Expected behaviour

Using SPEC_OPTS="--bisect" should work the same as using --bisect on the command line.

Actual behaviour

RSpec prints its bisection startup message, then continuously spawns new processes and does not terminate.

Steps to reproduce

cat > blah_spec.rb <<EOS
RSpec.describe 'blah' do
  it 'is true' do
    expect(true).to be_truthy
  end
end
EOS

rspec blah_spec.rb --bisect # Works as normal

SPEC_OPTS="--bisect" rspec blah_spec.rb # loops infinitely spawning new rspec instances

Discussion

I'm not sure if this is necessarily a case you'll want to handle, but it was a problem we ran into recently while trying to bisect order-dependent failures from CI. We use a separate test runner, knapsack, to manage parallelisation of our build. Since this controls RSpec invocation, it's frequently necessary to use SPEC_OPTS to add extra options such as --fail-fast and --bisect.

When --bisect is specified at the command line, RSpec spawns child processes each time it runs a bisect round, removing --bisect from the original CLI arguments so the child processes don't try to start a new bisection.

When --bisect is specified in SPEC_OPTS, however, it's passed through to the child processes, which then start a new bisection, launch more child processes, and so on and so on.

I'd be happy to submit a PR to address this problem, but first wanted to check whether you think it's worth the extra complexity that would be required to support --bisect via SPEC_OPTS, or whether this is an option you only want to support as a CLI argument.