Continuation of c9a43126

😢

Currently, Travis CI assumes that the project is in Ruby in the
absence of the `language` key.
This behavior may change in the future. (For example, switch to a "blank"
image which *may* be created.)

Emulate `jruby --dev` which improves JRuby's startup time.

See
https://github.com/jruby/jruby/wiki/Improving-startup-time#use-the---dev-flag

Follow @headius' idea and give Java more memory to reduce GC time.

This will make easier to compare run time between Ruby 2.1 and trunk

Not sure what's causing them suddenly, but it seems unlikely there's
much we can do about it. For the small amount of coverage this
particular job gains us, it's hurting us more, by making CI failures
unremarkable.

This reverts commit 5d2d3010.

cc http://blog.travis-ci.com/2014-04-28-upcoming-build-environment-updates/

/cc @Brixen

Ruby 2.1.0 includes the json gem 1.8.1 by default so we need bundler 1.5.1
for `bundle install` to work. To fix this reverse the downgrade to 1.3.5
and wrap the `run_generator` call with a block that resets `THOR_DEBUG`.

Since rbx and jruby builds are being allowed to fail lets receive the
build information early enabling this option

See http://about.travis-ci.org/blog/2013-11-27-fast-finishing-builds/

Bundler sets the THOR_DEBUG environment variable causing Thor to
re-raise exceptions on exit. This is breaking the railties tests
for generators where we are trying to capture error output.

The PR bundler/bundler#2794 fixes this by adding a per-instance
config `:debug` but this has not been released yet so to prevent
errors going unnoticed downgrade bundler to 1.3.5 until such a
time that 1.5.2 has been released.

This reverts commit 875dd3edfd34eaa20120b6589b0d1905230e948f.

The issue was fixed in rbx-2.2.3

Rubinius has problems with bundling when shared gems are disabled (ie.
when --path or --deployhment is used). There is no real reason to
install it with --path on TravisCI, so I'm dropping it.

Before Rubinius 2.2.0, some parts of the stdlib were lazily loaded, even
if not included in the Gemfile. Now it's 100% required to include the
gem 'rubysl', otherwise the stdlib is not available entirely, breaking
everything.

Also, the rubysl-test-unit gem was depending on minitest ~> 4.7, thus causing
conflicts with Rails's dependency on minitest 5. That is not the case
anymore since rubysl-test-unit 2.0.2, so it is completely safe to
include it fully.

When Travis is going to update RVM to 1.24, it'll be safe to use rbx-2,
thus picking new versions automatically.

Fixes for LOT of bugs have now landed in Jruby 1.7.6, thus we no longer
need to run tests against the master.

for broken tests that are still remaning:
1. If problem is in rails - we fix that test/code
2. if its a JRuby bug - we skip that test & file a bug report for jruby.

Many of the failing tests in rails test suite fail because of missing encoding support, example #11739


And according to recent announcement, lot of encoding issues have been fixed in current master.  http://ruby.11.x6.nabble.com/Big-encoding-patches-landed-on-master-td4993719.html


So lets test against JRuby master & incase there are still bugs. it would be easier to report them to JRuby.

This adds the retry behaviour used for other commands to the bundler installation.