Version Packages

[github-actions]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

wrangler@0.0.16

Patch Changes

• #364 3575892 Thanks @threepointone! - enhance: small tweaks to wrangler init

  • A slightly better package.json
  • A slightly better tsconfig.json
  • installing typescript as a dev dependency

• #380 aacd1c2 Thanks @GregBrimble! - fix: ensure pages routes are defined correctly

  In e151223 we introduced a bug where the RouteKey was now an array rather than a simple URL string. When it got stringified into the routing object these were invalid.
  E.g. [':page*', undefined] got stringified to ":page*," rather than ":page*".

  Fixes 🐛 BUG: 404 on Remix #379

• #329 27a1f3b Thanks @petebacondarwin! - ci: run PR jobs on both Ubuntu, MacOS and Windows

  • update .gitattributes to be consistent on Windows
  • update Prettier command to ignore unknown files
    Windows seems to be more brittle here.
  • tighten up eslint config
    Windows seems to be more brittle here as well.
  • use the matrix.os value in the cache key
    Previously we were using running.os but this appeared not to be working.

• #347 ede5b22 Thanks @threepointone! - fix: hide wrangler pages functions in the main help menu

  This hides wrangler pages functions in the main help menu, since it's only intended for internal usage right now. It still "works", so nothing changes in that regard. We'll bring this back when we have a broader story in wrangler for functions.

• #360 f590943 Thanks @threepointone! - fix: kv:key get

  The api for fetching a kv value, unlike every other cloudflare api, returns just the raw value as a string (as opposed to the FetchResult-style json). However, our fetch utility tries to convert every api response to json before parsing it further. This leads to bugs like kv:get errors out reading --preview key if key is not in JSON format #359. The fix is to special case for kv:key get.

  Fixes kv:get errors out reading --preview key if key is not in JSON format #359.

• #373 6e7baf2 Thanks @petebacondarwin! - fix: use the appropriate package manager when initializing a wrangler project

  Previously, when we initialized a project using wrangler init, we always used npm as the package manager.

  Now we check to see whether npm and yarn are actually installed, and also whether there is already a lock file in place before choosing which package manager to use.

  Fixes wrangler init should support yarn #353

• #363 0add2a6 Thanks @threepointone! - fix: support uppercase hotkeys in wrangler dev

  Just a quick fix to accept uppercase hotkeys during dev.

• #331 e151223 Thanks @petebacondarwin! - fix: generate valid URL route paths for pages on Windows

  Previously route paths were manipulated by file-system path utilities.
  On Windows this resulted in URLs that had backslashes, which are invalid for such URLs.

  Fixes Bug: generated routes on windows are incorrectly escaped (404) #51
  Closes Fix Pages Functions routing for windows path separator #235
  Closes Fix Pages Functions path routing on Windows #330
  Closes Fixed broken routing on Windows #327

• #338 e0d2f35 Thanks @threepointone! - feat: environments for Worker Sites

  This adds environments support for Workers Sites. Very simply, it uses a separate kv namespace that's indexed by the environment name. This PR also changes the name of the kv namespace generated to match wrangler 1's implementation.

• #329 e1d2198 Thanks @petebacondarwin! - test: support testing in CI on Windows

  • Don't rely on bash variables to configure tests
    The use of bash variables in the npm test script is not supported in Windows Powershell, causing CI on Windows to fail.
    These bash variables are used to override the API token and the Account ID.

    This change moves the control of mocking these two concepts into the test code, by adding mockAccountId() and mockApiToken() helpers.

    • The result is slightly more boilerplate in tests that need to avoid hitting the auth APIs.
    • But there are other tests that had to revert these environment variables. So the boilerplate is reduced there.

  • Sanitize command line for snapshot tests
    This change applies normalizeSlashes() and trimTimings() to command line outputs and error messages to avoid inconsistencies in snapshots.
    The benefit here is that authors do not need to keep adding them to all their snapshot tests.

  • Move all the helper functions into their own directory to keep the test directory cleaner.

• #380 aacd1c2 Thanks @GregBrimble! - refactor: clean up pages routing

• #343 cfd8ba5 Thanks @threepointone! - chore: update esbuild

  Update esbuild to 0.14.14. Also had to change import esbuild from "esbuild"; to import * as esbuild from "esbuild"; in dev.tsx.

• #371 85ceb84 Thanks @nrgnrg! - fix: pages advanced mode usage

  Previously in pages projects using advanced mode (a single _worker.js or --script-path file rather than a ./functions folder), calling pages dev would quit without an error and not launch miniflare.

  This change fixes that and enables pages dev to be used with pages projects in advanced mode.

