[Refactor] switch to an internal replacement for pkg-up and read-pkg-up

[mgwalker]

Inspired by #2046, I dug into what read-pkg-up provides and found it to be very simple, for this particular case. Its downstream dependencies build in some safety and options that aren't necessary when you already know you're working with a package.json, and in this library's case, when you know you don't want to "normalize" it. Since read-pkg-up drags several libraries along with it, at least two of which don't actually add any functionality to eslint-plugin-import, and its use here is pretty straightforward, it seemed moderately reasonable to me to reproduce the functionality internally.

[coveralls]

Coverage increased (+14.7%) to 84.198% when pulling e3a44f8 on mgwalker:remove-read-pkg-up into e871a9a on benmosher:master.

[ljharb]

I'm not super stoked on copy-pasting, but removing this dep is valuable.

[silverwind]

The replacement [email protected] is very out of date and contains security vulnerabilites to boot. I'd say drop the dependency completely and replace it with a small function:

const {dirname} = require('path');
const {accessSync} = require('fs');

function findUpSync(dir, file) {
  const path = join(dir, file);
  try {
    accessSync(path);
    return path;
  } catch {}
  const parent = dirname(dir);
  return parent === dir ? null : findSync(parent, file);
}

Use via findUpSync(filePath, 'package.json').

[mgwalker]

Ergh, sorry for the failing tests. I didn't realize that eslint-module-utils was a published package as well as a local package. I haven't had a chance to dig into it very far yet, but my working assumption is that something changed in npm's order of operations when a dependency is listed in both dependencies and devDependencies. In npm 5.x and 6.x, it seems npm install will fetch from the published package if available. Since that one doesn't include the updates in this PR, the tests fail. In later versions of npm, it uses the local version instead, and everything is fine.

My apologies if this is known issue and there's an established workflow for updating eslint-module-utils separately from eslit-plugin-import.

[ljharb]

There isn’t; you could make the changes in two PRs, and we could land and publish the utils one first?

altho i wouldn’t expect that to be needed - we use linklocal in this repo so everything should be using the latest local code.

[mgwalker]

Oh, well in that case, maybe just switching the dependencies to the local package will do the job. Pushed a new commit.

[mgwalker]

Also as you mentioned, the CVE got updated, so I guess closing this PR as moot would also make sense. 🤷🏻‍♂️

[ljharb]

Another alternative from a bit ago is #1017.

[ljharb]

went ahead and replaced pkg-up as well, and rebased. this should be good to go.

[codecov]

Codecov Report

Merging #2047 (9ccdcb7) into main (7c382f0) will increase coverage by 0.01%.
The diff coverage is 90.00%.

@@            Coverage Diff             @@
##             main    #2047      +/-   ##
==========================================
+ Coverage   84.02%   84.03%   +0.01%     
==========================================
  Files          94       96       +2     
  Lines        3017     3032      +15     
  Branches      898      899       +1     
==========================================
+ Hits         2535     2548      +13     
- Misses        482      484       +2     

Impacted Files	Coverage Δ
utils/readPkgUp.js	81.81% <81.81%> (ø)
src/core/packagePath.js	80.00% <100.00%> (ø)
src/rules/no-extraneous-dependencies.js	100.00% <100.00%> (ø)
src/rules/no-import-module-exports.js	100.00% <100.00%> (ø)
src/rules/no-relative-packages.js	95.00% <100.00%> (ø)
src/rules/no-unused-modules.js	97.43% <100.00%> (ø)
utils/pkgUp.js	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 7c382f0...9ccdcb7. Read the comment docs.