Behavior of Thread::new() on passed instances

The original implementation (see also #38) is designed so that Thread.new(a) makes a full copy of a in the newly created thread (unless mrb_type(a) == MRB_TT_DATA).
As a consequence, changes to a in any given thread are not visible to the other threads. This way, passing information between threads becomes a hard task.

Since I am still not sure about the reasons behind this choice, I have made an experimental branch in my own fork (pbosetti/mruby-thread) where the C macro MRB_THREAD_COPY_VALUES can be set to enable the original behavior (make copies) or leave undefined to switch to a "share the same instances" behavior.

The script examples/data.rb can be used to compare the two cases: when MRB_THREAD_COPY_VALUES is set, the changes to globals $data and $ary made in the subthread are not visible into the main thread. Conversely, if MRB_THREAD_COPY_VALUES, any change to instances passed to Thread::new() are visible to other threads.

The question is: can I make a PR for this?

Thanks,
Paolo

Start by reading the original implementation referenced in #38 and the experimental branch pbosetti/mruby-thread. Compare the two MRB_THREAD_COPY_VALUES behaviors using examples/data.rb, then determine whether shared instances are an accepted design and what tests or documentation would define the intended Thread::new behavior.