Created attachment 49948 [details]
An example of such a file
A bit of context : The french ministry of economics (Bercy, for the french of us) provides some useful forms to local administrations (towns, counties, and the such). Those documents are provided in RTF format, and they've often been problematic : random OOo (in the old times) crashes, formatting issues, etc.
Still, the problems seemed to become scarce over the time, and I had a lot less issues with LibO 3.3. Sadly, LibO 3.4 might have included a regression : the very same file that somehow worked on 3.3 freezes 3.4 (blank screen right before the "File import" progress bar completes, 50% CPU usage forever)
It's easy to reproduce, at least on a Windows machine (yet to test on other OSes) : run LibreOffice 3.4, open the given file, and enjoy. The very same process on 3.3 gives us rather quickly a readable content (progress bar goes unnoticed).
LO 3.4.3 OOO340m1 (Build:302)
Ubuntu 10.04.3 x86
Linux 2.6.32-34-generic Russian UI
I reproduce the problem under Ubuntu 10.04 FR (x86_64) :
- LibO 3.4.3 : CPU at 100%, I need to kill soffice
- master : crash with the following message:
Segmentation fault (core dumped)
- LibO 3.3.3 packaged by Ubuntu (PPA) : opens the file without problem.
Assigned to Miklos. Please feel free to reassign to default if you can't work on this bug. :-)
Best regards. JBF
Thanks, will have a look.
Created attachment 52768 [details]
Error followed backtrace
I put the error followed by the bt. (made on Debian Wheezy x86-32 with sources updated this day from git branch master).
Hope it could help.
Thanks, the bug is probably in DomainMapper_Impl::SetFieldFFData() - will have a look once I find some spare time soon. :)
Yes, it crashes at this line :
FieldContextPtr pContext = m_aFieldStack.top();
But I don't know why m_aFieldStack is empty. I suppose it's better to know why it's empty than to put a check of non emptiness of this field in the DomainMapper_Impl::SetFieldFFData()
In fact the problem was earlier, the importer tried to parse fly frames inside tables, which it should not do.
should fix the issue. If I were at it, I also fixed the import of the 'è' character and the state of the checkboxes (the filter marked them as checked, while they were not).
No crash in my current build of the master (Ubuntu 10.04 x86_64).
Thank you very much Miklos :-)
No crash confirmed with LO 3.5.1. Closing. Thank you.
Should have added keyword regression (comment #2).
Best regards. JBF