On 3/19/07, Steve Loughran <[EMAIL PROTECTED]> wrote:
Xavier Hanin wrote: > On 3/19/07, Steve Loughran <[EMAIL PROTECTED]> wrote: >> >> Xavier Hanin wrote: >> >> > I've just checked in a build.xml doing the packaging trick, hope it >> will >> > fix >> > your gump build. >> >> thank you? >> >> > >> > - Xavier >> > >> > Sorry for the trouble, >> >> No, this is exactly the kind of success that Gump is designed for: to >> catch a regression on an OSS project, before you even think about >> releasing it. We have just managed to identify and fix a problem before >> it hit the field -which is why any project that can build under Gump, >> should be registered under gump. Its a lot easier to fix a regression >> when it went in the previous week, than when it only shows up in >> shipping code, because that means it takes a whole release cycle to get >> the fix in, and it runs the risk of breaking somebody else if they >> actually depend on the regression's behaviour > > > Agreed, but I even prefer avoiding the break :-) But you're right, this > continuous integration is really a good thing, it adds even more tests to > Ivy, which is really good! > If you look at what breaks Ant in the field, its things gump doesnt test -classloader and classpath setup (gump takes over) -bad system installations (ant.sh from ant1.6 via jpackage but ANT_HOME set to Ant1.7, etc) -windows systems with spaces and quotes in classpaths -the shell scripts, again, mostly on windows. so be aware that Gump doesnt catch thesel; you still need a release cycle Thanks for the tip. We'll take care of this kind of things for our
release. - Xavier
