launchpad-dev team mailing list archive

[Robert Collins <robert.collins@xxxxxxxxxxxxx>]

On Thu, Apr 14, 2011 at 12:47 PM, Martin Pool <mbp@xxxxxxxxxxxxx> wrote:
>> This is a good point; my prototype schema will never show 0 as nonzero
>> - it will only show nonzero (calculated using source data) as possibly
>> higher-nonzero (calculated using the summary table).
>
> As John mentioned, saying there are 3 critical bugs when there are
> actually 2 is nearly as alarming.
>
> In short, if there are so few bugs in a class that they could all
> appear on a screen (say <50) and the user could count them by hand,
> people will notice if it's not precise.  I don't know if that is
> reasonable to implement in your approach.
>
> I think there is a history of bugs to support the thesis people do
> notice discrepancies at small absolute values.

They may, but the current situation is that we cache things so the
discrepancy hangs around for a while. As I've said *repeatedly* this
is feasible to updated in realtime, and would *only* show discrepancy
due to multiple-path subscriptions.

>> Indeed. OTOH I think with 2 critical *private* bugs you'd been quite
>> unlikely to be subscribed twice. IMBW.
>
> If you mean having two different subscription paths to see the bug, I
> think it does happen.  Subscription is used for a bunch of things in
> Launchpad, including the dupefinder offering to subscribe you (and I'm
> pretty sure that's independent of whether you're already implicitly
> subscribed.)  I think another case is apport bug filing directly
> subscribing you even if you're already in a team that can see it.

It does happen yes. We may be able to come up with something better,
but I can't tell if you are saying:
A 'no do not do this, we prefer accuracy and timeouts'
or
B 'if you can do better than you describe that would be awesome'
or
C 'have you considered the psychological impact of this corner case'

For A, I think most folk disagree; for B I don't have anything better
to offer yet, not doable in the short term [and these queries we're
trying to handle are sitting at 6+ seconds *hot*]. For C, yes I have;
I think its a fairly rare corner case, tolerable if we documented, and
certainly something we can improve on in future [but requires more
work].

-Rob