On Fri, Jul 23, 2010 at 9:48 PM, Bruno Wolff III bruno@wolff.to wrote:
I see this is just the general bind umount. So probably the problem is something pinning one of the bind mounts. That should go away on exit and then the deferred umounts should all get processed. I need to review the original bug to check that, but I think that is what I remember now.
This is about exactly what is happenning when an exception bubbles up and a bind mount has already been set, especially when we get into buildTransaction() this will always occur if we don't use a lazy umount.
To reproduce the problem reliably, just enter a nonreachable repository in a kickstart file and you should get a screen full starting with unable to find repository which is normal, to twice trying to unmount and nopt being able to plus a traceback because of it.
I'll look at getting this committed tomorrow after I have had some sleep. I suspect we probably want to keep the bug open as this isn't a fix of the root cause and as long as we think that is possible / practical we should keep it open.
Sounds like a fine idea.