Hi antony, 2 different things: 1/ PlatformLogger was patched (doLog method) to avoid string operations (message formatting) for disabled logs (patch submiited on JDK8 and JDK7u): http://mail.openjdk.java.net/pipermail/jdk7u-dev/2012-April/002751.html
2/ I looked 2 hours ago on JDK7u AND JDK8 source codes and both still have: - log statements WITHOUT log level check : if (log.isLoggable(PlatformLogger.FINE)) log.fine(...); - string operations (+) in log calls that could be improved using the message format syntax (String + args): for example, avoid using PlatformLogger.fine(String + ...) in favor of using PlatformLogger.fine(String msg, Object... params) I reported in my previous mail several cases where the isLoggable() call is missing and leads to useless String operations but also method calls (Component.paramString() for example). Finally, I also provided other possible cases (using grep); maybe there is a better alternative to find all occurences of String operations in log calls. Regards, Laurent 2013/3/18 Anthony Petrov <anthony.pet...@oracle.com> > Hi Laurent, > > Normally we fix an issue in JDK 8 first, and then back-port the fix to a > 7u release. You're saying that in JDK 8 the problem isn't reproducible > anymore. Can you please investigate (using the Mercurial history log) what > exact fix resolved it in JDK 8? > > -- > best regards, > Anthony > > On 03/18/13 15:09, Laurent Bourgès wrote: > >> Dear all, >> >> I run recently netbeans profiler on my swing application (Aspro2: >> http://www.jmmc.fr/aspro) under linux x64 platform and I figured out >> that a lot of char[] instances are coming from String + operator called >> by sun.awt.X11 code. >> >> I looked at PlatformLogger source code but found not way to disable it >> completely: maybe an empty logger implementation could be interesting to >> be used during profiling or normal use (not debugging). >> Apparently JDK8 provides some patchs to avoid String creation when the >> logger is disabled (level). >> >> However, I looked also to the sun.awt code (jdk7u repository) to see the >> origin of the string allocations: >> XDecoratedPeer: >> public void handleFocusEvent(XEvent xev) { >> ... >> * focusLog.finer("Received focus event on shell: " + xfe); >> * } >> >> public boolean requestWindowFocus(long time, boolean timeProvided) { >> ... >> * focusLog.finest("Real native focused window: " + >> realNativeFocusedWindow + >> "\nKFM's focused window: " + focusedWindow); >> *... >> * focusLog.fine("Requesting focus to " + (this == toFocus ? "this >> window" : toFocus)); >> *... >> } >> >> XBaseWindow: >> public void xSetBounds(int x, int y, int width, int height) { >> ... >> * insLog.fine("Setting bounds on " + this + " to (" + x + ", " + >> y + "), " + width + "x" + height); >> *} >> >> XNetProtocol: >> boolean doStateProtocol() { >> ... >> * stateLog.finer("**doStateProtocol() returns " + res); >> *} >> >> XSystemTrayPeer: >> XSystemTrayPeer(SystemTray target) { >> ... >> * log.fine(" check if system tray is available. selection owner: >> " + selection_owner); >> *} >> void addTrayIcon(XTrayIconPeer tiPeer) throws AWTException { >> ... >> * log.fine(" send SYSTEM_TRAY_REQUEST_DOCK message to owner: " + >> selection_owner); >> *} >> >> XFramePeer: >> public void handlePropertyNotify(XEvent xev) { >> ... >> stateLog.finer("State is the same: " + state); >> } >> >> However I only give here few cases but certainly others are still >> present in the source code; maybe findbugs or netbeans warnings could >> help you finding all of them. >> >> I advocate the amount of waste (GC) is not very important but String >> conversion are also calling several toString() methods that can be >> costly (event, Frame, window ...) >> >> Finally, I ran few grep commands on the sun.awt.X11 code (jdk7u) and you >> can look at them to see all string + operations related to log statements. >> >> PS: I may help fixing the source code but I have no idea how to >> collaborate (provide a patch ?) >> >> Regards, >> Laurent Bourgès >> >
