Introducing Rubyfmt

@penelope_zone ✨Introducing✨ rubyfmt

@penelope_zone

@penelope_zone Updates

@penelope_zone This is gonna get real emotional

@penelope_zone Penelope She/Her Trans Woman

@penelope_zone Ruby Central Director

@penelope_zone No longer an RSpec maintainer

@penelope_zone Thanks to Jon, Myron, Andy, and David

@penelope_zone Rubyfmt is gonna be my open source focus for
the next while

@penelope_zone What is Rubyfmt?

@penelope_zone @penelope_zone

@penelope_zone

@penelope_zone x

@penelope_zone I can’t unsee this

@penelope_zone Ruby autoformatter

@penelope_zone

@penelope_zone No formatting related configuration options

@penelope_zone “Unix tool”

@penelope_zone stdin/file name

@penelope_zone stdout/in place

@penelope_zone Not a gem!

@penelope_zone Probably > 6 months until something you can use
daily

@penelope_zone Why are you doing this?

@penelope_zone Principle

@penelope_zone

@penelope_zone "TDD results in better code"

@penelope_zone This is another way of discussing tradeoffs

@penelope_zone My Principles

@penelope_zone Do one thing well Correctness beats simplicity Speed is
needed at any cost

@penelope_zone Do one thing well

@penelope_zone

@penelope_zone I u Rubocop team

@penelope_zone Formatting Metrics Linting Naming Security Style Bundler Gemspec Rails

@penelope_zone

@penelope_zone Formatting Metrics Linting Naming Security Style Bundler Gemspec Rails

@penelope_zone

@penelope_zone $%&%'()*%%*+,)$%&%)%('- )'%$*($,%&%'%(&,-$()+$- +-%)$&+'')-%',*&)$&&'(% )$)-)&$*(+$%'-)&%+)*$'$ &*%)'-%$'-'$(*-(--$,)-- %&*,$+&*$*,))$)**,()'(* ,)++$-'&*'*'-*$)(-$($(+ '(%))$$),-$$+*$*(,%*(%- -&$(**,+&+'(&,,+)'&$)')

@penelope_zone

@penelope_zone Rubyformat is literally a different kind of tool

@penelope_zone

@penelope_zone Doing one thing extremely well

@penelope_zone By definition configuration options imply doing multiple things

@penelope_zone Support every possible combination

@penelope_zone Support every possible combination ❌

@penelope_zone Support one consistent style ✅

@penelope_zone bars = if foo baz else qux end

@penelope_zone bars = “one” \ “two”

@penelope_zone Rubyfmt always indents two relative to the parent construct

@penelope_zone bars = if foo baz else qux end

@penelope_zone bars = “one” \ “two”

@penelope_zone This is a tradeoff

@penelope_zone Correctness beats simplicity

@penelope_zone How do you parse Ruby?

@penelope_zone

@penelope_zone The parser output is really simple to deal with

@penelope_zone $ srb tc -p parse-tree -e 'a' Send {
receiver = NULL method = <U a> args = [ ] }

@penelope_zone $ srb tc -p parse-tree -e ‘a(1)’ Send {
receiver = NULL method = <U a> args = [ Integer { val = "1" } ] }

@penelope_zone

@penelope_zone For like 99% of Ruby programs this will never
be a problem

@penelope_zone I refuse to build an auto formatter which could
parse your program wrong

@penelope_zone

@penelope_zone Disadvantages

@penelope_zone Ripper requires a booted Ruby interpreter

@penelope_zone Running Ruby is slower than not running Ruby

@penelope_zone Ripper's output is painful

@penelope_zone $ rr 'a' [:program, [[:vcall, [:@ident, "a", [1, 0]]]]]

@penelope_zone $ rr 'a' [:program, [[:method_add_arg, [:fcall, [:@ident, "a", [1,
0]]], [:arg_paren, [:args_add_block, [[:@int, "1", [1, 2]]], false]]]]]

@penelope_zone Rubyfmt uses ripper to ensure compatibility

@penelope_zone Building against this is 90% of the dev time

@penelope_zone I would be done if I used whitequark

@penelope_zone I may end up separating Ripper from the Ruby
interpreter as a general purpose tool

@penelope_zone It’s totally a good thing that Sorbet and Rubocop
did not do this

@penelope_zone Speed is needed at any cost

@penelope_zone Gofmt executes on 25ms on even very large files

@penelope_zone

@penelope_zone How fast is fast enough?

@penelope_zone 16ms

@penelope_zone 16ms 16ms 16ms 16ms 16ms 16ms 16ms 100ms

@penelope_zone 100ms on 3k lines of Ruby code is the
goal

@penelope_zone ruby —disable=gems -e ‘’

@penelope_zone 16ms 16ms Ruby boot 25ms

@penelope_zone ruby -e ‘'

@penelope_zone 16ms 16ms Ruby boot with gems 75ms 16ms 16ms
16ms

@penelope_zone We can’t run the parser in the remaining 25ms

@penelope_zone So we can't use gems

@penelope_zone Speed is needed at any cost

@penelope_zone 100 100 100 100 100 100 100 100

@penelope_zone bundle exec <anything>

@penelope_zone 100m 100m 100m 100m 100m 100m 100m 100m Bundler
boot 100 100 100 100 100 100 100 100

@penelope_zone So we definitely can’t use bundler

@penelope_zone bundle exec rubocop <4 line file> 800ms 100m 100m
100m 100m 100m 100m 100m 100m Bundler boot Rubocop 100 100 100 100 100 100 100 100

@penelope_zone And we definitely can’t inherit from Rubocop

@penelope_zone So what does it look like today?

@penelope_zone

@penelope_zone So the complete Ruby version isn’t fast enough

@penelope_zone

@penelope_zone Parse.y isn’t really separable from the ruby interpreter

@penelope_zone ❤

@penelope_zone

@penelope_zone The initial test of the Rust version is that
it is fast enough

@penelope_zone I am essentially building an entire set of tooling
in Rust to work with Ruby source code

@penelope_zone Summary

@penelope_zone Do one thing well Correctness beats simplicity Speed is
needed at any cost

@penelope_zone These principles imply a tonne of work

@penelope_zone Rubyfmt is the most technically complex project I have
ever worked on

@penelope_zone Sweating the details

@penelope_zone Thanks to

@penelope_zone https://github.com/ penelopezone/rubyfmt

@penelope_zone That’s all @penelope_zone [email protected]

Introducing Rubyfmt

Introducing Rubyfmt

More Decks by Penelope Phippen

Featured

Transcript