Test on ruby-head, allow failures

[sodabrew]

@tamird @cbandy The query timeout workaround is failing on the uninitialized constant Timeout::ExitException in ruby-head. I wonder if they renamed this upstream!? I'll be sorting through https://github.com/ruby/ruby/blob/trunk/lib/timeout.rb soon...

https://travis-ci.org/sodabrew/mysql2/jobs/79014190

[sodabrew]

Yep: ruby/ruby@0f663b2

[tamird]

Sigh. Easy enough to solve by defining an alias at load-time.

[cbandy]

Maybe it's because I wasn't paying attention to the initial implementation...

Aren't there other places where interrupts can affect the protocol state? Are there some defines that help us do this around specific C calls in the extension?

[sodabrew]

There are other places where the interruption would cause a problem, but Tamir zoomed in on the most "obvious" and common point. This issue is certainly not a blocker at the moment, I am just getting ready for the upcoming Ruby release in December.

[jeremy]

Just hit this upgrading Active Record to support 0.4.0 (we test on Ruby trunk, too).

[jeremy]

Note that the ExitException removal was backported, too, but with an alias for compatibility: ruby/ruby@4027090#diff-32f7163092f0e8e64a3936fba1a3d59dR47

[tamird]

@jeremy sadly that won't help here as this library still supports all the way back to 1.8.7 (though the relevant versions are those that support Thread.handle_interrupt and don't have the backport: 2.0 and 2.1).

[jeremy]

@tamird Right, the backports only demonstrate that aliasing ExitException = Error is the way to go. See #671 😁