Is it too messy to test this? Feels like it could brake again in the future

Is it too messy to test this?

Anything is possible, but, yes, it was a PITA

Updated.

Added test fails in master with:

  1) Failure:
TestIntegrationPumactl#test_require_dependencies [/mnt/c/Greg/GitHub/puma/test/test_integration_pumactl.rb:200]:
Expected "uninitialized constant Puma::DSL::LogWriter\n" to include "Command restart sent success".

1. Added a test, but since the previous use of pumactl (TestIntegration#cli_pumactl) started it via Puma::ControlCLI.new, any files loaded in the test process might corrupt the test. So, added TestIntegration##cli_pumactl_spawn, which runs pumactl from a command line in a sub-process.

2. Previously, Puma::CLI was lazy loaded. Changed Puma::Configuration and Puma::StateFile to also lazy load.

3. Note that the test has two uses of pumactl, one using a config file, and one using a state file.

Lastly, GitHub Actions appears to be messed up (again). Same branch passed in my fork.

@dentarg

Thanks. Don't know if you looked at the 2nd run, and for those that aren't often looking at Actions job logs, Actions seems to intermittently have problems with external connections.

I've seen problems reaching keyservers, OS package sites (installing packages with apt, brew, etc), and even RubyGems. In the second run here, one of the JRuby jobs failed when installing ragel, see https://github.com/puma/puma/actions/runs/5414643664/jobs/9853016363?pr=3187#step:4:111