Let Selenium's WebDriver traffic bypass WebMock

WebMock hooks into every Net::HTTP request your tests make, including the WebDriver commands Selenium sends to chromedriver. A system spec sends hundreds of them. For each one, WebMock builds a request signature, matches it against all stubs and adds it to its request registry. In one project, profiling showed this took about 10% of the Ruby CPU time of the system and browser automation specs.

Settings like WebMock.disable_net_connect!(allow_localhost: true) or VCR's ignore_localhost don't help. They only decide what happens after WebMock has processed the request.

Selenium talks to a chromedriver on the same machine, so there is nothing to stub or record. This monkey patch lets its HTTP client use the original Net::HTTP implementation:

# spec/support/webmock.rb

# Selenium talks to chromedriver on this machine, so it bypasses WebMock (and VCR, which hooks into it).
module SeleniumHttpClientWithoutWebMock
  private

  def new_http_client
    WebMock::HttpLibAdapters::NetHttpAdapter::OriginalNetHTTP.new(server_url.host, server_url.port, nil)
  end
end

# Selenium::WebDriver::Remote::Http::Default#new_http_client is private. After an update, check that the
# patch is still called (e.g. no WebDriver requests in WebMock::RequestRegistry), then raise the version here.
if Gem::Version.new(Selenium::WebDriver::VERSION) > Gem::Version.new('4.44.0')
  raise "Please check whether SeleniumHttpClientWithoutWebMock still works with selenium-webdriver #{Selenium::WebDriver::VERSION}"
end

Selenium::WebDriver::Remote::Http::Default.prepend(SeleniumHttpClientWithoutWebMock)