Make sure to also rescue non-standard error

[RusHibaM]

A potential way to solve #268

I tried to raise the SystemExit again as we have this test case:

raises DeadWorker when using exit so people learn to not kill workers and do not crash main process

If I just rescue all the exceptions, this test will fail since no DeadWorker will be raised when the worker exits.

I'm also thinking about allowing users to specify the non-standard errors that they wish to rescue. By doing this, we may not need to handle all different edge cases and just let the users choose what to rescue.

Parallel.each([1,2,3], extra_exceptions_to_handle: [Exception]) do |index|
end

[grosser]

we'll also need a test that explains the usecase / would not work with the old code

[RusHibaM]

we'll also need a test that explains the usecase / would not work with the old code

Sure, will add a test case.

BTW, I noticed that Parallel.map can work in isolation was failing in Travis for my changes. I'm a little bit confused as it seems to be passing locally on my machine.

Update:
Ah.. never mind, the Parallel.map can work in isolation is now passing in Travis...

[grosser]

1.19.0 ... let's see if this comes back to bite us :D

[RusHibaM]

1.19.0 ... let's see if this comes back to bite us :D

Haha, let's see. 😄

[valscion]

Neat change! I wonder if this list should also have NoMemoryError and Interrupt that should probably not be rescued either. This way the list of "exceptions not to rescue" would match what exceptions RSpec itself does not rescue:

https://github.com/rspec/rspec-support/blob/673133cdd13b17077b3d88ece8d7380821f8d7dc/lib/rspec/support.rb#L132-L140

    module AllExceptionsExceptOnesWeMustNotRescue
      # These exceptions are dangerous to rescue as rescuing them
      # would interfere with things we should not interfere with.
      AVOID_RESCUING = [NoMemoryError, SignalException, Interrupt, SystemExit]

      def self.===(exception)
        AVOID_RESCUING.none? { |ar| ar === exception }
      end
    end

Our usage of parallel won't really care about this, though — I merely spotted this change when updating parallel recently and thought you might want to know about how RSpec does a similar thing ☺️ (PR implementing this in rspec-support originally is here: rspec/rspec-core#2063 in case you're interested)

[grosser]

awesome, I copied that in 1.19.1