Add fix for Windows Subsystem for Linux

[manuth]

This PR will add support for Windows Subsystem for Linux.

Windows Subsystem for Linux seems to have quite a weird treatment with file and directory-names.
While directories are case-sensitive (directory ./tests exists while ./TeSts doesn't), files don't seem to be (both reSOLVE.js and resolve.js exist).

This PR adds WSL-support by checking whether both directories and files are case-sensitive before setting CASE_SENSITIVE_FS.

[ljharb]

Thanks! Can you provide a test that would fail without this fix?

[manuth]

https://github.com/benmosher/eslint-plugin-import/blob/3f46ccfe203fa5aad5767699df339a9b2f0294e3/tests/src/core/resolve.js#L165-L186

These tests fail when running npm tests on WSL without this fix being applied.
Is there some way to put this into a unit-test?

[coveralls]

Coverage increased (+2.6%) to 97.741% when pulling 8118170 on manuth:wsl-fix into 62b554b on benmosher:master.

[coveralls]

Coverage remained the same at 97.738% when pulling 4619037 on manuth:wsl-fix into 3f46ccf on benmosher:master.

[coveralls]

Coverage remained the same at 97.738% when pulling 4619037 on manuth:wsl-fix into 3f46ccf on benmosher:master.

[coveralls]

Coverage remained the same at 97.738% when pulling 4619037 on manuth:wsl-fix into 3f46ccf on benmosher:master.

[coveralls]

Coverage remained the same at 97.738% when pulling 4619037 on manuth:wsl-fix into 3f46ccf on benmosher:master.

[ljharb]

ahhh so the existing tests do cover it, but we're not actually running tests on WSL - just on normal windows.

If you already know how to add that to appveyor or travis, then that would be amazing, but if not, that is also OK and we can punt it to later.

[manuth]

Ouch... looks like apt just invoked a form which requires user-interaction... might take a while until all the jobs run into a timeout :/

[manuth]

@ljharb is it possible to run console.log(require("./utils/resolve").CASE_SENSITIVE_FS) on appveyor for test-purposes?

Though WSL works now, the npm-tests still fail which is very strange.
An error about the case-sensitive tests is reported even if they actually should be skipped (cause of CASE_SENSITIVE_FS being set to true)

[manuth]

@ljharb as expected, "CASE_SENSITIVE_FS" is true, but still the case sensitive tests aren't skipped.

Any idea what might be buggy here?
Probably a function binding problem?

[manuth]

@ljharb I can't help it - CASE_SENSITIVE_FS definitely is true and still the tests aren't skipped...
Anyways the WSL jobs are up and running now.

Any other idea on how to fix it?

[manuth]

@ljharb sorry for spaming you with notifications
I found out what's causing the problem.

When only running npm run tests-only, pretest isn't executed, which actually would symlink ./utils as module eslint-module-utils. The fact that pretest isn't executed causes npm to use eslint-module-utils from the npm-registry instead of ./utils.

Should I change the .travis.yml and appveyor.yml files to execute pretest

[ljharb]

Yes, that sounds like a great idea.

[manuth]

@ljharb Alright, made it!
The WSL-jobs now should pass.
After merging this PR I could create another one for this here, if you want:
https://github.com/benmosher/eslint-plugin-import/blob/3f46ccfe203fa5aad5767699df339a9b2f0294e3/appveyor.yml#L40-L42

After doing so, this project's fully Windows compatible (using WSL, at least)

[ljharb]

Yes, that would be amazing, thanks. I don't have any Windows machines to test, so Appveyor is the only validation this project really has.

[ljharb]

Looks like node 8 and 14 "Native" tests are still failing. I've rebased this (and squashed and added a changelog entry), so let's see what happens.

[manuth]

Looks like Babel is causing trouble
I'll try to update it and create a PR

[ljharb]

LGTM pending tests

[ljharb]

hmm, now all the native tests are failing - but on the coveralls step, all the tests are passing on node 10 and 12 (node 14 still fails with "Cannot find module 'C:\projects\eslint-plugin-import\node", and node 8 still fails with a ton of "Path must be a string. Received undefined" errors)

However, all those tests were already failing the same way on master, so this still seems worth landing for now, as-is.