Running rubocop without bundle exec significantly reduces run time (w/ benchmarks)

[mculp]

I read in a rubocop PR that running rubocop (and erblint) without bundle exec speeds up boot time by ~300ms for the individual who made the post. So, I removed bundle exec from my rubocop and erblint lefthook configs and tried it, and it indeed reduce run time consistently and significantly.

Config changes

I noticed that after I did this, the pre-commit script changed:

I guess I have two questions:

1. I'm wondering what exactly this is doing and why it changed after removing bundle exec from the config? I've noticed these files changing before, too. I don't have pnpm on my machine. The first section is just moved and I don't have pnpm on my machine. So it doesn't appear to actually change how anything is executed.

2. To make this safe for my team, who could be running a different setup, should I check that rubocop exists with type rubocop or something similar or is that something this script handles?

Generating a binstub and using bin/rubocop does not give the performance boost, either.

A note for anyone else interested, if you run rubocop --start-server or add --server to the rubocop command, it will start a rubocop server with its files already loaded and the second time you run rubocop, it can drop your runtime significantly more.

bundle exec rubocop --no-server ...

rubocop --no-server ...

bundle exec rubocop --server ... (second run with server running)

rubocop --server (second run with server running)

Note: I used git restore .git-hooks/ to remove the pre-commit and prepare-commit-msg scripts between these first four. The updated scripts seem to drop it a bit more. Seems consistent ~0.26-0.27s without the scripts and ~0.18-0.19s with. Maybe the extra time is Lefthook generating the new scripts? I don't see anything from the scripts that would change anything.

rubocop --server with updated scripts

[mrexox]

Hey! Very cool suggestion, I really like the boost using rubocop --server. This is really useful!

I want to answer a few questions:

1. Lefthook uses a hook template which finds the lefthook executable and runs it. Since lefthook is distributed as NPM package, Ruby Gem, simple binary, it can be missing in $PATH, so that's why we do bundle exec, pnpm, npx, and check node_modules. Adding pnpm was a new thing in last released version, and you might have updated lefthook recently, so this change appeared.
2. You can provide a fail_text with instructions to install rubocop globally :) Or you can create a more sophisticated solution providing not a command, but a script, which checks for rubocop and installs it is it is missing. This requires some "dances" with lefthook but IMHO providing a fail_text with s fixing command is just enough.