On Aug 2, 2010, at 2:52 PM, Martin Buchholz wrote:
Hi Chris (or others),
Here's a fix for a little bug I introduced in jdk7-b94
http://bugs.sun.com/view_bug.do?bug_id=6955504
6955504: (str) String[Builder/Buffer].append(char[],int,int) throws
OutOfMemoryError in b94
http://cr.openjdk.java.net/~martin/webrevs/openjdk7/oome2aioobe/
I couldn't resist sneaking in a little optimization.
I'm a little disappointed that I broke a JCK test, but that this was
not noticed by openjdk QA until several builds later. I'm thinking
it's part of the job of integrators to ensure __NO__REGRESSIONS__
and that includes jck tests (which are much easier for integrators
to run than for external contributors).
The JCK tests are NO easier to run for anyone, it only gets easier if
you have repeatedly run it over and over.
But the JCK team is making progress to make it easier, and we will try
and run it more often.
In defense of the integrators (who are not the ones that created the
regressions), they are doing the best they can
to find all our mistakes given the resources they have.
* If they don't integrate quickly enough, people complain that
their favorite change did not make it in
* If they integrate too quickly, people complain about too many
integrations and the regressions created by quicker integrations
* If they find regressions and hold up an integration, people are
upset that all the other changes are being held up
* If the integration goes smoothly, silence. Kind of a thankless
job, like a postal worker,... oh dear... please be nice to our
integrators.;^}
And it is impossible to guarantee no regressions given the time and
resources available to an integrator.
Avoiding regressions is a shared responsibility we all have.
In this case, we need to figure out how to get regular runs of JCK
into the integration process.
-kto
Martin
