Vladimir, I think we are in agreement: (1) bugs are not a reason to delay a release, (2) we need better testing.
Can you please log jira cases for the 2 testing improvements and 1 bug cited in your message. By the way I am ready to check in https://issues.apache.org/jira/browse/CALCITE-507 right after the release, and will then start thinking about how to achieve https://issues.apache.org/jira/browse/CALCITE-579. Julian On Sat, Apr 11, 2015 at 11:42 AM, Vladimir Sitnikov <[email protected]> wrote: > Regarding 'half of the world', I believe it is not the last time we are > going to face the issue. > > I think it makes more sense to go ahead (release 1.2 as is), and start > using CI to test such issues (e.g. to have CI runs in different time zones). > > Regarding 'releasing non-buildable' versions, I do not like that in > general, however I think we never can give 100% guarantee, so it is fine to > discuss/agree on the case by case basis. > > For instance, Calcite 1.2 uses identityHashCode as _unique_ identifiers of > statements. > Should I elaborate why that is inherently broken? > Not sure if we should veto 1.2 due to the defect ('calcite executes wrong > sql'), but that is clear example of 'never releasing due to fixes here and > there'. > > Well, those kind of 'random' issues can be captured by durability testing. > It would be nice to have a long running test (e.g. a couple of days) to > test against memory leaks/random failures. > > Regards, > Vladimir Sitnikov
