#35801: Signals are dispatched to receivers associated with dead senders
------------------------------+------------------------------------
Reporter: bobince | Owner: (none)
Type: Bug | Status: new
Component: Core (Other) | Version: 5.1
Severity: Normal | Resolution:
Keywords: | Triage Stage: Accepted
Has patch: 1 | Needs documentation: 0
Needs tests: 0 | Patch needs improvement: 1
Easy pickings: 0 | UI/UX: 0
------------------------------+------------------------------------
Changes (by Simon Charette):
* has_patch: 0 => 1
* needs_better_patch: 0 => 1
Comment:
Thanks for all the extra details Sjoerd, this is greatly appreciated.
Here's how I think we can resume the problem in a few bullet points
1. `Signal` doesn't provide a way to connect to non-weak senders
2. `Signal.receivers` doesn't hold a strong reference to senders as it
uses `id` to retrieve them by lookup and under memory pressure `id` can
return the same value for two distinct objects
3. The migration framework creates and disposes of `ModelBase` instances
(aka `Model` subclasses) which creates a lot of memory pressure (see
#29898) and cycles through `sender` for all model signals which makes the
perfect conditions for `id(SomeFakeModel) == id(SomeOtherFakeModel)`
collisions.
With all that said I can see a few solutions which should be considered.
First making `Signal.connect(weak=True)` also use `weakref` for senders
like
[https://github.com/bobince/django/compare/main..signals_from_dead_senders
bobince implemented] seems like a good step forward. If you could open a
PR from your branch that also tests `Signal(use_caching=True)` (which
should work as it already uses a `WeakKeyDictionary` for its sender cache)
we should be land this work independently of the following proposed
solutions. I believe we should have `weakref.ref(sender)` raise a
`TypeError` instead of silencing though as the proposed code now holds a
''strong'' reference to `sender` in `receivers` which could make things
much worst but this can be discussed further during code review.
That should leave us with what should be done with signal receivers
connected with `weak=False`. In the case of `django-dirty-fields`'s usage
Sjoerd I think they should switch to `weak=True` as their `reset_state`
receiver is defined at the module level anyway so it's causing more harm
than good. There are albeit rare use cases for `weak=False` and it's a
public API so I wonder if the migration framework should do a better job
at clearing after itself by implementing `__del__` logic for `AppConfig`
and fake models for the core signals at least.
--
Ticket URL: <https://code.djangoproject.com/ticket/35801#comment:7>
Django <https://code.djangoproject.com/>
The Web framework for perfectionists with deadlines.
--
You received this message because you are subscribed to the Google Groups
"Django updates" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion visit
https://groups.google.com/d/msgid/django-updates/01070194323cd1c9-36597853-ebc7-468d-bd75-8ed66c0334b8-000000%40eu-central-1.amazonses.com.
