Jean-Francois.Doyon@CCRS.NRCan.gc.ca a écrit :
Hello,
Does setting a persistent object as attribute to another persistent object, and then accessing it directly through the web, create a leak?
For example:
from OFS.Image import Image
class myclass(Image):
def __init__(self) self.thumbnail = Image(id='thumbnail')
and then access it using something like: There's no /leak/, but the thumbnail attribute is a persistent attribute of instances of 'myclass'.
http://host.com/myinstance/thumbnail
Wouldn't this leave the thumbnail lying around after the GC cycle, since once the system is done with the parent instance, the refcount falls to zero, it's cleaned out, and the thumbnail is orphaned, creating a memory leak? If I'd want a per request created and garbaged Image instance I'd do something like this :
class myclass(Image): def thumbnail(self) : """a doc""" return Image(id='thumbnail') But if you want a per database transaction (it should make some valuable cache for each http connection) created and garbaged objects you should use zope volatile (_v_) attribute : class myclass(Image): def thumbnail(self) : """a doc""" if not hasattr(self, '_v_thumb') : self._v_thumb = Image(id='thumbnail') return self._v_thumb
I was told at some point that this was OK, but now that I'm hunting memory leaks and learning more about that, this no longer seems sensible.
And since I do it all over the place, it'd explain my horrible leak of AcquisitionWrappers and HTTPRequests and so on ... As well as high refcount for CMFImage.Image ...
Memory leaks appears in python in only one case : if two objects mutually reference them and are not referenced elsewhere. In zope this cannot happen as any object orphaned in the ZODB tree will be garbaged.