Posts Tagged ‘carbon’

Data Loss through WindowServer Crashes in Leopard

Fortunately, Mac OS X put an end to those dreaded times when your Mac would crash at least once daily, with all unsaved work lost. That is, almost. At least for me, the issue is back after I upgraded to Leopard in fall, even if the crash frequency is “only” once every one to two weeks.

The symptom is always the same: when typing away in a Carbon (Rosetta?) application on my MacBook Pro, the screen freezes for a few seconds and then goes blue, while the WindowServer starts up again, presenting me with a login window. No need to say that the WindowServer takes all currently open applications with it, in effect trashing all unsaved work.

Digging in /Library/Logs/CrashReporter/ reveals that the crash always happens at the same location somewhere in CarbonCore:

Date/Time:       2009-03-10 17:00:46.770 +0100
OS Version:      Mac OS X 10.5.6 (9G55)
Report Version:  6

Exception Codes: KERN_PROTECTION_FAILURE at 0x00000000be255460
Crashed Thread:  0

Thread 0 Crashed:
0   ...ple.CoreServices.CarbonCore  0x94a47719 UCKeyTranslate + 363
1          0x929e8937 CGSUniCodeForKeyAndFlags + 152
2          0x928baee6 CGXCheckForHotKey + 499
3          0x9288b416 sPostContinuation + 1759
4          0x929de924 postAfterTapID + 266
5          0x929df88b postFilteredEventTapData + 496
6          0x92910ac7 _XPostFilteredEventTapData + 221
7          0x92810967 CGXWindowServer_server + 117
8          0x9289bc3b rendezvousHandler + 155
9          0x9288c70e CGXPostPortData + 158
10          0x9288c504 CGXRunOneServerPass + 556
11          0x92888b0f CGXRunOneServicesPass + 377
12          0x92893e9e CGXServerLoop + 161
13          0x92821fb9 CGXGetRootAdminCredentials + 0
14  WindowServer                    0x00001ff4 main + 24
15  WindowServer                    0x00001fbe start + 54

So far, I’ve only experienced the crash while I was working in Excel 2004 or Eudora, both Carbon PPC applications that run under Rosetta emulation.

I filed the bug with Apple on 2008-12-05 as rdar://6423987, also see the OpenRadar copy. It was acknowledged as a duplicate of rdar://5432883; Apple says engineering is still investigating.

It seems this bug is somewhat widespread and old (at least going back to December 2007). I could find the following reports that are probably describing the same issue:

Now this sort of bug is extremely exasperating, as it is bound to lead to data loss. Have you experienced the same issue? Please comment below, and, most importantly, file a bug at in which you mention the original Radar ID (5432883) so that Apple gets an idea how widespread the issue actually is.

20 comments March 11th, 2009




Posts by Month