I stumbled across this very slight but significant difference when trying to inject some HTTP headers into an application being tested with Capybara.

If you’re running a server-side authentication scheme such as LDAP or Shibboleth, you can rely on server variables to get the name of the person who’s authenticated to use your application. Usually it’s a variable like “REMOTE_USER” or something else. When you’re testing, you’ll want your test app to add that variable to the server’s headers.

Using Capybara::RackTest::Driver, this looks something like:

Capybara.register_driver(:driver_name) do |app|
  Capybara::RackTest::Driver.new(app,
                                 respect_data_method: true,
                                 headers: { 'HTTP_REMOTE_USER' => 'joe' })
end

And if we query the variable in our test:

>  request.env.fetch("HTTP_REMOTE_USER")
=> "joe"

Now let’s use the Capybara::Poltergeist::Driver instead so we can test some Javascript. At first, that looks like:

Capybara.register_driver(:poltergeist) do |app|
  driver = Capybara::Poltergeist::Driver.new(app, js_errors: true, timeout: 90)
  driver.headers = { 'HTTP_REMOTE_USER' => 'joe' }
  driver
end

But look what happens to our ENV hash:

>  request.env
=> {"HTTP_HTTP_REMOTE_USER"=>"joe"}

The Poltergeist driver is appending another HTTP_ on to the variable. I don’t have a definitive explanation, but I suspect this is because of a section of the CGI specification rfc3875 that dictates how variables sent from the client are inpreted by the server.

The gist is, since Poltergeist is a more fully-fledged client/server setup, it’s going to do the variable conversion for you, whereas the RackTest driver doesn’t and you have to do it yourself.

In any case, to fix the above problem, just don’t add the “HTTP” prefix to the variable if you’re using Poltergeist.