Segmentation fault with Ruby 3.4 + Rails 8.1 when coverage is enabled (negative lineno in class_eval)

[george-bit]

Steps to reproduce

1. Use Ruby 3.4.x, Rails 8.1.x, view_component 4.2.0
2. Enable code coverage with SimpleCov (or any coverage tool)
3. Run bin/rails test

The segfault occurs during component compilation when class_eval is called with a negative line number (-1).

Root cause: In v4.2.0, Template::File#initialize sets lineno = -1 for Rails 8.1+ with ERB handler:

# lib/view_component/template.rb lines 26-31
lineno =
  if Rails::VERSION::MAJOR >= 8 && Rails::VERSION::MINOR > 0 && details.handler == :erb
    - 1
  else
    0
  end

This is then passed to class_eval at line 119:

@component.class_eval <<~RUBY, @path, @lineno

When Ruby's coverage is enabled, negative line numbers in eval/class_eval trigger a segmentation fault. This is documented in https://bugs.ruby-lang.org/issues/19363.

Expected behavior

Tests run without crashing.

Actual behavior

Segmentation fault during eager loading/component compilation.

Backtrace:

  /vendor/bundle/ruby/3.4.0/gems/view_component-4.2.0/lib/view_component/template.rb:119: [BUG] Segmentation fault at 0x0000000000000024
  ruby 3.4.8 (2025-12-17 revision 995b59f666) +PRISM [x86_64-linux]

System configuration

Rails version: 8.1.2

Ruby version: 3.4.8

Gem version: 4.2.0

---

This works on 4.1.1 which always uses lineno: 0.

[joelhawksley]

Thank you for the report! Would you be interested in working on a fix?

[george-bit]

I've submitted a PR with a fix: #2541

The fix: Detects whether Ruby's Coverage module is running and adapts the lineno value accordingly:

• Coverage OFF → uses -1 (correct line numbers)
• Coverage ON → uses 1 (avoids segfault, line numbers ~2 off)

Root cause: The -1 lineno introduced in v4.2.0 triggers a known Ruby bug (bugs.ruby-lang.org/issues/19363) where negative line numbers in class_eval cause segfaults when coverage is enabled. Testing showed lineno=0 also crashes, so 1 is the minimum safe value.

Trade-off: This is the best fix possible given an underlying Ruby bug. Most users get correct line numbers; only CI/coverage users see slightly off line numbers—but their tests run instead of crashing.

I've also raised a question in the PR about whether there's a cleaner long-term solution, since the current approach involves a chain of workarounds for Rails 8.1's annotation feature.