This patch applies to several methods that deal with filepaths in Dir,
File, File::Stat, IO, and Kernel. All of these methods now raise an
argument error if a null byte is detected in a path argument.

Some of these methods are also slightly refactored, mostly by replacing
context.runtime with just runtime where applicable.

* Not used anymore and we can get the current method to achieve
  RubyCallStack.getCurrentDeclaringModule().

… MRI does.

…s InternalMethod.

* Only used by super and Module.nesting which are not interested in RubyProc.
* For backtrace, this is done separetely with the call nodes,
  so there is no influence from this change.

These are both rarely encountered in scope bodies so it would be nice to remove an
explicit impl from the in the body interpreter.

Fix the bug with size with each_slice without block

* No need to reconstruct the barrier, it does not break on interrupt.
* No need for tricky locking scheme.
* Entirely unsynchronized leaveThread.

…and implement #local_variables

when you have embedded gems inside a jarfile and the filename was something
like myjar-20150210+behappy.jar then the rubygems could not find those gems.

Sponsored by Lookout Inc.

it removes the case my.jar!/f%20o.rb which matches the behaviour of
file:f%20o.rb

Sponsored by Lookout Inc.

Fixed arity issue with RubyMethod

This reverts commit 21576dd.

* No need to reconstruct the barrier, it does not break on interrupt.
* No need for tricky locking scheme.
* Entirely unsynchronized leaveThread.
* Fix: main thread needs to call enterThread().

Make filepaths with null byte fail

* Kill all threads except main in RubyContext.shutdown.
* Add aliases to KillNode.

* This is the behavior of MRI.
* If it turns out we still want it, we can add it in a finally block in run().
* Do not catch Throwable but just Ruby exceptions.

* They did not run anymore since they were waiting indefinitely for the global lock.
* Stealing the global lock is not nice at all.
* This assumes we do not re-enter the SafepointManager.

Revert "Fixed arity issue with RubyMethod"

…irectly the VirtualFrame.

* We do not need the global lock to poll anymore.

…tics.

* RubyThread.join() already raises if needed.
* A node returning null is a bug.

* This reverts commit b2446b7.
* The current SafepointManager needs for threads to have the global lock to know which RubyThread it is.

Our File.expand_path always canonicalizes a path, whereas MRI's does not.  By forcing the entire SPEC_TEMP_DIR through File.expand_path, we'll get uniformly wrong behavior, making any String comparisons using SPEC_TEMP_DIR and file paths pass.

…propagating whether

subtrees have assignments in them and having more specific nodes makes it easier to know
whether I need to check whether it has an assignment or not in it (numeric types cannot).

…fies truffle and IR

since it does not need to do implicit nil check AND a null check.