New self-organizing/self-optimising threading model (no API changes)

Using a simple, hysteresis based approach… currently in this first
version the optimizer recursively moves modules out of a thread until a
certain timing target is reached; the "modules moved out" just mentioned
are then partitioned into new threads according to their latest time profile.

Very simple, but very effective.
Todo: Implement merging threads with (again) good behaving workloads.

i3pystatus/core/modules.py

@@ -2,6 +2,7 @@ from threading import Thread
 import time
 
 from .util import SettingsBase
+from .threads import AutomagicManager
 
 __all__ = [
     "Module", "AsyncModule", "IntervalModule",
@@ -28,13 +29,23 @@ class AsyncModule(Module):
     def mainloop(self):
         """This is run in a separate daemon-thread"""
 
-class IntervalModule(AsyncModule):
+class IntervalModule(Module):
     interval = 5 # seconds
+    managers = {}
+
+    def registered(self, status_handler):
+        if self.interval in IntervalModule.managers:
+            IntervalModule.managers[self.interval].add(self)
+        else:
+            am = AutomagicManager(self.interval)
+            am.add(self)
+            IntervalModule.managers[self.interval] = am
+
+    def __call__(self):
+        self.run()
 
     def run(self):
-        """Called every self.interval seconds"""
+        """Called approximately every self.interval seconds
 
-    def mainloop(self):
+        Do not rely on this being called from the same thread at all times.
-        while True:
+        If you need to always have the same thread context, subclass AsyncModule."""
-            self.run()
-            time.sleep(self.interval)

i3pystatus/core/threads.py

@@ -0,0 +1,131 @@
+
+import sys
+import threading
+import time
+import traceback
+
+if hasattr(time, "perf_counter"):
+    timer = time.perf_counter
+else:
+    timer = time.clock
+
+class StderrExcDelegate:
+    def __call__(self, exc):
+        traceback.print_exception(*sys.exc_info(), file=sys.stderr)
+
+class ExceptionWrapper:
+    def __init__(self, workload, exc_delegate):
+        self.workload = workload
+        self.exc_delegate = exc_delegate
+
+    def __call__(self):
+        try:
+            self.workload()
+        except Exception as exc:
+            self.exc_delegate(exc)
+
+    def __repr__(self):
+        return repr(self.workload)
+
+class WorkloadWrapper:
+    def __init__(self, workload):
+        self.workload = workload
+        self.time = 0.0
+
+    def __call__(self):
+        tp1 = timer()
+        self.workload()
+        self.time = timer() - tp1
+
+    def __repr__(self):
+        return repr(self.workload)
+
+class Thread(threading.Thread):
+    def __init__(self, target_interval, workloads=None):
+        super().__init__()
+
+        self.workloads = workloads if workloads is not None else []
+        self.target_interval = target_interval
+        self.daemon = True
+
+    def __iter__(self):
+        return iter(self.workloads)
+
+    def __len__(self):
+        return len(self.workloads)
+
+    def pop(self):
+        return self.workloads.pop()
+
+    def push(self, workload):
+        self.workloads.append(workload)
+
+    def run(self):
+        while self.workloads:
+            for workload in self.workloads:
+                workload()
+
+            self.workloads.sort(key=lambda workload: workload.time)
+
+            filltime = self.target_interval - self.time
+            if filltime > 0:
+                time.sleep(filltime)
+
+    @property
+    def time(self):
+        return sum(map(lambda workload: workload.time, self.workloads))
+
+class AutomagicManager:
+    def __init__(self, target_interval):
+        self.target_interval = target_interval
+        self.upper_bound = target_interval * 1.1
+        self.lower_bound = target_interval * 0.7
+
+        initial_thread = Thread(target_interval, [self.wrap(self)])
+        self.threads = [initial_thread]
+
+        initial_thread.start()
+
+    def __call__(self):
+        separate = []
+
+        for thread in self.threads:
+            separate.extend(self.optimize(thread, thread.time))
+
+        self.create_threads(self.partition(separate))
+
+    def wrap(self, workload):
+        return WorkloadWrapper(ExceptionWrapper(workload, exc_delegate=StderrExcDelegate()))
+
+    def optimize(self, thread, vtime):
+        if len(thread) > 1 and vtime > self.upper_bound:
+            remove = thread.pop()
+            return [remove] + self.optimize(thread, vtime - remove.time)
+        return []
+
+    def partition(self, workloads):
+        timesum = 0.0
+        new_threads = []
+        current_thread = []
+        for workload in workloads:
+            current_thread.append(workload)
+            timesum += workload.time
+
+            if timesum > self.lower_bound:
+                new_threads.append(current_thread)
+                current_thread = []
+                timesum = 0
+
+        return new_threads
+
+    def create_threads(self, threads):
+        for workloads in threads:
+            self.create_thread(workloads)
+
+    def create_thread(self, workloads):
+        thread = Thread(self.target_interval, workloads)
+        thread.start()
+        self.threads.append(thread)
+
+    def add(self, workload):
+        self.threads[0].push(self.wrap(workload))