On Friday 29 November 2002 18:26, Manuel Amador wrote: > El jue, 28-11-2002 a las 18:06, Michael Kreitzer escribi=C3=B3: > > 'cause it all starts when I click on the shortcuts on my desktop that= are > > URL's for either sftp connections or smb connections. That's when th= e > > fam binary itself starts using 100% CPU... I don't kno wmuch about kd= e's > > architecture so I'm sure it could be something else... I'm just fishi= ng > > for answers. > > and it enters a strange tight loop. All applications which are > subscribed to the fam instance experiment odd behavior meanwhile. I > disabled fam for that reason. > > But what bothers me is that, if I do dd if=3D/dev/zero of=3Dfile, and I= have > a konqueror window open, fam provokes a situation in konqueror, where > konqueror updates a zillion times. FAM should send update events to > konqueror at a reasonable pace (half a second maybe?)! Or maybe the See other mail. Configure FAM to only use STAT. > subscribing application should be able to tell it how often it needs to > know about updates (And the default should still be half a second). Half a second is way to much when the only solution possible is polling o= ver=20 NFS. The application would need to check for remote mounts (OK, KDirWatch= =20 already does this). Better let a tool like FAM decide on this. Aside from that, an application usually always wants events "as fast as=20 possible". So there's no need for applications to specify any time interv= als.=20 Using a kernel service like dnotify, the make the sense at all. > And FAM also causes unmountable removable devices. Even when I've > closed all Nautilus/Konqueror windows, I can't eject/unmount media > because FAM insists on keeping track of files in it to see if they've > changed. Absurd! FAM should let go when an unmount is triggered by the That shouldn't happen. When you kill applications doing file watching wit= h=20 FAM, FAM will see the broken IPC and finish watching of all files for tha= t=20 application: for dnotify, this means closing the corresponding files. The= =20 watching app can be *every* KDE application using a KFileDialog, for exam= ple.=20 So better shut down KDE :-) > user/system. On Mandrake, which has supermount, FAM causes horrible > CD-ROM errors (you eject the CD, CD is gulped again, if you manage to > snatch it before the drive closes, you still have /mnt/cdrom listing > files from the CD-ROM which isn't even there anymore). Hmm. Does a STAT of files in unmounted directories make supermount to do = a=20 mount? That's bad. This is a problem with KDirWatch (without FAM), too. So KDE *needs* to be aware of unmounting? Should we supply an "umount" wrapper with KDE doing a DCOP broadcast? Josef > > luck, > > >> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to > >> unsubscribe << >> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<