It previously only rescued Errno::ECONNREFUSED and Errno::EBADF. On a
few occasions, I have also seen the Errno::ETIMEDOUT exception raised by
Net::HTTP from this method. Since the Net::HTTP calls are the only thing
in this method that should be making any kind of system calls, a rescue
of all SystemCallError exceptions (the superclass of all of the Errno
exceptions) will catch these as well as any other system call errors
triggered by Net::HTTP.
When they were added in 09d6335, responsive? already only hit
/__identify__, which does not hit the server, but gets special-cased in
Capybara's Identify server class (and runs instantly).
So it's not clear what constitutes a "slow" server that might time out
at boot, and thus why server_boot_timeout needs to be configurable. A
server that is slow to respond to HTTP requests would not trigger the
boot timeout. All we really seem to be waiting for within the timeout
window is for Thin or WEBrick to come up.
In order to run Selenium via Capybara on my virtual machine, you need to be
able to bind the server IP to a net interface on the same subnet as the VM.
This patch exposes an configuration variable to allow you to bind the
test server to any net interface on your local machine.
Falls back to 127.0.0.1 if the config var is not set.
John Wilger