Bad entry points in `package.json`

[benmccann]

Expected Behavior

UMD build should be listed in browser. main is server entry-point

https://docs.npmjs.com/cli/v7/configuring-npm/package-json#main
https://docs.npmjs.com/cli/v7/configuring-npm/package-json#browser

Current Behavior

Vite is unable to load Chart.js

Possible Solution

Put UMD build in browser. Add "type": "module"

Steps to Reproduce

https://github.com/benmccann/vite-chartjs-bug

Context

Chart.js does not work in SvelteKit

Environment

• Chart.js version: 3.4.1

[kurkle]

That information is somewhat different compared to this:
https://nodejs.org/api/packages.html#packages_main

It suggests using conditional exports

Also no mention of browser field in there.

[benmccann]

I don't think the npm and Node docs conflict. They both say that main is the server entry point

I don't think we'd have to use conditional exports. That's for if you want to a different package to be loaded on the server when you do require vs import. That's more configurable, but also a little harder to maintain, so if we can get away without doing it that seems easier to me

Node.js is a server-side runtime, so not surprising that they don't mention the browser field as it wouldn't be applicable to their users

[kurkle]

vitejs/vite#4209 seems related

[benmccann]

vitejs/vite#4209 seems related

I don't think it is. That bug is related to CommonJS packages. Chart.js doesn't provide a CommonJS package

[kurkle]

Chart.js provides UMD package, which is the current main and is compatible with AMD and CommonJS. A missing type would make Node think its CommonJS and the UMD file provided would work for that.

What I worry about in changing the type to module and main to the esm build is that some module loaders might not work anymore (like requirejs). I worry, because I don't know and we don't have any tests for that.

[benmccann]

Thanks for the help in investigating this! After the various findings, I think this isn't a great description of what's happening, so I've filed a clean issue with a better explanation: #9436