I switched our real project to forkmode="perTest" and found that it runs even slower (much slower), so it's clearly doing better than one VM per test. (I haven't ruled out something like one VM per class, though.) I won't have time to play more with the code until Monday, but I will look at it then.
Laura On Fri, Jul 23, 2010 at 1:10 AM, Stefan Bodewig <bode...@apache.org> wrote: > On 2010-07-22, Laura Dean wrote: > > > My team has a unit test suite that ran in about a minute and a half > > with ant 1.6.5 (and 1.7.1), but now takes over 4 minutes with ant > > 1.8.1 (or 1.8.0). Has anyone else had this problem? The closest I've > > found via google is here, but it doesn't shed much light on the > > subject: > > > http://netbeans.org/bugzilla/show_bug.cgi?id=182263 > > Ant 1.8.0 changed some parts of the process execution logic to adapt to > problems with hanging grandchild processes on Windows. This changes > have a side-effect (which we cannot seem to avoid) that causes any > forked process to use up at least about half a second. With 1.8.1 the > same logic changes have been applied to other OSes as well. > > So if you build process was forking a lot of small processes, a major > slowdown is to be expected. > > > We're running junit with fork=yes and forkmode=once. > > forkmode="once" should just create a single new process and you'd only > be paying the overhead once - not for each test. This is the first > thing I'd look into, makesure there really only is a single VM forked. > > > For the curious, I've created a simplified project here: > > http://web.mit.edu/lgdean/Public/test-project.zip . > > I'll look into it. > > Stefan > > --------------------------------------------------------------------- > To unsubscribe, e-mail: user-unsubscr...@ant.apache.org > For additional commands, e-mail: user-h...@ant.apache.org > >
