I've been using Ant to build a fairly large web application (about twelve thousand .class files and various other things in a multi-phase build) for several years. We've generally been pretty happy with Ant performance. Recently, however, we noticed that if we ran Ant 1.7 on the same build configuration the process was very noticeably slower, and involved long (many seconds) periods of high CPU load. Through some poking it appeared that the <copy> task was incurring the load, which seemed odd.
So today I ran a build with JMP (Java Memory Profiler) just out of curiosity, after having downloaded and built the source. At the point of the large copy, I noticed that the "contains" method on java.util.ArrayList was being called very many times, and that that was resulting in a tremendous number of calls to compare Resource instances. It seems to me that the code in Union.getCollection() is the culprit. For reasons unknown to me it builds up its result with an obviously n^2 loop to avoid duplicates in the result collection. By experimentally replacing the "ArrayList" with "LinkedHashSet", the excessive CPU load is completely cured. I realize that there might be some issue using LinkedHashSet, because it's from a fairly recent Java version, but the code could also work by just keeping an explicit java.util.HashSet around in parallel with the ArrayList being assembled. For comparison, a build on an already-built source tree (i.e., something that just runs through the <copy> task without ending up doing anything) takes 15 seconds with my "fixed" Ant 1.7, 45 seconds without the fix, and 15 seconds with Ant 1.6.5. (My machine is a fairly modern dual-processor P4 with lots of memory, running Linux.) (If this is a known and solved problem, my apologies; I can't find any mention of it on the web.) -- [ I love pre-moistened towelettes ] Mike McNally -- [EMAIL PROTECTED] --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
