On Wed, Nov 12, 2008 at 6:33 PM, Christian López Espínola <[EMAIL PROTECTED]> wrote:
> Right now I updated again, and another fatal error occurred when > starting ArgoUML (stacktrace later). That was me. I apologize. I discovered the problem soon after the commit, but got interrupted before I could commit the fix, so things were probably broken for at least an hour. >From the point of view of getting the build fixed as quickly as possible, direct mail to the offending developer is probably the best method if you can identify them (SVN history is good for this). Mail to the list is good for shaming the offender and making them less likely to screw up in the future, but since list email is often filtered, it may get a less timely response. > I'm wondering if we are following any basic best practices before > commiting, or maybe I'm using a strange configuration (recently I > switched from branch/0.26.x to trunk again). > If I'm not doing anything wrong, we should be more carefully about > testing before committing. We should try to launch unit tests, or at > least, try to launch the application. Well, "best practices" is a bit of a misnomer. Practices can always be improved. The project rules are pretty simple: 1. don't break the build 2. don't break the tests 3. don't break the application With sufficient test coverage #2 & #3 are equivalent, but ArgoUML is a long way from that. Doing work in small, independent, isolated increments and running a complete test suite before committing every change will almost guarantee that we don't run into problems. In practice however, I, as well as others, I suspect make tradeoffs of completness vs expediency. If I think it's a low risk change (Javadoc update, simple refactoring), I might commit without testing and count on the Hudson continuous integration framework to catch any problems. If I'm working on multiple things at the same time, I might count on Mylyn's context tracking along with my own oversight to distinguish multiple, hopefully indepenent, change sets as I commit them. If I screw up again, as I'm almost certain to, send me an email directly (and CC the list for good measure) and I'll look at it as soon as possible. In this particular case, the problem was fixed long before I saw the email, but one never knows what the future holds. Tom --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
