A few issues

[Gustavo Niemeyer]

On Wed, Apr 10, 2013 at 7:22 PM, William Reade
<william.reade at canonical.com> wrote:
> On Wed, 2013-04-10 at 18:39 -0300, Gustavo Niemeyer wrote:
>> I believe this is incorrect. Python has the "installed" status as
>> well, and I believe it covers that case. "pending" means that the
>> agent data is missing entirely, which isn't the situation described.
>
> If it's running the "install" hook, it's not yet "installed". We do
> have, and pass through, an "installed" agent-state, but many charms
> don't spend long there before graduating to "started".

I stand corrected. It is certainly confusing to have no difference
between never seen and running the install hook, but if we can blame
someone for that confusion, that's myself. Sorry for that.

>> Yes, I didn't even know --fake-series existed. How does that work? Can
>> we make it easier for people just trying things out to not have to
>> read a lot of theory before it will work?
>
> There are two obvious improvements that need to be made to --fake-series
> (which is merely the latest (and hopefully among the last) of various
> short-term-focused responses to the lack of series handling). We should
> certainly default to uploading fake tools for the default-series
> regardless (which I think would have resolved your problem); and we
> should expose it for upgrade-juju as well (with similar defaults).
>
> This should hit the vast bulk of use cases; and the imminent changes to
> the tools selection logic would have caused a clearer failure earlier,
> so I think this is in hand.

+1.  I think "fake-series" is something extremely vague to a normal
person that wants to try out tip. I think I get what it does, but
that's because I still know something about the internals.

>> It's not clear to me how that's relevant. JSON already formatted
>> machine ids as strings, both in Python and in the Go version we
>> implemented. This issue is only relevant in YAML, and it was broken
>> after the internal change in machine ids.
>
> My contention is that inconsistent JSON output is a real issue, and the
> YAML change is collateral damage from the fix; but that bringing the two
> representations in line will still help more than it will hurt.

Understood.


gustavo @ http://niemeyer.net