def test_gc_list_modify_silent(self):
#print "test_gc_list_modify_silent"
clearCurrentLeaks()
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 0)
l = [0, "test1"]
d = {'a': 7, 'b': "test6"}
wkl = WeakReference(l)
wkd = WeakReference(d)
wkl2 = weakref.ref(l)
wkd2 = weakref.ref(d)
self.assertIsNotNone(wkl.get())
self.assertIsNotNone(wkd.get())
self.assertIsNotNone(wkl2())
self.assertIsNotNone(wkd2())
DemoExtension.listSetIndex(l, 0, d)
del d
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 4)
#monitor.listAll()
runGC()
#monitor.listAll()
self.assertFalse(monitor.lastClearGraphValid)
self.assertIsNotNone(wkl.get())
self.assertIsNone(wkd.get())
self.assertIsNotNone(wkl2())
self.assertIsNotNone(wkd2())
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 3)
self.assertIsNotNone(l[0])
self.assertEqual(len(l[0]), 2)
del l
runGC()
#monitor.listAll()
self.assertTrue(monitor.lastClearGraphValid)
# For some reason resurrected objects persist one more
# gc-cycle in principle. So we have to run gc again before
# we can observe wkd2 to die. It is currently unclear whether
# this behavior is a JyNI-bug or natural Java-gc behavior,
# but for now (with some evidence) we assume the latter.
# Note: Since WeakRef-support wkd2 actually keeps the native
# referent alive for one more cycle. So also the monitor-refcount
# test only passes after another gc-run now.
runGC()
#monitor.listAll()
self.assertTrue(monitor.lastClearGraphValid)
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 0)
self.assertIsNone(wkl2())
self.assertIsNone(wkd2())
def test_gc_list_modify_silent(self):
#print "test_gc_list_modify_silent"
clearCurrentLeaks()
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 0)
l = [0, "test1"]
d = {'a': 7, 'b': "test6"}
wkl = WeakReference(l)
wkd = WeakReference(d)
wkl2 = weakref.ref(l)
wkd2 = weakref.ref(d)
self.assertIsNotNone(wkl.get())
self.assertIsNotNone(wkd.get())
self.assertIsNotNone(wkl2())
self.assertIsNotNone(wkd2())
DemoExtension.listSetIndex(l, 0, d)
del d
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 4)
runGC()
self.assertFalse(monitor.lastClearGraphValid)
self.assertIsNotNone(wkl.get())
self.assertIsNone(wkd.get())
self.assertIsNotNone(wkl2())
self.assertIsNotNone(wkd2())
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 3)
self.assertIsNotNone(l[0])
self.assertEqual(len(l[0]), 2)
del l
runGC()
self.assertTrue(monitor.lastClearGraphValid)
# For some reason resurrected objects persist one more
# gc-cycle in principle. So we have to run gc again before
# we can observe wkd2 to die. It is currently unclear whether
# this behavior is a JyNI-bug or natural Java-gc behavior,
# but for now (with some evidence) we assume the latter.
# Note: Since WeakRef-support wkd2 actually keeps the native
# referent alive for one more cycle. So also the monitor-refcount
# test only passes after another gc-run now.
runGC()
self.assertTrue(monitor.lastClearGraphValid)
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 0)
self.assertIsNone(wkl2())
self.assertIsNone(wkd2())
def test_gc_list_modify_silent(self):
#print "test_gc_list_modify_silent"
l = [0, "test1"]
d = {'a': 7, 'b': "test6"}
wkl = WeakReference(l)
wkd = WeakReference(d)
wkl2 = weakref.ref(l)
wkd2 = weakref.ref(d)
self.assertIsNotNone(wkl.get())
self.assertIsNotNone(wkd.get())
self.assertIsNotNone(wkl2())
self.assertIsNotNone(wkd2())
DemoExtension.listSetIndex(l, 0, d)
del d
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 4)
#print "run1"
runGC()
#print "run1 done"
self.assertFalse(monitor.lastClearGraphValid)
self.assertIsNotNone(wkl.get())
self.assertIsNone(wkd.get())
self.assertIsNotNone(wkl2())
self.assertIsNotNone(wkd2())
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 3)
self.assertIsNotNone(l[0])
self.assertEqual(len(l[0]), 2)
del l
#print "run2"
runGC()
#print "run2 done"
self.assertTrue(monitor.lastClearGraphValid)
self.assertEqual(len(monitor.getCurrentNativeLeaks()), 0)
self.assertIsNone(wkl2())
