feat: new wait for idle

[dimaqq]

Same as #1219 but using dumb classes instead of async generator.

[james-garner-canonical]

Thanks, it probably seems silly but this is significantly easier for me to follow. check and Loop.next look good to me.

I've commented on your FIXME note -- also a minor refactor suggestion to put the the error checking loops behind a raise_on_error flag (and likewise for blocked).

Also a few nitpicks on the tests just to do my due diligence, but I'm guessing you're fine with things as is. The most significant note is that it would probably be worth duplicating the check tests for blocked/errors to cover the "raise_on_{blocked,error}" = False cases (unless you've already covered that and I've just missed it).

[james-garner-canonical]

I notice there are a couple of failing integration tests (they all passed last run in #1219) -- related to this refactor, or just more test flakiness?

[dimaqq]

failed (quaranteed) integration test:

juju.errors.JujuError: refreshing charm "ch:amd64/focal/hello-juju-8": unexpected content-type from server "text/html"

and stable integration test:

Post "https://api.charmhub.io/v2/charms/refresh": net/http: TLS handshake timeout

looks like charmhub had a glitch, restarted.

[dimaqq]

tests passed. https://github.com/juju/python-libjuju/actions/runs/12404794817

[dimaqq]

@james-garner-canonical I've implemented most suggestions, pelase review :)

[dimaqq]

real charm CI runs:

• https://github.com/canonical/mysql-router-operator/actions/runs/12409228356?pr=196
• https://github.com/dimaqq/seldon-core-operator/actions/runs/12409256210

[james-garner-canonical]

Looks great! Real charm CI runs are reassuring too.

Interesting how this charm tests run has two timeouts with old wait for idle in test_upgrade_from_edge (3.4/stable on focal and jammy) -- which both passed with new wait for idle. Since it's a timeout I guess this could be a flaky test generally, but I do wonder if this indicates something -- was old wait for idle more eager to bail out with a timeout error rather than check everything? I don't think this necessarily needs further investigation right now, just noting it

[dimaqq]

/merge