Penny Gale

I've been thinking about this (and also reading a lot of code). In an OF charm, we usually initialize one instance of the Framework class. It contains, among other things,...

@rwcarlsen the invocation to super does a few more things, like map events to a handler. It's actually pretty cool, in that you can define completely separate classes with completely...

Heh, the pseudo singleton nature of some of this stuff was already a problem in the testing harness: https://github.com/canonical/operator/blob/45e1e5a9e4e99492d09423436b8378f955c4f0a3/ops/testing.py#L131 (Ran into this while trying to write a RollingOpsManager with a...

Thank you for filing this!

I suspect that a lot of people using the lib would just pass -f, and create their own problems detecting typos. I like the simplicity of having the library work...

Ideally, we'd implement this as part of keeping up with core Juju features, when they implement the replacement for this concept (goal state is mostly deprecated, per my current understanding)....

Thank you for the well and thoroughly written bug. I agree that scoped statuses would be highly useful. I'll sketch out a design, and see when we can slot the...

After sleeping on this, and talking to some folks about this, I have more thoughts. Mostly, I'm a little concerned about charms routinely setting, then unsetting a "blocked" status. If...

The root cause is probably that the indirection in the Relation and RelationMapping classes means that we're not actually updating the original Relation object when we call `add_relation_unit`. In production,...

@rwcarlsen thanks for the clarification. I I like the idea of solving this as a property. We could even make it a property only on the testing model backend (I...