[WIP] PR for CI: Regions and other fun

[xFrednet]

Volatile branch, most likely not worth reviewing yet

[xFrednet]

Status Update:

The current branch provides a new regions module to construct a Region object. Creating a new bridge object using the Region() constructor will also create a new region. The write barrier is implemented but isn't being called in most places. This results in the LRC and OSC being almost always wrong.

There is also a mechanism to mark all open regions as dirty, however, currently it is not being called, which makes regions remain clean.

Here is a working example:

from regions import Region, is_local
class A:
    pass

r1 = Region()
r2 = Region()

# Check ownership
r1.owns(r1) == True
r2.owns(r1) == False
is_local(r1) == False

# Dictionary assignments have some write barriers
a = A()
r1.a = a

is_local(a) == False
r1.owns(a) == True

The invariant can upper bound the LRC and OSC values and catches incorrect references.

These are the high level TODOs:

• Add more methods to the Region type (dissove(), close(), is_dirty(), better repr etc)
• Add more test for regions, methods and objects in the standard library
• Handle name argument to Region constructor
• Handle de-allocation of the bridge object (Merge into local and regiondata.name = NULL)
• Add write barrier to dictionary and bytecode interpreter
• Add a way to mark types and modules as Pyrona aware
• Mark open regions as dirty, if Pyrona unaware C code is called
• Store the bridge RC on close and validate the bridge RC on open, to detect dirty states on closed regions
• Rebase cowns, and wire them up to regions
• Support weak references, maybe include them in the LRC or add special handling for regions
• Merging a region into the local region should open subregions, if the merge didn't happen for error handling

Current approach: Try to break the implementation in the REPL. On failure, add write barriers or do other fixes. Turn passing code into tests.

---

Bug 1:

>>> r.rr = regions.Region(name="two")
>>> r
<Region name='duck' _lrc=3 _osc=1 is_dirty=False>
>>> r.rr
<Region name='two' _lrc=1 _osc=0 is_dirty=False>
>>> 

The OSC of "duck" is not updated when "two" closes. Maybe because it doesn't close? o.O

---

Bug 2

import sys
import regions
r = regions.Region("duck")
sys.getrefcount(r)

This increases the RC but doesn't inform LRC

---

Bug 3

>>> r = Region()
>>> r
<Region name=None _lrc=2 _osc=0 is_dirty=False>
>>> r
<Region name=None _lrc=3 _osc=0 is_dirty=False>
>>> r
<Region name=None _lrc=4 _osc=0 is_dirty=False>

The rc is never decremented