bundler/inline: perform installation from a forked child by casperisfine · Pull Request #7941 · ruby/rubygems

casperisfine · 2024-08-19T11:17:43Z

Unless of course fork isn't available.

Alternate: #7930, #7933
Fix: #7930, #7933

When bundler inline has to install gems, it loads more dependencies than when it goes through the fast path of all gems being installed.

One of them is securerandom so if trying to use bundler/inline with a gem that have a dependency on securerandom that don't match the default version, the script fails with Gem::LoadError.

This can be preproduced on Ruby 3.2.x, after making sure to gem uninstall securerandom so only the default gem remains, and then running the following script:

require 'bundler/inline'

gemfile do
  source 'https://rubygems.org'
  gem 'activesupport', '7.2.0' # depends on securerandom >= 0.3
end

require 'securerandom'

casperisfine · 2024-08-19T12:18:05Z

+          # If the install succeeded, we need to refresh gem info
+          Bundler.reset!
+
+          builder = Bundler::Dsl.new
+          builder.instance_eval(&gemfile)
+          builder.check_primary_source_safety
+
+          definition = builder.to_definition(nil, true)
+          def definition.lock(*); end
+          definition.validate_runtime!


Of course it's harder than I expected, because since the install is done in a subprocess the parent need to refresh the gem info. I'm super un-familiar with rubygems/bundler do I don't know what the clean way of doing this, but it seems to work except for native gems :/

There is 3 failing specs left I don't know how to fix:

rspec ./spec/runtime/inline_spec.rb:193 # bundler/inline#gemfile installs subdependencies quietly if necessary when the install option is not set, and multiple sources used rspec ./spec/runtime/inline_spec.rb:218 # bundler/inline#gemfile installs quietly from git if necessary when the install option is not set rspec ./spec/runtime/inline_spec.rb:323 # bundler/inline#gemfile installs gems with native extensions in later gemfile calls

Sign in to view

+        else
+          do_install.call
+        end
      end


Unless of course fork isn't available. Alternate: ruby#7930, ruby#7933 Fix: ruby#7930, ruby#7933 When bundler inline has to install gems, it loads more dependencies than when it goes through the fast path of all gems being installed. One of them is `securerandom` so if trying to use bundler/inline with a gem that have a dependency on securerandom that don't match the default version, the script fails with `Gem::LoadError`. This can be preproduced on Ruby 3.2.x, after making sure to `gem uninstall securerandom` so only the default gem remains, and then running the following script: ```ruby require 'bundler/inline' gemfile do source 'https://rubygems.org' gem 'activesupport', '7.2.0' # depends on securerandom >= 0.3 end require 'securerandom' ```

hsbt · 2026-03-11T06:36:30Z

@casperisfine @byroot Sorry to noisy notification. I finally resolved the CI failure and rebased with the current master branch. Maybe Ruby 3.2 will also reach EOL and this issue will no longer be an issue.

However, the default gems will continue to remain in rubygems and bundler, so I will merge this.

byroot · 2026-03-11T07:32:12Z

No worries, I'm not notified on pushes. Thanks for the merge!

`script/run_snippets.sh` runs the snippets through `bundler/inline`, which installs the gems and sets up the runtime in one process. Since bundler 4.0.15 that produces a `$LOAD_PATH` built from specs that never learned their real `require_paths`, so every gem is assumed to use the default `lib`. `concurrent-ruby`, an ActiveSupport dependency, uses `lib/concurrent-ruby`, and the first snippet dies before running a single example: Running avoid_fixture_name_collision.rb There was an error while trying to load the gem 'rails'. Gem Load Error is: cannot load such file -- concurrent/map CI hits this on every run because its gem home starts out empty, so the snippets are always the ones installing those gems. Bisecting the released gems against a fresh gem home puts the break at exactly 4.0.15, which shipped on 2026-06-24 -- the day after the last green build on `main`. The cause is ruby/rubygems#9618: `Installer#install` now calls `release_resolution_memory!`, discarding the memoized remote index before installing. `Source::Rubygems#install` annotates the resolved `EndpointSpecification` in place rather than swapping it, and `EndpointSpecification#load_paths` only reports the gemspec's real `require_paths` while that annotation is present. Any re-resolution after the install therefore rebuilds pristine specs that report `<gem>/lib`. It is still present in 4.0.18; `master` avoids it only incidentally, by installing in a forked child (ruby/rubygems#7941, never backported to 4.0). Pin until there is a release we can move to.

