harmony-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mark Hindess" <mark.hind...@googlemail.com>
Subject Re: classlib test suite status emails?
Date Tue, 11 Apr 2006 15:06:19 GMT
On 4/11/06, Geir Magnusson Jr <geir@pobox.com> wrote:
>
> George Harley wrote:
> > Geir Magnusson Jr wrote:
> >>
> >>
> >> George Harley wrote:
> >>> Hi Geir,
> >>>
> >>> From my point of view the answer is yes, I ran the tests for the
> >>> patched module. The build/test server picked up the downstream
> >>> breakages in the other modules. I fixed matters so that the builds
> >>> ran again.
> >>
> >> Right - I thought we considered the build/test server a safety to keep
> >> accidents from getting away from us, but given the delay and the
> >> distribution of people throughout the planet working on this (rather
> >> than all sitting in the same office), we want to avoid breakages at
> >> all costs. (Again, I'm guilty of the same thing...)
> >>
> >>>
> >>> Elsewhere in this thread I am trying to cooperate with others to make
> >>> it simpler to run more - just the necessary - tests prior to making a
> >>> commit.
> >>
> >> I guess the question is "what is necessary"? Given a dependency chain
> >>
> >> A <- B <- C
> >>
> >> I don't think it's only the immediate dependencies of a given module
> >> A, but rather the full dependency graph, because a "1 away" module (C)
> >> might use a direct dependency (B) in a way different than the test
> >> cases for that direct dependency do (especially since we aren't quite
> >> up to speed with doing implementation testing...)
> >
> > Hi Geir,
> >
> > Sure, point taken. Testing A and then its set of immediate dependants
> > (just B in this example) will greatly reduce the risk of downstream
> > module breakages but will not completely catch all such events.
> > Actually, subject to the quality of the tests in C, it is possible that
> > even testing A, B and C will not find any such submerged problems.
>
> Indeed.
>
> >
> > I am not an expert in the theory of testing ( as if you had to be told
> > :-) ) but instinct alone tells me that making it convenient and quick to
> > run a module's tests and those of its dependants will definitely
> > *improve* matters - it would certainly have helped me last night before
> > I broke the build.
>
> There is a cultural element to it - maybe we keep tabs on who breaks the
> build, and that decides who buys the beer next time we all meet...

Not that I object to George buying me beer, but this seems a little
harsh considering he only broke 15 tests out of ~5400.  Would he have
to buy everyone dinner if he broken them all? ;-)

Regards,
-Mark.

--
Mark Hindess <mark.hindess@googlemail.com>
IBM Java Technology Centre, UK.

---------------------------------------------------------------------
Terms of use : http://incubator.apache.org/harmony/mailing.html
To unsubscribe, e-mail: harmony-dev-unsubscribe@incubator.apache.org
For additional commands, e-mail: harmony-dev-help@incubator.apache.org


Mime
View raw message