Re: [Opensync-devel] python plugin confusion

[Mark E. <ma...@mp...>]

On Sun, 2010-08-22 at 00:17 +0200, Daniel Gollub wrote:
> On Sunday, August 22, 2010 04:09:35 am Mark Ellis wrote:
> > In short, the problem seems to be in OSyncList memory allocation, which
> > uses GSlice. I've pretty much figured out that the code using OSyncList
> > and therefore GSlice runs through once ok, but falls over the second
> > time around. If I had to guess, I would say that something is
> > reinitialising the slice allocator unsafely in another thread, but I'm
> > guessing wildly.
> 
> Could you provie for this issue a small demo application using OSyncList to 
> reproduce the problem? I would look into that then ..
> 

Hi Daniel

I don't believe it's OSyncList specifically, but instead something to do
with threading and initialisation of the slice allocator.

If you want to reproduce it, just set up a sync group containing the
python sample plugin, all latest svn, and run a discover on it with
osynctool, you'll get a sugfault. Now run it again with
G_SLICE=always-malloc and it won't segfault. It won't work either, but
that's an entirely different problem !

I've done a lot of analysis, rebuilding libopensync and glib with extra
logging, and at some point the slice allocator is getting a completely
bizarre and invalid address to allocate from. This address is then
passed out into an OSyncList allocation request and segfaults.

I'm going away for a couple of weeks today, but I'll try and sort out my
notes and send them on.

Mark