Rollup v2.0.0

[lukastaegert]

This PR contains:

• bugfix
• feature
• refactor
• documentation
• other

Are tests included?

• yes (bugfixes and features will not be merged without tests)
• no

Breaking Changes?

• yes (breaking changes will not be merged unless absolutely necessary)
• no

List any relevant issue numbers:
See list of contained PRs at the bottom

Description

This is the release branch for Rollup@2.
See https://github.com/rollup/rollup/projects/2 for the current state of the new release.

At the moment, the following changes are contained in this branch:

Breaking Changes

• Rollup now requires at least Node 10 to run, or a sufficiently modern browser ([v2.0.0] Update minimum required Node version to 10 #3346)

• The file structure of Rollup's ESM builds has changed:

  • The main ESM entry point is now at rollup/dist/es/rollup.js instead of rollup/dist/rollup.es.js
  • The ESM browser build is at rollup/dist/es/rollup.browser.js instead of rollup/dist/rollup.browser.es.js

  In general, the ESM builds now follow the same naming scheme as the CJS builds but are located in the rollup/dist/es subfolder instead of rollup/dist ([v2.0.0] Bundle acorn, allow node ESM import fo Rollup, update dependencies #3391)

• The "watch.chokidar" option no longer accepts a boolean value but only an object of parameters that is passed to the bundled Chokidar instance. Chokidar installations by the user will be ignored in favour of the bundled instance ([v2.0.0] Bundle chokidar #3331)

• Modules that are completely tree-shaken will no longer be listed as part of any chunks in generateBundle

• The experimentalOptimizeChunks and chunkGroupingSize options have been removed

• acorn plugins can only be used if they accept a passed-in acorn instance instead of importing it themselves. See Request: give token types and etc to plugins acornjs/acorn#870 (comment) for what needs to be done to make plugins compatible that do not support this yet ([v2.0.0] Bundle acorn, allow node ESM import fo Rollup, update dependencies #3391)

• Emitted chunks now have the TypeScript type UInt8Array instead of Buffer. A Buffer can still be used, though ([v2.0.0] Remove @types dependencies #3395)

• The TypeScript types no longer use ESTree types for AST nodes but a very generic type that does not contain information specific to certain node types ([v2.0.0] Remove @types dependencies #3395)

• The signature of the writeBundle plugin hook has been changed to match generateBundle: The bundle object is now passed as second parameter instead of first and the first parameter is the output options ([v2.0.0] Change writeBundle signature to match generateBundle #3361)

• The following plugin hooks have been removed:

  • ongenerate: use generateBundle instead
  • onwrite: use writeBundle instead
  • transformBundle: use renderChunk instead
  • transformChunk: use renderChunk instead

• You can no longer access this.watcher on the plugin context.

• The transform hook can no longer return dependencies.

• The treeshake.pureExternalModules option will now show a deprecation warning when used: use treeshake.moduleSideEffects: 'no-external' instead

• Using import.meta.ROLLUP_ASSET_URL_<..> and import.meta.ROLLUP_CHUNK_URL_<..> in code will now show warnings: use import.meta.ROLLUP_FILE_URL_<..> instead

• The resolveAssetUrl hook will now show a deprecation warning when used: use resolveFileUrl instead

• The following plugin context functions will show warnings when used:

  • this.emitAsset: use this.emitFile
  • this.emitChunk: use this.emitFile
  • this.getAssetFileName: use this.getFileName
  • this.getChunkFileName: use this.getFileName
  • this.isExternal: use this.resolve
  • this.resolveId: use this.resolve

• Directly adding properties to the bundle object in the generateBundle is deprecated will show a warning (removing properties is allowed, though): Use this.emitFile

• Accessing chunk.isAsset on the bundle is deprecated: Use chunk.type === 'asset' instead

• The error code for a missing name property when targeting UMD has been changed to MISSING_NAME_OPTION_FOR_IIFE_EXPORT to emphasize this is needed for the IIFE part of UMD (more explained error code for name-less umd bundle  #3393)

Features

• Rollup now bundles Chokidar for a better watch experience ([v2.0.0] Bundle chokidar #3331)
• Rollup now bundles acorn again, removing its only external dependency ([v2.0.0] Bundle acorn, allow node ESM import fo Rollup, update dependencies #3391)
• Do not consider empty imports from side-effect-free modules for chunking and hoist side-effect imports if necessary ([v2.0.0] Avoid empty imports from side-effect-free chunks #3369)
• Rollup can now be imported as an ES module in Node via import {rollup} from 'rollup'. Note that this relies on Node's experimental conditional package exports feature and is therefore itself experimental ([v2.0.0] Bundle acorn, allow node ESM import fo Rollup, update dependencies #3391)
• systemjs can be used as format alias for system ([v2.0.0] Rename "esm" to "es" in docs and errors #3381)

Bug Fixes

• Unknown output options now trigger a warning when using the JavaScript API ([v2.0.0] Remove active deprecations #3352)
• Rollup will no longer introduce Node types into TypeScript projects that do not use them ([v2.0.0] Remove @types dependencies #3395)
• Generate correct sourcemaps when tree-shaking occurs in a multi-file bundle (Update magic-string to fix sourcemaps #3423)

Pull Requests

• #3331: Bundle Chokidar (@lukastaegert)
• #3343: Remove experimentalOptimizeChunks (@lukastaegert)
• #3346: Update minimum required Node version to 10 (@lukastaegert)
• #3352: Remove active deprecations (@lukastaegert)
• #3361: Change writeBundle signature to match generateBundle (@lukastaegert)
• #3369: Avoid empty imports from side-effect-free chunks (@lukastaegert)
• #3381: Rename esm to es everywhere, add systemjs alias (@lukastaegert)
• #3391: Bundle acorn, allow importing Rollup as Node ES module, update dependencies (@lukastaegert)
• #3393: Better error code for name-less umd bundle (@rail44)
• #3395: Remove @types dependencies (@lukastaegert)
• #3423: Update magic-string and fix sourcemaps (@lukastaegert)

[codecov]

Codecov Report

Merging #3347 into master will increase coverage by 1.71%.
The diff coverage is 98.28%.

