I'm not suggesting that we have to run JBlanket every time before we
commit. What I'm saying is that before a developer adds a new module into
the daily build _for the first time_, it would be appropriate for them to
invoke the cruisecontrol target locally to find and remove as many problems
as they can. With respect to JBlanket, it would be usually be good enough
to test it on a small configuration containing the new module, as opposed
to hackystat-ALL. If that takes a while to run, so be it. (Note that
hackystat-ALL was actually failing in less than 10 minutes when JBlanket
was blowing up, so if there's a problem, you'll probably find out pretty
quickly. If JBlanket runs OK, and you have to wait an hour or so, that
seems better than failing the global build the next day.)
It's not just jblanket that can have problems. There were also some
problems with JavaDoc that I pointed out to Christoph which he hadn't
noticed, which would have been obvious with 'ant cruisecontrol'.
Cheers,
Philip
--On Wednesday, March 16, 2005 7:28 PM -1000 Aaron Kagawa
<[EMAIL PROTECTED]> wrote:
Actually, there might be a small problem in requiring developers to run
JBlanket before integration in to the Daily Build. See
<http://www.mail-archive.com/[email protected]/msg00553.html>.
This email shows that that running the tests with JBlanket takes 96
minutes on my brand new dell computer. Where as running the tests without
JBlanket (including a freshStart) takes only 10 minutes.
So.. Is running a freshStart good enough? If not, there seems to be a
trade off with running JBlanket before integration. Wait more than 96
minutes to be absolutely sure or just commit and you might fail the build.
thanks, aaron
At 05:25 PM 3/16/2005, you wrote:
--On Wednesday, March 16, 2005 10:37 AM -1000 "(Cedric) Qin ZHANG"
<[EMAIL PROTECTED]>
wrote:
We can assume Christoph has tested everything locally. No one is
expected to run JBlanket locally. So it should not be his fault.
Actually, I am not sure I agree. I think we should expect someone to run
JBlanket
locally when working on the initial integration of a module into the
build system (which
is what Christoph is doing now). What he should be invoking locally is:
ant cruisecontrol
which would expose the problems in his local environment.
Of course, once everything's OK, we would not expect a developer to check
jblanket
operations on a daily basis.
Cheers,
Philip
