Apache OpenOffice (AOO) Bugzilla – Issue 124421
crash on Mac if there are problems with network connectivity
Last modified: 2014-04-04 21:23:21 UTC
Created attachment 82856 [details] Diagnostic Report Using the 64 bit version of OpenOffice on Mac OS X the application crashes without network connectiviy after 2 minutes. Steps to reproduce: (1) End OpenOffice application (no OpenOffice process running) (2) Disconnect all networks (wireless and wired) (3) Wait some time (I don't know exactly how long) (4) Start OpenOffice and wait 2 minutes or do something within OpenOffice The diagnostic report attached to this bug report has the following entry: "terminating with unexpected exception of type com::sun::star::ucb::InteractiveNetworkGeneralException abort() called" After reconnecting networks the problem disappears. This behaviour has been produced with the current revison 1576748 from trunk, but appears with earlier revisions of the 64 bit version too.
I assume that the automatic check for updates is active and that this functionality triggers the crash as it tries to setup an HTTP connection. @Thorsten Wagner: Did you experience the same problem with AOO 4.0.1 or earlier versions?
The crash occurs reproducible after 35 seconds measured from starting OpenOffice. AOO 4.0.1 is not affected. I tried disabling automatic updates, unfortunately without success. The problem still exists and disappears after connecting to the network again.
Sorry, maybe my previous test was wrong: The problem disappears when disabling automatic updates now.
Some more precise information would help to find out whether other OS also are affected. @Thorsten Wagner (a) 4.0.1 still not affected or only no Update search done? (b) Does the crash appear when you do 'Tools -> Options -> OO -> Online Update -> [Search Now]'? (b2) If yes after what time?
(In reply to Rainer Bielefeld from comment #4) > Some more precise information would help to find out whether other OS also > are affected. > > @Thorsten Wagner > (a) 4.0.1 still not affected or only no Update search done? > (b) Does the crash appear when you do 'Tools -> Options -> OO -> Online > Update > -> [Search Now]'? > (b2) If yes after what time? (a) AOO 4.0.1 seems to be not affected. (b) The crash occurs about 30 seconds from start of OpenOffice without doing anything. I didn't try to originate a manual search. I will give a feedback about manual search after investigation.
(1) With automatic updates turned on the crash occurs about 30 seconds after starting OpenOffice. (2) Without automatic updates there is no crash, even when originating a manual search (see attached screenshot).
Created attachment 82944 [details] Screenshot of message after manual search (no network connectivity, no crash)
I got access to a Mac OS X 10.9.2 system and I was able to reproduce the described crash. It occurs with AOO 4.1.0 Beta, but not with AOO 4.0.1 --> marking it as a regression. I will perform a further check with a new build of recent branch AOO410.
(In reply to Oliver-Rainer Wittmann from comment #8) > I got access to a Mac OS X 10.9.2 system and I was able to reproduce the > described crash. > > It occurs with AOO 4.1.0 Beta, but not with AOO 4.0.1 > --> marking it as a regression. > > > I will perform a further check with a new build of recent branch AOO410. The new build showed the same crash.
I checked this issue on Windows 7. Result: no crash. --> crash seems to be platform-dependent
Created attachment 83008 [details] patch to disable type_info generation for the OSX64 uno bridge Jürgen found out that the problem only happened when the dynamic loading of the typeinfo symbol for InteractiveNetworkGeneralException failed and the uno bridge had to generate the typeinfos themselves. Disabling the generation of synthetic typeinfos and handling the consequences gracefully seems to be a better approach.
Created attachment 83009 [details] patch to force the availability of type_info symbols for ucb exceptions The patch before allows the OSX64 bridge to handle missing type_info symbols of UNO exceptions gracefully. This patch forces the availability of type_info symbols that are thrown by the usb module. So the bridge-patch solves the general problem of missing type_info symbols by handling them gracefully. This patch solves the particular problem that having the type_infos of thrown exceptions is always good. So the both patches address different aspects of the problem that caused the reported crash.
grant showstopper flag because the root cause is deeper in the UNO bridge which is new for 64 bit and the clang compiler
Comment on attachment 83009 [details] patch to force the availability of type_info symbols for ucb exceptions it can't be bad to help the compiler to force the generation of the necessary type info. And it's harmless because never called.
Comment on attachment 83008 [details] patch to disable type_info generation for the OSX64 uno bridge looks ok so far, but we should try to find a more general fix for UNO to ensure that the rtti exception type info is always generated. And we should check if cppu::unoThrow is really necessary in this context or if it is better to throw the exception directly.
"hdu" committed SVN revision 1582359 into trunk: #i124421# disable generated type_info for the OSX64 uno bridge
"hdu" committed SVN revision 1582360 into branches/AOO410: #i124421# disable generated type_info for the OSX64 uno bridge
"hdu" committed SVN revision 1582365 into trunk: #i124421# force the availability of type_info symbols thrown by the ucb module
"hdu" committed SVN revision 1582368 into branches/AOO410: #i124421# force the availability of type_info symbols thrown by the ucb module
Fixed with the commits above on trunk and the AOO410 branch.
"hdu" committed SVN revision 1582709 into trunk: #i124421# use OSL_ debug helpers instead of tools DBG_ stuff
"hdu" committed SVN revision 1582710 into branches/AOO410: #i124421# use OSL_ debug helpers instead of tools DBG_ stuff
Verified on branch AOO410. Rev. 1583666
Reviewed successful with trunk revision 1582709