w00t indeed. This just saved me serious time this morning already. On Wed, Jun 10, 2009 at 6:35 PM, Scott Blum <sco...@google.com> wrote:
> w00t!! > > > On Wed, Jun 10, 2009 at 5:47 PM, Freeland Abbott <fabb...@google.com>wrote: > >> As of r5537, my no-change "ant build" takes 1:55 instead of 19:43, and >> there's still some easy work to do, albeit with obviously diminishing >> returns >> >> Most of that difference is due to a rather annoying timestamp >> consideration with directory entries in jars; my patch introduces a new Ant >> task, LatestTimeJar, to resolve it. >> >> The issue is---was---that in general, we jar both >> .../src/com/google/gwt/.../Foo.java and also >> build/out/.../com/google/gwt/.../Foo.class. The jar file will have one >> directory entry for "com/", the existence of which is actually important to >> GWT as Scott pointed out in the first-round review comments. But the two >> directories have different touch dates, and we archived the first-named, >> which was usually from .../src/..., with an "old" date by svn. The second >> build would therefore notice that the *second* instance of "com/" was >> newer than the archived "com/", and therefore jar it again. (Because we did >> "updates," the entry would have been new after that second cycle. In some >> cases, notably the servet API classes in alldeps.jar, we had up to four such >> duplicates, though.) Worse, everything downstream of that error also had to >> be redone... including the samples. >> >> >> >> >> > > > > --~--~---------~--~----~------------~-------~--~----~ http://groups.google.com/group/Google-Web-Toolkit-Contributors -~----------~----~----~----~------~----~------~--~---