@@            Coverage Diff             @@
##           master    #3347      +/-   ##
==========================================
+ Coverage   93.29%   95.00%   +1.71%     
==========================================
  Files         172      171       -1     
  Lines        6114     5831     -283     
  Branches     1823     1722     -101     
==========================================
- Hits         5704     5540     -164     
+ Misses        218      157      -61     
+ Partials      192      134      -58     

Impacted Files	Coverage Δ
src/Graph.ts	97.35% <0.00%> (+4.40%)	⬆️
src/ModuleLoader.ts	98.52% <0.00%> (-0.01%)	⬇️
browser/crypto.ts
src/utils/chunkAssignment.ts	100.00% <0.00%> (ø)
src/watch/fileWatcher.ts	96.42% <0.00%> (ø)
src/watch/watch-proxy.ts	100.00% <0.00%> (ø)
src/watch/fsevents-importer.ts	40.00% <0.00%> (ø)
src/utils/parseOptions.ts	100.00% <0.00%> (ø)
src/utils/mergeOptions.ts	100.00% <0.00%> (+7.95%)	⬆️
cli/run/batchWarnings.ts	97.63% <0.00%> (+2.36%)	⬆️
... and 2 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f458cbf...4610f4d. Read the comment docs.

[rollup-bot]

Thank you for your contribution! ❤️

You can try out this pull request locally by installing Rollup via

npm install rollup/rollup#release-2.0.0

or load it into the REPL:
https://rollupjs.org/repl/?circleci=9771

[Andarist]

q: wasnt "es" format supposed to be renamed to "esm"? or you have decided not to proceed with this one?

[lukastaegert]

Yes it was. The problem is that this caused an inconsistency with plugins that still receive "es" as format specifier, which was confusing to say the least. Asking around for opinions, I agree with the following reasoning:

• CJS modules are called cjs
• AMD modules are called amd
• SystemJS modules are called system
• Why should ES modules be called esm and not es?

So it is much more consistent with the other format specifiers. Nevertheless this is not a breaking change: esm will continue to work as an alias and is not marked for deprecation, and module is supported as well.

[kzc]

I'm not convinced that bundling acorn and magic-string is a good idea. If there's a bug in one of those packages (as is currently the case with magic-string) there's no way for a user to supply an alternate package with a fix via yarn resolutions or other methods. The users would have to wait for the upstream package fix followed by a new rollup release.

[lukastaegert]

I see your point and we might revert this policy at some point. At the moment, bundling provides some slight size and parsing time advantages and allows for consistent behaviour between Node and browser builds.

I usually try to be quick on updating merged dependencies, so unless you want to use a Github build this should not be too big an issue. Let's see how this pans out.

[NotWoods]

Can you update compilerOptions.target in the tsconfig.json file to "es2018"? That'll tell Typescript to any syntax that can run in Node 10 and save us a few bytes.

[lukastaegert]

Thanks for checking, this has actually already been done in one of the PRs included here:

rollup/tsconfig.json

Line 12 in 3f1052b

"target": "es2018"

[wessberg]

Hey Lukas. Great work, and congratulations on the v2 milestone :-)

Please let me know if you'd like me to create a new issue in relation to the following question. I'd like to understand if you have any plans for bringing back chunk grouping? For production applications built with Rollup with hundreds or thousands of chunks due to heavy use of code splitting, you might end up with many chunks weighing less than 1 kilobyte. My team is relying heavily on Rollup, but we're currently serving so many chunks to our users, most of which are less than or around 1kb in size, that it hurts performance. I've been following progress of the experimentalChunkOptimization flag with the hopes that it might help us due to the chunk grouping feature. However, I've also noticed that the feature didn't really work (hence the experimental nature of the feature, and I'm guessing that may be the reason why you removed it), but I think it would be extremely valuable, at least to me and my team, and generally for those who rely on Rollup for building applications in a progressive and performance-sensitive way.

[lukastaegert]

So to give you some idea, these are the topics I want to work on next:

• Reduce the number of facade chunks. Facade chunks are reexport chunks that are generated whenever a chunk that contains an entry point would also contain additional exports. The plan is to add an option to configure Rollup more for web apps, which would get rid of facades for static entries. And later add an option to allow for "tainted" namespaces in dynamic imports, i.e. namespaces that contain additional exports. This would get rid of many facades for dynamic imports.
• Fix hashing by completely reworking how it is done.
• Refactor chunking so that it is no longer part of the "build" phase but moved to the "output generation" phase. This is a profound refactoring but would be a necessary step in order to implement a better "optimizeChunks". At that point, I would also consider adding a plugin API to allow plugins to directly control the chunking procedure. Depending how it is done, "optimizeChunks" could also become a plugin then.

However, it would take quite a bit of time to get there. For the time being, I would also have a look at manualChunks, which can basically be used to hand merge chunks. If all static entry points are always loaded together, one easy optimization could be to make all of them manual chunks. As a rule of thumb, an entry that is always loaded initially can always be made a manual chunk without risk. Manual chunks work like a flood fill paint, adding all statically imported modules to a chunk, unless they are already part of another manual chunk.