• #383 969c887 Thanks @threepointone! - fix: remove redundant process.cwd() calls in wrangler init

  Followup from feat: wrangler init offers to create a starter worker #372 (comment), just removing some unnecessary calls to process.cwd()/path.join(), since they're already relative to where they're called from.

• #329 ac168f4 Thanks @petebacondarwin! - refactor: use helpers to manage npm commands

  This change speeds up tests and avoids us checking that npm did what it is supposed to do.

• #348 b8e3b01 Thanks @threepointone! - chore: replace node-fetch with undici

  There are several reasons to replace node-fetch with undici:

  • undici's fetch() implementation is set to become node's standard fetch() implementation, which means we can just remove the dependency in the future (or optionally load it depending on which version of node is being used)
  • node-fetch pollutes the global type space with a number of standard types
  • we already bundle undici via miniflare/pages, so this means our bundle size could ostensibly become smaller.

  This replaces node-fetch with undici.

  • All instances of import fetch from "node-fetch" are replaced with import {fetch} from "undici"
  • undici also comes with spec compliant forms of FormData and File, so we could also remove formdata-node in form_data.ts
  • All the global types that were injected by node-fetch are now imported from undici (as well as some mistaken ones from node:url)
  • NOTE: this also turns on skipLibCheck in tsconfig.json. Some dependencies oddly depend on browser globals like Request, Response (like @miniflare/core, jest-fetch-mock, etc), which now fail because node-fetch isn't injecting those globals anymore. So we enable skipLibCheck to bypass them. (I'd thought skipLibCheck completely ignores 'third party' types, but that's not true - it still uses the module graph to scan types. So we're still typesafe. We should enable strict sometime to avoid anys, but that's for later.)
  • The bundle size isn't smaller because we're bundling 2 different versions of undici, but we'll fix that by separately upping the version of undici that miniflare bundles.

• #357 41cfbc3 Thanks @threepointone! - chore: add eslint-plugin-import

  • This adds eslint-plugin-import to enforce ordering of imports, and configuration for the same in package.json.
  • I also run npm run check:lint -- --fix to apply the configured order in our whole codebase.
  • This also needs a setting in .vscode/settings.json to prevent spurious warnings inside vscode. You'll probably have to restart your IDE for this to take effect. (re: node:-prefixed modules not recognised as builtins import-js/eslint-plugin-import#2377 (comment))

  (I'd also like to enforce using node: prefixes for node builtin modules, but that can happen later. For now I manually added the prefixes wherever they were missing. It's not functionally any different, but imo it helps the visual grouping.)

• #372 05dbb0d Thanks @threepointone! - feat: wrangler init offers to create a starter worker

  We got feedback that wrangler init felt incomplete, because the immediate next thing folks need is a starter source file. So this adds another step to wrangler init where we offer to create that file for you.

  Fixes wrangler init should populate example code #355

• #384 8452485 Thanks @petebacondarwin! - refactor: use xxhash-wasm for better compatibility with Windows

  The previous xxhash package we were using required a build step, which relied upon tooling that was not always available on Window.

  This version is a portable WASM package.

• #334 536c7e5 Thanks @threepointone! - feat: wasm support for local mode in wrangler dev

  This adds support for *.wasm modules into local mode for wrangler dev.

  In 'edge' mode, we create a javascript bundle, but wasm modules are uploaded to the preview server directly when making the worker definition form upload. However, in 'local' mode, we need to have the actual modules available to the bundle. So we copy the files over to the bundle path. We also pass appropriate --modules-rule directive to miniflare.

  I also added a sample wasm app to use for testing, created from a default workers-rs project.

  Fixes wasm/textfile support for --local mode in dev #299

• #329 b8a3e78 Thanks @petebacondarwin! - ci: use npm ci and do not cache workspace packages in node_modules

  Previously we were caching all the node_modules files in the CI jobs and then running npm install. While this resulted in slightly improved install times on Ubuntu, it breaks on Windows because the npm workspace setup adds symlinks into node_modules, which the Github cache action cannot cope with.

  This change removes the node_modules caches (saving some time by not needing to restore them) and replaces npm install with npm ci.

  The npm ci command is actually designed to be used in CI jobs as it only installs the exact versions specified in the package-lock.json file, guaranteeing that for any commit we always have exactly the same CI job run, deterministically.

  It turns out that, on Ubuntu, using npm ci makes very little difference to the installation time (~30 secs), especially if there is no node_modules there in the first place.

  Unfortunately, MacOS is slower (~1 min), and Windows even worse (~2 mins)! But it is worth this longer CI run to be sure we have things working on all OSes.