, TaggedTemplateExpression, BinaryExpression (strict equality)), prevent /*#__PURE__*/ comments from leaking across ObjectExpression properties, and detect PURE comments inside TemplateLiteral interpolations.

  • Reject new import.defer(...) and new import.source(...) as a parse-time SyntaxError, matching the spec — ImportCall is a CallExpression and is not a valid operand of new. Parenthesized forms (new (import.defer(...))) remain valid and continue to throw TypeError at runtime as before.

  • Escape # characters that appear inside a path-shaped request's directory portion before passing the request to the resolver, so projects located in directories like /home/user/proj#1 (and tools like webpack-dev-server that build entry requests with query strings) resolve correctly. The escape only kicks in when the request contains both a # in the path portion and a ? query string — paths without a query keep their existing semantics.

  • Silence unhandled rejection from the prefetch trigger when chunk loading fails. The ensureChunkHandlers.prefetch runtime created Promise.all(promises).then(...) whose result is discarded by __webpack_require__.e. If chunk loading rejected (e.g. chunkLoadTimeout), that dangling chain produced an unhandled rejection. Prefetch is best-effort, so a no-op rejection handler is now attached.

  • Align require() of an ES module with Node.js's require(esm) "module.exports" named-export convention. When CommonJS require() resolves to an ES module that exports a binding with the literal string name "module.exports" (e.g. export { value as "module.exports" }), require() now returns the value of that export instead of the module's namespace object — matching Node.js v22.12+/v23+ behavior and easing migration of dual ESM/CJS libraries that rely on module.exports = …. The unwrap applies to plain require(), require().foo, calls (require()(…)), destructuring, and to CJS wrappers like module.exports = require(esm) / exports.x = require(esm).

  • Remove outdated @types/eslint-scope package from dependencies.

  • Fix export * resolution when a star-reexported module re-exports a name back to the importer cyclically. Previously, in a graph where a does export * from "./b"; export * from "./c"; and b does export { foo } from "./a"; while c provides the actual foo binding, webpack hoisted foo from b into a's namespace without per-name cycle detection — emitting a getter chain (a.foo → b.foo → a.foo) that threw "Maximum call stack size exceeded" at runtime. The TC39 ResolveExport algorithm requires the cyclic branch to return null and the star loop to fall through to the non-cyclic source.

  • Preserve using declaration initializers when the inner graph optimization is enabled.

  • Fixed typescript types.

  • Bump webpack-sources to ^3.4.1 and feed asset bytes into hashes via the new Source.prototype.buffers() API. For large ConcatSource/ReplaceSource outputs this avoids the intermediate Buffer.concat that source.buffer() performs, removing a peak-memory spike equal to the source's total size on each hashed asset (AssetGenerator.getFullContentHash, CssIcssExportDependency content hashing, and RealContentHashPlugin). A small benchmark on a 64 MiB ConcatSource shows ~64 MiB lower peak external memory and ~45% faster hashing.

A whole project to check? Paste its package.json in the dependency check: every deprecated, outdated or vulnerable dependency, with what to upgrade to. Free, no account.