Consider @babel/core being bundled to improve CLI startup times.

[stefanpenner]

Feature Request

A cold require of @babel/core for the first time, can take 400ms->500ms even on a fast machine.
This is quite high, especially for smaller projects, as their complete babel compilation time may be less than that.

Idea

Thoughts on mimic'ing yarn, and bundling @babel/core with webpack? My experiments suggest this can lead to a nice boost, with relatively minimal cost.

// after a clean install
load babel-core.bundle.js: 79.151ms
load babel-core: 449.058ms

// once the disk is warm
load babel-core.bundle.js: 79.151ms
load babel-core: 170.081ms

early POC: https://github.com/stefanpenner/babel-core.bundle

Some issues

It does appear, additional configuration/cleanup will be required to ensure various internals resolve correctly. As doing so via my naive approach appears to break some babel plugins. This may or may not be a show stopper.

For example:

   Error: Cannot find module '/Users/spenner/src/babel/broccoli-babel-transpiler/node_modules/@babel/plugin-transform-strict-mode/lib/index.js'
      at r (node_modules/@babel/core/dist/babel-core.bundle.js:1:443324)
      at A (node_modules/@babel/core/dist/babel-core.bundle.js:1721:50205)
      at t.loadPlugin (node_modules/@babel/core/dist/babel-core.bundle.js:1721:48335)
      at m (node_modules/@babel/core/dist/babel-core.bundle.js:1:445869)
      at t.map (node_modules/@babel/core/dist/babel-core.bundle.js:1:445008)
      at Array.map (<anonymous>)
      at h (node_modules/@babel/core/dist/babel-core.bundle.js:1:444997)
      at d (node_modules/@babel/core/dist/babel-core.bundle.js:1:444947)
      at t (node_modules/@babel/core/dist/babel-core.bundle.js:1:444350)
      at /Users/spenner/src/babel/broccoli-babel-transpiler/node_modules/@babel/core/dist/babel-core.bundle.js:1:89995
      at plugins.n (node_modules/@babel/core/dist/babel-core.bundle.js:1:443820)
      at k (node_modules/@babel/core/dist/babel-core.bundle.js:1:451988)
      at O (node_modules/@babel/core/dist/babel-core.bundle.js:1:451411)
      at t.buildRootChain (node_modules/@babel/core/dist/babel-core.bundle.js:1:447284)
      at f (node_modules/@babel/core/dist/babel-core.bundle.js:1:459282)
      at t.default (node_modules/@babel/core/dist/babel-core.bundle.js:1721:51447)
      at a (node_modules/@babel/core/dist/babel-core.bundle.js:1721:58326)
      at Object.o (node_modules/@babel/core/dist/babel-core.bundle.js:1721:58138)

Question

So if this startup perf boost via bundling sounds appealing, I'll gladly explore/debug the above issues further.

Just let me know.

[nicolo-ribaudo]

I think that we should bundle each package individually. This will bring big benefits to @babel/core, @babe/traverse and @babel/types, since they consists in multiple packages.

Maybe using rollup instead of webpack would be better, since we don't need most of webpack's features ans our files are es modules? We already have the logic to build @babel/parser using rollup, we just need to expand it to other packages.

[stefanpenner]

@nicolo-ribaudo ya more granular bundles should achieve the same perf win, with less or no downside.