# For some reason resurrected objects persist one more
# gc-cycle in principle. So we have to run gc again before
# we can observe wkd2 to die. It is currently unclear whether
# this behavior is a JyNI-bug or natural Java-gc behavior,
# but for now (with some evidence) we assume the latter.
#print "run3"
runGC()
#print "run3 done"
self.assertIsNone(wkd2())
def run3():
JyNI.JyRefMonitor_setMemDebugFlags(1)
JyWeakReferenceGC.monitorNativeCollection = True
import DemoExtension
#JyNI.JyRefMonitor_setMemDebugFlags(1)
#JyWeakReferenceGC.monitorNativeCollection = True
l = [0, "test1"]
# print l
# DemoExtension.listSetIndex(l, 0, 100.7)
# print l
d = {'a': 7, 'b': "test6"}
#We create weak reference to l to monitor collection by Java-GC:
wkl = WeakReference(l)
wkd = WeakReference(d)
wkl2 = weakref.ref(l)
wkd2 = weakref.ref(d)
#Note:
#=====
#To make this work we'll have to ensure that d gets a GCHead even though
#l doesn't recognize the insertion of d and thus would not explore it.
#In fact every C-stub object must get a GCHead when going native, regardless
#of whether it is inserted somewhere or not, because it might be silently
#inserted and a GCHead is the only way to detect this and fix it.
#Add a call (e.g. to size) to the dict in l on native side to test that
#d still works (which it currently shouldn't).
#Later use d's GCHead to detect the invalid graph and resurrect d's java-part.
#Take care of the GIL here. CStubs must release the GIL when they detect a
#failing backend, so the GC-thread can acquire it and resurrect the backend.
#How to fix the weak reference etc?
#l[0] = d
print "weak(l): "+str(wkl.get())
print "weak(d): "+str(wkd.get())
print "weak2(l): "+str(wkl2())
print "weak2(d): "+str(wkd2())
print "make l native..."
#l[0] = d
#DemoExtension.argCountToString(l)
#l[0] = d
DemoExtension.listSetIndex(l, 0, d)
print "Delete l... (but GC not yet ran)"
#del l
del d
#l = None
#print "weak(l) after del: "+str(wkl.get())
print "weak(d) after del: "+str(wkd.get())
print "weak2(d) after del: "+str(wkd2())
print ""
print "Leaks before GC:"
monitor.listLeaks()
print ""
print "calling Java-GC..."
System.gc()
time.sleep(2)
# if monitor.lastClearGraphValid:
# print "valid graph"
# else:
# print "invalid graph"
# monitor.lastClearGraphValid = False
print "weak(l) after GC: "+str(wkl.get())
#print "l after GC: "+str(l)
print "weak(d) after GC: "+str(wkd.get())
print "weak2(l) after GC: "+str(wkl2())
print "weak2(d) after GC: "+str(wkd2())
wkd = WeakReference(l[0])
print ""
#monitor.listWouldDeleteNative()
print ""
#print "leaks after GC:"
#monitor.listLeaks()
print "l[0], i.e. d after gc:"
#print l[0]
#print len(l[0])
print "------"
print "del l..."
