GC - Remote reference count

[bosley]

It may not play out well, but I had an idea for a garbage collection scheme I want to try.

• Global "collector" class that contains a std::set of std::size_t.
• When a collector_c receives an alloc request to produce a , it creates some struct that it ties a reference count to the object by allocating it, and storing its pointer in the set.
• When a dies, it will visit the collector, indicating that it has been nuked and the collector will handle the ref counting / deallocation.

This is really just an abstracted reference counter, but what it would buy the cell_c type is the following:

• No more need for the actual counter overhead incurred by the parent ref_counted_ptr class
• We could decorate all complex types (fns, dicts, etc) as a collected type, and then reference them only by a raw pointer, thus allowing us to move all data into the cell union.

This would ensure we continue to have 0 memory leaks while reducing the overall size of the cell by an estimated 16 bytes.

[bosley]

So I ran this on bosleylab (much slower than everything on the planet) and I got this differential in speed:

BEFORE:

test	time (s)	time (ms)
mandelbrot.nibi	0.1383s	138.3382ms
primality.nibi	1.2408s	1240.7899ms
4sieve.nibi	1.306s	1306.0257ms
leibniz.nibi	0.1469s	146.8661ms

AFTORE:

test	time (s)	time (ms)
mandelbrot.nibi	0.8463s	846.293ms
primality.nibi	8.0331s	8033.0782ms
4sieve.nibi	13.4166s	13416.5957ms
leibniz.nibi	0.9215s	921.5019ms

The performance hit was due to the fact that we introduced new hops for every copy and delete. The remote GC thing I figured would be slower, but not THIS much slower.

The performance hit was not worth the 4-bytes saving on cell_c