Examples for post_save signal usage

[arski]

Hi, could you please elaborate/add specific examples how FieldTracker will and will NOT work with post_save signal handlers. Im referring to this in the docs:

Due to the implementation of FieldTracker, post_save signal handlers relying on field tracker methods should only be registered after model creation.

Thanks!

[romgar]

@carljm you had some use cases in mind when writing this?

[carljm]

I think @treyhunner (author of FieldTracker) might have better memory. IIRC it's just that FieldTracker itself relies on a post_save signal handler, and signal handlers run in the order they are registered, so if you have a post_save handler that needs to use FieldTracker data from that save, it needs to be registered later. Normally this isn't a problem if you register your listener in the normal way, outside of model class creation (that is, at module level or in an AppConfig.ready method); it's only a potential problem if you register the handler in e.g. a field contribute_to_class method, or something else that occurs during model class creation.

[arski]

@carljm you mean FieldTracker clears it's "changed" states in a post_save handler, so basically your handler must get there before that happens? Sounds a bit flaky to be used reliably.

I tested with a post_save handler in my handlers.py file and it doesn't seem to be passed anything to instance.tracker.changed() e.g., so I guess that is too late already :(

[carljm]

@arski No, that's the opposite of what I described.

But it looks like I was wrong anyway, FieldTracker doesn't use post_save at all, it patches the model save method. I've never really used FieldTracker and don't recall what that doc sentence is about (nor do I know why your test with post_save isn't working, it should work AFAIK); hopefully @treyhunner can clarify.

[kezabelle]

I believe that the documentation line being discussed was added by Trey after we briefly did change to using a signal, which we subsequently reverted. The root issue was #83 (pickling), #130 (PR) fixed that, but #143 meant it got reverted.

[carljm]

Thanks @kezabelle. I guess maybe that sentence should just be removed then.