`script/run_snippets.sh` runs the snippets through `bundler/inline`, which installs the gems and sets up the runtime in one process. Since bundler 4.0.15 that produces a `$LOAD_PATH` built from specs that never learned their real `require_paths`, so every gem is assumed to use the default `lib`. `concurrent-ruby`, an ActiveSupport dependency, uses `lib/concurrent-ruby`, and the first snippet dies before running a single example: Running avoid_fixture_name_collision.rb There was an error while trying to load the gem 'rails'. Gem Load Error is: cannot load such file -- concurrent/map CI hits this on every run because its gem home starts out empty, so the snippets are always the ones installing those gems. Bisecting the released gems against a fresh gem home puts the break at exactly 4.0.15, which shipped on 2026-06-24 -- the day after the last green build on `main`. The cause is ruby/rubygems#9618: `Installer#install` now calls `release_resolution_memory!`, discarding the memoized remote index before installing. `Source::Rubygems#install` annotates the resolved `EndpointSpecification` in place rather than swapping it, and `EndpointSpecification#load_paths` only reports the gemspec's real `require_paths` while that annotation is present. Any re-resolution after the install therefore rebuilds pristine specs that report `<gem>/lib`. It is still present in 4.0.18; `master` avoids it only incidentally, by installing in a forked child (ruby/rubygems#7941, never backported to 4.0). Reported upstream as ruby/rubygems#9781. Pin until there is a release we can move to.

casperisfine force-pushed the inline-fork branch from 754c404 to e715b14 Compare August 19, 2024 11:18

casperisfine mentioned this pull request Aug 19, 2024

bundler/inline: re-exec script after installation #7933

Closed

casperisfine force-pushed the inline-fork branch 2 times, most recently from 72db816 to c9c77d2 Compare August 19, 2024 12:12

casperisfine commented Aug 19, 2024

View reviewed changes

Playgirlkaybraz11 reviewed Sep 22, 2024

View reviewed changes

Comment thread bundler/lib/bundler/inline.rb

This comment was marked as spam.

Sign in to view

Playgirlkaybraz11 reviewed Sep 22, 2024

View reviewed changes

Comment thread bundler/lib/bundler/inline.rb

else

do_install.call

end

end

This comment was marked as spam.

Sign in to view

hsbt added the bundler: bug fix label Nov 20, 2025

hsbt force-pushed the inline-fork branch 2 times, most recently from 413e8d3 to 11962a7 Compare January 14, 2026 01:37

byroot and others added 2 commits March 11, 2026 12:17

Reset Gem::Specification and mark inline sources

4ce1510

hsbt force-pushed the inline-fork branch from 11962a7 to 4ce1510 Compare March 11, 2026 05:02

hsbt removed the bundler: bug fix label Mar 11, 2026

hsbt self-assigned this Mar 11, 2026

hsbt added the bundler: feature label Mar 11, 2026

hsbt force-pushed the inline-fork branch 2 times, most recently from bf83056 to 21586e7 Compare March 11, 2026 06:04

Configure definitions and YAML status after forked install

80e37bb

hsbt force-pushed the inline-fork branch from 21586e7 to 80e37bb Compare March 11, 2026 06:05

hsbt merged commit 6fd37f4 into ruby:master Mar 11, 2026
95 checks passed

pirj mentioned this pull request Aug 17, 2026

bundler/inline builds a wrong $LOAD_PATH for gems it installs itself (regression in 4.0.15) #9781

Closed

Navigation Menu

Uh oh!

Uh oh!

bundler/inline: perform installation from a forked child - #7941

bundler/inline: perform installation from a forked child#7941
hsbt merged 3 commits into
ruby:masterfrom
casperisfine:inline-fork

casperisfine commented Aug 19, 2024

Uh oh!

casperisfine Aug 19, 2024

Uh oh!

This comment was marked as spam.

Uh oh!

This comment was marked as spam.

Uh oh!

hsbt commented Mar 11, 2026

Uh oh!

Uh oh!

byroot commented Mar 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Uh oh!

Uh oh!

Conversation

casperisfine commented Aug 19, 2024

Uh oh!

casperisfine Aug 19, 2024

Choose a reason for hiding this comment

Uh oh!

This comment was marked as spam.

Uh oh!

This comment was marked as spam.

Uh oh!

hsbt commented Mar 11, 2026

Uh oh!

Uh oh!

byroot commented Mar 11, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants