[meld] matchers.helpers: Revert to using multiprocessing instead of threading

[Kai Willadsen <kaiw src gnome org>]

commit 07ec4fd9b01d8efdf74cc268dc35aa4f5791c48b
Author: Kai Willadsen <kai willadsen gmail com>
Date:   Sun Jul 16 08:58:45 2017 +1000

    matchers.helpers: Revert to using multiprocessing instead of threading
    
    Using threads here is (mostly) blocking our UI thread and I don't really
    understand why. Until I can figure this out, pushing us back to
    multiprocessing fixes our UI interactivity at a fairly minor performance
    penalty.
    
    Python's GIL means that throwing our comparison on to a thread doesn't
    actually get us real multi-CPU concurrency... but it shouldn't be
    causing the comparison thread to starve the main thread. It's not
    actually blocking either, because some UI events *do* get through while
    a comparison is running, just not many.
    
    My current suspicion is that this is some kind of interaction between
    the GIL, thread scheduling and GLib's event loop. However, that's a
    guess because I don't have any other explanation.

 meld/matchers/helpers.py |    8 ++++----
 1 files changed, 4 insertions(+), 4 deletions(-)
---
diff --git a/meld/matchers/helpers.py b/meld/matchers/helpers.py
index 86c23ac..55cbe59 100644
--- a/meld/matchers/helpers.py
+++ b/meld/matchers/helpers.py
@@ -1,7 +1,7 @@
 
 import logging
+import multiprocessing
 import queue
-import threading
 import time
 
 from gi.repository import GLib
@@ -12,7 +12,7 @@ from meld.matchers import myers
 log = logging.getLogger(__name__)
 
 
-class MatcherWorker(threading.Thread):
+class MatcherWorker(multiprocessing.Process):
 
     matcher_class = myers.InlineMyersSequenceMatcher
 
@@ -45,12 +45,12 @@ class CachedSequenceMatcher(object):
 
     def __init__(self):
         self.cache = {}
-        self.tasks = queue.Queue()
+        self.tasks = multiprocessing.JoinableQueue()
         # Limiting the result queue here has the effect of giving us
         # much better interactivity. Without this limit, the
         # result-checker tends to get starved and all highlights get
         # delayed until we're almost completely finished.
-        self.results = queue.Queue(5)
+        self.results = multiprocessing.JoinableQueue(5)
         self.thread = MatcherWorker(self.tasks, self.results)
         self.task_id = 1
         self.queued_matches = {}