Apparently my workaround with do end did not fix it in travis, but did fix it locally on my macbook: https://travis-ci.org/logstash-plugins/logstash-input-http_poller/jobs/238068792

I'm stumped as to how to actually fix this

Braces versus do/end should not make any difference, nor should environment. I suspect there's some dependency-loading issue here, like on one env things are loading in a different order and failing.

Can you narrow down what's actually failing in the app?

@headius you're right, I just got lucky playing the roulette wheel when I changed the braces. It looks like it's something to do with require in rspec that's causing the bug.

If I add require 'rspec/matchers/built_in/raise_error.rb' to the top of my _spec file that seems to fix it.

It looks like the requires should be happening here: https://github.com/rspec/rspec-expectations/blob/master/lib/rspec/matchers.rb#L8 which brings us to:
https://github.com/rspec/rspec-support/blob/3-6-maintenance/lib/rspec/support.rb#L14

While that code is a bit convoluted I don't see anything that seems unsafe there.

BTW, I don't know if you noticed this @headius , but in the error it says:

uninitialized constant RSpec::Matchers::BuiltIn::RaiseError
       Did you mean?  RSpec::Matchers::BuiltIn::RaiseError

So it seems like JRuby simultaneously knows the constant exists, and doesn't know it exists, since the recommendation is for the thing that is missing! Any ideas?

I assume there's not a lot of concurrency in the specs, right?

Do they run in some randomize order?

@headius they are randomized but not concurrent. Some of the specs may spawn additional threads, but the actual expectations are all in main.

The confusing thing to me is that my reading of the rspec code indicates that that the require should happen correctly at that point. Additionally, I can access that constant before the specs run. I suspect that if I replaced my duplicate require with a reference to RSpec::Matchers::BuiltIn::RaiseError that would work as well. I can access that constant correctly before any tests run as well.

This is all very confusing.

I suspect this is related to concurrency issues in requires or autoloads (which should be fixed by #5764), or else concurrency issues in RSpec similar to rspec/rspec-dev#121.

In any case this issue is rather stale and superseded by others so we'll close it as a dupe.