Community
Participate
Working Groups
Based on some comments in bug 420078, it became obvious we should have a "releng test" to look for regressions in "dirt produced" during a build. (I know I once "regressed" when a "file copy" became necessary). And, just as well generate the report each build, rather than have to remember to do it every once in a while. Best to do the "report" as part of the production build, since that's when the "aggregation directory" is in a known state. The "releng test" can operate on that report ... eventually making sure it was of (near) zero length ... but even before "completely clean", we can probably spot regressions by simply checking a "known length" and making sure it did not get larger.
I've created a "dirtReport" script, and will put its output under the usual "releng build logs" location, such as, under http://download.eclipse.org/eclipse/downloads/drops4/N20150226-0100/buildlogs.php http://git.eclipse.org/c/platform/eclipse.platform.releng.aggregator.git/commit/?id=5e3837e597e2ef42b924e153ef5803f0b584a75e I'll "let it run" a few days, before adding the releng test to check it's size, just to make sure it is "reliable". And to be explicit, the releng test won't be designed to fail, just because there is dirt (or, else it would always fail) .... but will simply fail if there ends up being more dirt, than in previous runs (i.e. regressions). Once we have "zero" dirt, then it will be a true "dirt test".
http://git.eclipse.org/c/platform/eclipse.platform.releng.git/commit/?id=233134310962ab4bfe494af3a6fe73b18f9d6bf8 includes similar "test of file size" for "unexpected comparator errors" log.
I should mention, the solution originally relied on "the best" URLConnection support found in Java 7, so increased the BREE from 1.5, to 1.7. But, then realized the "old" getContentLength (as opposed to getContentLengthLong) covers lengths up to roughly "2 Gig" and our "logs" should never get anywhere near that size. So, changed back to 1.5, and used the older "getContentLength()". Also, modified so that test allows different "download host" to be specified. It defaults to "download.eclipse.org" (always used during "production") but having the variable allows my "local test machine" to still work with that test. http://git.eclipse.org/c/platform/eclipse.platform.releng.git/commit/?id=1fc3fc909543ecf31e96f10bce20aa3862ea5820
Seems well in http://download.eclipse.org/eclipse/downloads/drops4/N20150302-2000/
