fix: show type on invalid semver error

[tjenkinson]

Based on this comment from @wraithgar

If we want to revert this, it needs to be done in a way that also sets up some sort of linting or other testing so that "does not use x internals from node" is part of the contract, which it wasn't before.

This configures some eslint plugins to prevent importing node builtins or dev dependencies.

It also removes the dependency on node util and updates the error message slightly.

References

Fixes #554

[wraithgar]

Is there really no eslint rule for this? I find that amazing if so.

[tjenkinson]

Hmm actually didn't think about that. https://github.com/import-js/eslint-plugin-import/blob/main/docs/rules/no-extraneous-dependencies.md looks like it may work will give it a go

[wraithgar]

Is there something special about rollup or is it a coincidence that it also only supports the same subset of node internals that webpack does?

[wraithgar]

https://github.com/import-js/eslint-plugin-import/blob/main/docs/rules/no-nodejs-modules.md Is probably our rule. You'll need to override it in tests and map.js.

eslint rules can be extended by adding a .eslintrc.local.js file.

[tjenkinson]

this may not be required. checking

[wraithgar]

It's already in the base ruleset https://github.com/npm/eslint-config/blob/9e87b58d9e049004c802239cab1fabbca67caf74/lib/index.js#L213

[tjenkinson]

Ah yep. I think this config is useful though? The default would allow importing from devDependencies

[wraithgar]

The file is already imported as a plugin, so the extra dependency isn't needed. I think a single new rule is all we need here with an allow of ./test and map.js.

[wraithgar]

cf https://github.com/npm/cli/blob/3d5bbcc86c484ef0b6c04ee0236c1072d36915d6/.eslintrc.local.json#L3 or https://github.com/npm/cacache/blob/de12400af2ab452f1e75a15edaf7ea51b294667b/.eslintrc.local.js#L4

[wraithgar]

Then this needs to be addressed in template-oss, that's not a rule that we'd want to only apply to a single one of our over 80 repos.

[tjenkinson]

Opened npm/template-oss#316

[tjenkinson]

@wraithgar so given this doesn't follow the standard structure npm/eslint-config#67 doesn't have any effect here, so is the custom config in this PR good to go?

[wraithgar]

Yeah we'll need a custom config but this one looks to be a bit different in approach than the one we landed on in eslint-config. We should pattern it after the one there but instead of lib and bin it should be bin, classes, functions, internal, and ranges (and now we begin to see why we standardized on lib).

The way this PR currently does it is by completely bypassing the no-extraneous-dependencies rule for the test files, which is a regression.

[tjenkinson]

Yep. done!

[tjenkinson]

Is there something special about rollup or is it a coincidence that it also only supports the same subset of node internals that webpack does?

By default rollup doesn't understand any node_module dependencies. There is a node-resolve plugin that adds that support for those to be found.

It also doesn't pollyfill any node builtins out of the box.

[ljharb]

An alternative would be to use the "browser" field to specify an alternative for util.inspect outside of node, which could be typeof, but then node usage would still get the nice util.inspect experience.

[tjenkinson]

Yeh I'm not sure if the benefit of inspect is really worth it for this case though?