On Tue, Apr 27, 2010 at 2:42 PM, Michal M <mich.mier...@googlemail.com> wrote: > On 27 Kwi, 23:21, Duncan Booth <duncan.bo...@invalid.invalid> wrote: >> Michal M <mich.mier...@googlemail.com> wrote: >> > I've just found out that one of objects is not destroyed when it >> > should be. This means that something was holding reference to this >> > object or part of it (i.e. method). Is there any way to check what >> > holds that reference? I am unable to do that just looking to the code >> > or debugging it because it is pretty complicated, but I am able to >> > invoke this situation again. >> >> See if this code helps: >> >> http://groups.google.com/group/comp.lang.python/browse_thread/thread/... >> >> It's pretty old so it may need some adjustment, but I wrote it >> to figure out exactly that sort of problem. > > Thanks you for answers. > I tried to use gc.get_referrers(self) in point in code when according > to me object should be destroyed earlier but I did not see anything > interesting. Then just realised that gc will not destroy object even > when itself holds reference to his own method. > > For example > > class A(object): > def a(self): > pass > def b(self): > self.method = self.a > def __del__(self): > print "A object deleted" > >>> a = A() >>> del a > A object delted >>> a = A() >>> a.b() >>> del a > ... nothing ... > > I thought gc would discover such circle loops but apparently it did > not.
No, it does, you just didn't give it long enough; for performance reasons, cyclical GC is only done every so often: >>> from weakref import ref >>> class A(object): ... def a(self): ... pass ... def b(self): ... self.method = self.a ... >>> a = A() >>> def bye(x): print "BYE:", x ... >>> b = ref(a, bye) >>> del a BYE: <weakref at 0x377990; dead> >>> a = A() >>> b = ref(a, bye) >>> a.b() >>> del a >>> >>> #indeed, it didn't get GC-ed immediately >>> from gc import collect >>> collect() #but if we force a cyclical GC run... BYE: <weakref at 0x4441b0; dead> Cheers, Chris -- http://blog.rebertia.com -- http://mail.python.org/mailman/listinfo/python-list
