Opened 17 years ago
Closed 16 years ago
#6778 closed (duplicate)
dispatch signal needed on many to many relationship alterations (add, delete)
Reported by: | Owned by: | Andrew Gibson | |
---|---|---|---|
Component: | Core (Other) | Version: | dev |
Severity: | Keywords: | dispatcher relation m2m | |
Cc: | George Song, Christian Schilling, rvdrijst | Triage Stage: | Accepted |
Has patch: | yes | Needs documentation: | yes |
Needs tests: | no | Patch needs improvement: | yes |
Easy pickings: | no | UI/UX: | no |
Description
When a model has a relation added, no signal is sent.
This breaks functionality where the model is internally caching processed relational data, and the model becomes associated with new (or disassociated with) items. There is no method to inform the original model that it's cache is stale.
it would seem appropriate that all functions with the property 'alters_data' (originally intended for template use) would have pre/post signals.
(django/db/models/fields/related.py, django/contrib/contenttypes/generic.py: add, create, remove, clear )
to give a trivial example:
class Attendee(models.Model): name = models.CharField(maxlength=255) class Conference(models.Model): name = models.CharField(maxlength=255) attendees = models.ManyToManyField(Attendee, related_name='conferences') def num_attendees(self): if not hasattr(self,"attendee_count"): setattr(self,"attendee_count",len(self.attendees.all())) return self.attendee_count def stale_cache(self): if hasattr(self,"attendee_count"): delattr(self,"attendee_count")
(generally, i'm looking to do more complex processes than take the len() of a queryset...)
a,b,c = Attendee(name="Alice"), Attendee(name="Bob"), Conference(name="PyCon") for i in (a,b,c): i.save() c.attendees.add(a) c.num_attendees() c.attendees.add(b) c.num_attendees()
obviously calls to stale_cache() would fix the problem, but the application-level programmer shouldn't necessarily have to deal with cache maintenance. (DRY... someone's bound to forget to update cache somewhere), it would be better of there is a signal to do this on relational updates.
Enabling this should be fairly straightforward:
- add appropriate signals to django.db.models.signals
- call the dispatcher from appropriate functions in the relational model managers, etc
Attachments (2)
Change History (11)
comment:1 by , 16 years ago
Triage Stage: | Unreviewed → Accepted |
---|
comment:2 by , 16 years ago
comment:3 by , 16 years ago
Cc: | added |
---|
comment:4 by , 16 years ago
Owner: | changed from | to
---|---|
Status: | new → assigned |
comment:5 by , 16 years ago
Has patch: | set |
---|
follow-up: 7 comment:6 by , 16 years ago
Needs documentation: | set |
---|---|
Patch needs improvement: | set |
Docs would be good (/docs/ref/signals.txt
)
Also, you can't use the new decorator format - Django has to be Python 2.3 compatible
by , 16 years ago
Attachment: | patch.diff added |
---|
comment:7 by , 16 years ago
Replying to SmileyChris:
Docs would be good (
/docs/ref/signals.txt
)
Also, you can't use the new decorator format - Django has to be Python 2.3 compatible
removed new decorator format, added docs and put the whole patch into one file
comment:8 by , 16 years ago
Cc: | added |
---|
comment:9 by , 16 years ago
Cc: | added |
---|---|
Resolution: | → duplicate |
Status: | assigned → closed |
Looks like a duplicate of #5390
This is a big issue. Right now, it is impossible to perform tasks when relations are added or removed. Consider this use-case:
How would I send a message to an User when he is added to the Survey? Hooking into post_save of Survey does not work, as its unrelated to the 'participants' manager. #6095 would offer a way to achieve this (by hooking management code into an intermediary model). But using intermediary models will break the admin interface (eg. 'filter_horizontal'), and that's a huge consequence to adding some management code.
So signals seem to be the best solution to this problem. I suggest we name those signals 'pre/post_add_relation' and 'pre/post_remove_relation'. As I see it, there are two issues.
We just have the sender and instance arguments to pass information.. maybe we can use a tuple to pass both related objects as instance and use the m2m field as sender?
Thinking of it, we could refractor the m2m manager to use a generic intermediary model. This model would then send the stock pre/post_save signals. This would allow a much cleaner implementation of #6095, too.