del l
System.gc()
time.sleep(2)
# if monitor.lastClearGraphValid:
# print "valid graph"
# else:
# print "invalid graph"
# monitor.lastClearGraphValid = False
print ""
monitor.listWouldDeleteNative()
print ""
#print "leaks after GC (l deleted):"
#monitor.listLeaks()
#print DemoExtension.argCountToString.__doc__
#monitor.listFreeStatus("dict")
#monitor.listAll()
#GlobalRef.processDelayedCallbacks()
print "weak(d) after GC2: "+str(wkd.get())
print "weak2(l) after GC2: "+str(wkl2())
print "weak2(d) after GC2: "+str(wkd2())
System.gc()
time.sleep(2)
print "weak(d) after GC3: "+str(wkd.get())
monitor.listWouldDeleteNative()
print ""
print "leaks after GC3:"
monitor.listLeaks()
print ""
print "===="
print "exit"
print "===="
def run3():
JyNI.JyRefMonitor_setMemDebugFlags(1)
JyWeakReferenceGC.monitorNativeCollection = True
import DemoExtension
#JyNI.JyRefMonitor_setMemDebugFlags(1)
#JyWeakReferenceGC.monitorNativeCollection = True
l = [0, "test1"]
# print l
# DemoExtension.listSetIndex(l, 0, 100.7)
# print l
d = {'a': 7, 'b': "test6"}
#We create weak reference to l to monitor collection by Java-GC:
wkl = WeakReference(l)
wkd = WeakReference(d)
wkl2 = weakref.ref(l)
wkd2 = weakref.ref(d)
#Note:
#=====
#To make this work we'll have to ensure that d gets a GCHead even though
#l doesn't recognize the insertion of d and thus would not explore it.
#In fact every C-stub object must get a GCHead when going native, regardless
#of whether it is inserted somewhere or not, because it might be silently
#inserted and a GCHead is the only way to detect this and fix it.
#Add a call (e.g. to size) to the dict in l on native side to test that
#d still works (which it currently shouldn't).
#Later use d's GCHead to detect the invalid graph and resurrect d's java-part.
#Take care of the GIL here. CStubs must release the GIL when they detect a
#failing backend, so the GC-thread can acquire it and resurrect the backend.
#How to fix the weak reference etc?
#l[0] = d
print "weak(l): " + str(wkl.get())
print "weak(d): " + str(wkd.get())
print "weak2(l): " + str(wkl2())
print "weak2(d): " + str(wkd2())
print "make l native..."
#l[0] = d
#DemoExtension.argCountToString(l)
#l[0] = d
DemoExtension.listSetIndex(l, 0, d)
print "Delete l... (but GC not yet ran)"
#del l
del d
#l = None
#print "weak(l) after del: "+str(wkl.get())
print "weak(d) after del: " + str(wkd.get())
print "weak2(d) after del: " + str(wkd2())
print ""
print "Leaks before GC:"
monitor.listLeaks()
print ""
print "calling Java-GC..."
System.gc()
time.sleep(2)
# if monitor.lastClearGraphValid:
# print "valid graph"
# else:
# print "invalid graph"
# monitor.lastClearGraphValid = False
print "weak(l) after GC: " + str(wkl.get())
#print "l after GC: "+str(l)
print "weak(d) after GC: " + str(wkd.get())
print "weak2(l) after GC: " + str(wkl2())
print "weak2(d) after GC: " + str(wkd2())
wkd = WeakReference(l[0])
print ""
#monitor.listWouldDeleteNative()
print ""
#print "leaks after GC:"
#monitor.listLeaks()
print "l[0], i.e. d after gc:"
#print l[0]
#print len(l[0])
print "------"
print "del l..."
del l
System.gc()
time.sleep(2)
# if monitor.lastClearGraphValid:
# print "valid graph"
# else:
# print "invalid graph"
# monitor.lastClearGraphValid = False
print ""
monitor.listWouldDeleteNative()
print ""
#print "leaks after GC (l deleted):"
#monitor.listLeaks()
#print DemoExtension.argCountToString.__doc__
#monitor.listFreeStatus("dict")
#monitor.listAll()
#GlobalRef.processDelayedCallbacks()
print "weak(d) after GC2: " + str(wkd.get())
print "weak2(l) after GC2: " + str(wkl2())
print "weak2(d) after GC2: " + str(wkd2())
System.gc()
time.sleep(2)
print "weak(d) after GC3: " + str(wkd.get())
monitor.listWouldDeleteNative()
print ""
print "leaks after GC3:"
monitor.listLeaks()
print ""
print "===="
print "exit"
print "===="