perf: rewrite npmlog

[H4ad]

This is a very risky PR since it could break a lot of things but I think it is worthy to have.

The loading graph before:

The loading graph after:

The loading time using hyperfine before:

$ hyperfine --warmup 3 "node ./lib/log.js"
Benchmark 1: node ./lib/log.js
  Time (mean ± σ):     114.9 ms ±   1.7 ms    [User: 101.4 ms, System: 37.9 ms]
  Range (min … max):   111.9 ms … 117.6 ms    25 runs

After:

$ hyperfine --warmup 3 "node ./lib/log.js"
Benchmark 1: node ./lib/log.js
  Time (mean ± σ):     104.1 ms ±   2.0 ms    [User: 91.6 ms, System: 34.4 ms]
  Range (min … max):    98.1 ms … 108.5 ms    29 runs

I also added another method called trackerRemoveAllListeners that can be used to never load the are-we-there-yet at https://github.com/npm/cli/blob/latest/lib/utils/display.js#L114

I think we should add more tests to gauge and also for tracker that didn't include mocked versions since I didn't even declare tracker and gauge during the tests and it passed all the tests because they are mocked.

[lukekarrys]

This is definitely a good idea! However I have a WIP PR that will remove npmlog and are-we-there-yet and gauge. It will add one new dependency proggy. Currently tests are not passing in that PR but as it gets close I would be interested in you taking a look at it for performance before it lands.

[H4ad]

Send me the branch, I can take a look and see if it will worth continuing the refactoring (by looking at the load time)

[lukekarrys]

npm/cli#7339

[lukekarrys]

The branch might even be completely broken at the moment. It's a WIP that I'm hoping to finish next week.

[H4ad]

Since you are working to remove npmlog, and the alternative looks faster, should I close this PR? I think you will deprecate this module, right?

[wraithgar]

Yes we will deprecate this module. PLEASE be involved as much as you want in the rewrite efforts. This is our overview issue where we're tracking and discussing the efforts: npm/statusboard#810