[filtered] Re: Crash in revision 1030637

[David Nolden]

Am Sonntag 04 Oktober 2009 11:55:27 schrieb Christoph Bartoschek:
> The question is how to handle such a situation. I've learned that
>  .kdevduchain is unreliable and has to be first deleted before reporting a
>  bug.
> 
> However if you say that each time deleting helps we have a real bug then
>  there should never be a situation where one should delete the directory
>  because the file format changed.
~/.kdevduchain is not unreliable, it is just a storage place that makes all 
duchain problems persistent.

The only times when the storage is at fault is when there is an assertion 
somewhere in "itemrepository.h", and that's nearly always due to system 
crashes and file corruption. There is a protection against this, but that 
doesn't work for example when the kernel crashes or in case of power-loss.

The problem is that the duchain is so complex, that it's very hard to reliably 
reproduce problems that appeared across many runs.

The crash you did see actually might be a 'valid' situation that just happens 
very rarely, and the "rm ~/.kdevduchain" changed the situation again.

The most useful action to do in that case would have been looking exactly at 
the crash and checking why it happened (was the context zero, was the 
declaration zero, was something unitialized, etc.)

Greetings, David