Philippe Bossut wrote:
Katie Capps Parlante wrote:
This approach seems pretty reasonable for Mozilla (or Safari, which I
hear has a similar system), where the intense focus is on a
particular well-known use case. In our case, we're continuing to
tweak the functionality and even change the formulation of the use
cases that we measure. I think it would be overly constraining at
this point to have a formal "back all regressions out" policy.
Agree. The "back out regressions" policy was also used on MacIE (and
MSN for OS X). It was extremelly efficient but, as Katie said, was
focusing on well-known use case and a particularly stable feature set.
I think we just had our first "use case" of performance regressions on
the trunk... I checked in a big change to calendar yesterday which
unfortunately had the rather unexpected side effect of causing new event
creation to be slow (again).. heikki noticed and filed a bug - thank
goodness because I didn't even realize I slowed stuff down (mostly
because I checked in at the end of the day... again, a historical graph
would have helped me help myself here)
But since we're at the beginning of a milestone, and there are other
architecture changes coming down the pipe in the next few weeks, I think
it makes sense to hold off on trying to optimize that case again, until
we've sorted out some of the other architecture stuff, like the hints
bit I keep talking about.
[And a further note about bug management: Sure, I have one more bug
against me, but it will be prioritized over time along with all the
others.. just because heikki filed it against me doesn't mean I need to
interrupt everything I'm doing to decide how to handle this bug]
Fortunately, we have a good baseline established in 0.6, and a bug filed
against me, to 'keep me honest' about actually fixing the regression,
one way or the other, for the next release.
Alec
I think that enforcing this policy on Chandler (or Cosmo for that
matter) right now would lead to Premature Optimization issues for any
significant feature we add.
Note that we (MacIE) also used a historized graph (as recommended by
Alec and others) and it's only when we got that implemented that
people (devs) could make sense of the data and we started to see
significant perf improvments.
Cheers,
- Philippe
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Open Source Applications Foundation "Dev" mailing list
http://lists.osafoundation.org/mailman/listinfo/dev
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Open Source Applications Foundation "Dev" mailing list
http://lists.osafoundation.org/mailman/listinfo/dev
