On Tue, May 14, 2002 at 02:06:28PM +0200, Steffen Neumann wrote:
> Here is another traceback of a venus crash,
> no data was lost, for debugging purpose only.
> 
> The user did (afaik) not do any serious writes,
> but used software installed on /coda  

This is that annoying startup bug I've been trying to fix. Looking at my
local copy of the source, the function where is assert is triggered
isn't even called anymore. Maybe I fixed it :) Probably not, I think I
might have only moved the problem to another location. And all of this
is because local-global repair is messing up local data structures. The
old code never noticed this and was simply doing CML optimization etc.
on objects that were moved into a temporary volume, leading to subtle
inconsistencies.

Nice log and gdb info, this stuff is hard to get.

> [ X(00) : 0000 : 13:51:48 ] fsobj::Recover: invalid fso (Dyn-Laptop-05, (0xffffffff.0xfffffffe.0x10)), attempting to GC...
> 0x256af508 : fid = ((0xffffffff.0xfffffffe.0x10)), comp = Dyn-Laptop-05, vol = 26283e88
>         state = Dying, stat = { 193, 5, 1020872166, 64, 0644, 0, File }, rc rights = 0

Temporary object with CML entries which has been removed, local-repair
never cleaned it up right. This info is very good, after reading this I
am starting to believe I was trying to fix this thing the wrong way.

Jan