[Jython-dev] Fw: A cumbersome weakref-observation

[Stefan R. <Ste...@gm...>]

Okay, sorry for bothering you. I already found out the reason for all this.

When handing a number to a Java-style weakref (or any other non-Jython-aware
Java method), Jython implicitly converts it to a Java-Float or Integer object,
while tuples and lists are kept PyTuple or PyList.
So the Java-style weakref actually points to a distinct object than the Jython-style one.

The reason for strings beeing "overpersistent" regarding Java-weakrefs is internation
I suppose. Jython hands the interned string to the weakref, but the interned string
is immortal anyway, so the weakref is never cleared.

Regards

Stefan


> Gesendet: Sonntag, 19. Juli 2015 um 18:43 Uhr
> Von: "Stefan Richthofer" <Ste...@gm...>
> An: "Jython Developers" <jyt...@li...>
> Betreff: Fw: A cumbersome weakref-observation
>
> Update: If insert a del-statement, strings are overpersistent for Java
> style wekarefs:
> 
> import time
> from java.lang import System
> from java.lang.ref import WeakReference
> import weakref
> 
> l = "test"
> wkl = WeakReference(l)
> wkr = weakref.ref(l)
> print "wkl: "+str(wkl.get())
> print "wkr: "+str(wkr())
> del l
> System.gc()
> time.sleep(2)
> print "wkl2: "+str(wkl.get())
> print "wkr2: "+str(wkr())
> 
> 
> ----
> Output is:
> wkl: test
> wkr: test
> wkl2: test
> wkr2: None
> 
> However with tuples and lists both variants work fine.
> 
> 
> > Gesendet: Sonntag, 19. Juli 2015 um 18:06 Uhr
> > Von: "Stefan Richthofer" <Ste...@gm...>
> > An: "Jython Developers" <jyt...@li...>
> > Betreff: A cumbersome weakref-observation
> >
> > Hello all, I just stumbled over some (seemingly?) strange behavior
> > regarding weak references:
> > 
> > import time
> > from java.lang import System
> > from java.lang.ref import WeakReference
> > import weakref
> > 
> > l = 123999.7
> > wkl = WeakReference(l)
> > wkr = weakref.ref(l)
> > print "wkl: "+str(wkl.get())
> > print "wkr: "+str(wkr())
> > System.gc()
> > time.sleep(2)
> > print "wkl2: "+str(wkl.get())
> > print "wkr2: "+str(wkr())
> > 
> > 
> > ----
> > Output is:
> > wkl: 123999.7
> > wkr: 123999.7
> > wkl2: None
> > wkr2: 123999.7
> > ----
> > With l = "test" the output is:
> > wkl: test
> > wkr: test
> > wkl2: test
> > wkr2: test
> > ----
> > "wkl2: None" happens for floats and ints, but not for strings, tuples and lists.
> > (These are the ones I tested so far.)
> > 
> > So the Java-style weak reference breaks for number-types, although the referent
> > is not gc'ed. Can anyone tell me how this comes? What is special about numbers
> > regarding java-style weak references?
> > 
> > I describe this behavior as "cumbersome" rather than a bug, because the official
> > weak-ref variant works fine (Why?). Still, there is no obvious reason why the Java-variant
> > should not work and I need to understand this issue. Maybe it even points to a subtle
> > bug. Looking at the weakref-module code I don't see special treatment of number-types
> > so far. What do I overlook?
> > 
> > Thanks in advance!
> > 
> > Stefan