Prevent loading all non-Bundler gems

[Bo98]

• Have you followed the guidelines in our Contributing document?
• Have you checked to ensure there aren't other open Pull Requests for the same change?
• Have you added an explanation of what your changes do and why you'd like us to include them?
• Have you written new tests for your changes? Here's an example.
• Have you successfully run brew style with your changes locally?
• Have you successfully run brew typecheck with your changes locally?
• Have you successfully run brew tests with your changes locally?

---

This prevents random user-installed gems from breaking brew in various ways, including but not limited to:

• Broken native extensions of critical components, like fileutils, causing brew to not load. [When brew is used, an error is reported：'/usr/local/opt/libyaml/lib/libyaml-0.2.dylib' not found #10520, Homebrew won't install on M1 Mac (macOS 11.2.1) #10684, many many discussions]
• Various warnings about user-installed gems that brew isn't responsible for. [always error output: Ignoring redcarpet-3.5.0 because its extensions are not built. Try: gem pristine redcarpet --version 3.5.0 #10649, Brew is 'Ignoring ffi-1.13.1 because its extensions are not built.' #10675, Ruby errors on every brew upgrade #10688]
• Breaking-change updates of gems being used which brew has not been tested against. [undefined method `assertions' for main:Object #2984]

Prior discussion in #7681.

[BrewTestBot]

Review period will end on 2021-02-25 at 18:08:29 UTC.

[Bo98]

I cannot reproduce that rubocop-ast error. Nor can I think of anything that could cause it.

It should be in the $LOAD_PATH:

brew/Library/Homebrew/vendor/bundle/bundler/setup.rb

Line 81 in 3f39c20

$:.unshift "#{path}/../#{ruby_engine}/#{ruby_version}/gems/rubocop-ast-1.4.1/lib"

and therefore unaffected by any GEM_PATH changes.

[MikeMcQuaid]

I'd be tempted to move this require to brew.rb instead. This file is also used by rubocops.rb.

[MikeMcQuaid]

I cannot reproduce that rubocop-ast error. Nor can I think of anything that could cause it.

Bootsnap? The Library/Homebrew/rubocops.rb load_path require?

[Bo98]

Bootsnap?

I don't think that actually is being used on Linux:

brew/Library/Homebrew/homebrew_bootsnap.rb

Line 10 in 74f7d2f

(ENV["HOMEBREW_PROCESSOR"] == "Intel")

HOMEBREW_PROCESSOR being "Intel" is a macOS-specific quirk.

The Library/Homebrew/rubocops.rb load_path require?

Call path appears to be brew bottle. It doesn't happen in my Linux docker though.

[Bo98]

I think I figured it out - we (intentionally) remove non-existent directories from the $LOAD_PATH. But because we Homebrew.install_bundler_gems! mid-run, some of the gitignored gems won't be in the $LOAD_PATH on first install. Removing the "exists" check causes this horrible bug again though:

/home/linuxbrew/.linuxbrew/Homebrew/Library/Homebrew/vendor/portable-ruby/2.6.3_2/lib/ruby/2.6.0/rubygems/core_ext/kernel_require.rb:54: [BUG] Segmentation fault at 0x0000000000000018

I'll play around with it.

[MikeMcQuaid]

I don't think that actually is being used on Linux:

brew/Library/Homebrew/homebrew_bootsnap.rb

Line 10 in 74f7d2f

(ENV["HOMEBREW_PROCESSOR"] == "Intel")

HOMEBREW_PROCESSOR being "Intel" is a macOS-specific quirk.

Whoops, should revert that as it's not intentional.

[Bo98]

The last commit should fix the Ruby crash issue once and for all.

[MikeMcQuaid]

brew/Library/Homebrew/test/dev-cmd/test_spec.rb

Lines 9 to 10 in 485f73a

	# randomly segfaults on Linux with portable-ruby.
	it "tests a given Formula", :integration_test, :needs_macos do

would be good to remove the comment and :needs_macos. This is looking good, though.

[BrewTestBot]

Review period ended.

[Bo98]

Enabling that test immediately finds a bug! Fixed that now.

(I've also made the brew test command print the stacktrace to stderr instead of stdout - it makes the integration test for it have much better output when it fails)

[MikeMcQuaid]

Thanks again @Bo98, looks like this is uncovering a bunch of 🐛 too. If the scope here is getting a little large feel free to extract any of it into a separate PR e.g. just the minitest changes.

[MikeMcQuaid]

Looks great 👍🏻

[carlocab]

Related? Homebrew/discussions#953

[Bo98]

We do not ship a rake script (though the gem was there) with portable Ruby, so that's half intentional.

I suspect the user is actually wanting to use system Ruby however, which would make sense for formula build scripts specifically. This probably broke because GEM_PATH and GEM_HOME is bleeding into formulae.

So we can either:

• Not use those envs and use alternative Gem API.
• Reset those ENVs in a build environment.

I'll try out the former just now because it looks interesting.

[Bo98]

#10726

[reitermarkus]

Is there a way now to allow specific Gems which are not in the Gemfile? The cask review action broke because it cannot load the git_diff gem.

[Bo98]

I have a fix for Homebrew.install_gem! here: #10735

For more complex scenarios, bundler still works (homebrew-formula-analytics).

[Bo98]

Merged now:

$ brew irb
==> Interactive Homebrew Shell
Example commands available with: `brew irb --examples`
irb(main):001:0> Homebrew.install_gem! "git_diff"
==> Installing 'git_diff' gem
Fetching git_diff-0.4.3.gem
irb(main):002:0> require "git_diff"
=> true
irb(main):003:0> defined? GitDiff
=> "constant"

Also works on subsequent runs - Homebrew.install_gem! will still set up the path of gems already installed.