Implement "exports" proposal

[guybedford]

This implements the package.json "exports" proposal (https://github.com/jkrems/proposal-pkg-exports/) with the following features:

• When "exports" are defined for a package ({ "exports": { "./x": "./x.js" }), resolving that package import 'pkg/x', will resolve the exports path (/path/to/pkg/x.js).
• When a subpath does not match "exports", an error is thrown (see test case).
• Support for "exports": false to indicate a package has no exports at all.
• Support for the "." export as an alias of the main.
• Support for folder exports ({ "./folder/": "./other-folder/" })

Support for "exports" when resolving from CommonJS is not currently implemented.

Checklist

• make -j4 test (UNIX), or vcbuild test (Windows) passes
• tests and/or benchmarks are included
• documentation is changed or added
• commit message follows commit guidelines

[MylesBorins]

Amazing work!

I think that we should remove "Support for the "." export as an alias of the main."... this recreates the issue with dual mode.

[devsnek]

this repo is two months behind master, you may want to re-open this in nodejs/node

[targos]

This repo is not behind master. It's updated once per day by a Jenkins job

[guybedford]

I've also added some documentation and included a folder mapping feature test.

[ljharb]

I think this should just be “any non-nullish value is passed through Object.keys to get the export list; any nullish value disables the feature”, because then “false” is just a convention instead of a magic special value.

[hybrist]

So:

1. null and undefined: Same as { './': './' }.
2. Anything else: Normalized as an object (e.g. Object.assign({}, pkgExports)).

Makes sense!

[ljharb]

I’d use { ...pkgExports } but yes, exactly!

[hybrist]

I think guiding people towards a convention here may still have value. How about adding the following sentence:

This is just a convention that works because false, just like {}, has no iterable own properties.

[ljharb]

Agreed and that text sounds good to me.

[hybrist]

Done!

[hybrist]

Support for "." is gone as is the special treatment of false.

[hybrist]

Afaict this now has all the pieces for the upstream PR, with CJS support and general polish / edge case hunting being follow-ups while it's experimental.

[hybrist]

I turned this into a PR against core: nodejs/node#28568

Should we close this PR?

[ljharb]

@jkrems i thought we were going to remove Support for the "." export as an alias of the main.

[hybrist]

@ljharb May be a misunderstanding: What I removed (and there's a test for it not working) is setting the un-postfixed package name export via the exports field. The test to verify that adding a . exports key doesn't work is here: https://github.com/nodejs/node/pull/28568/files#diff-a512c0b2a7207547e7e8e8fb17bd1aa5R9

[hybrist]

Or rather the actual test is here: https://github.com/nodejs/node/pull/28568/files#diff-3fa3ea8413322c7b78e358f1a6f6e613R26

[Trott]

Should this be closed at this point?