Re: [Tsp-devel] Blackboard problem !

[Eric Noulard]

2009/4/15 Frederik Deweerdt <address@hidden>:
>
> I did reproduce the problem on x86 full 32 bits... So it's seems that
> the bug has always been present on 0.8.3. It is unclear how it slipped
> through as bb_simu + tsp_bb_provider + tsp_stdout_client is a classic
> test combo.
>
> I didn't take time to dig deeper, but the cause of the problem in
> tsp_data_sender is that dimension is 0, so the encoder correctly
> returns 0.

Fred I'm taking a look at this and I do not quite understand what you are
talking about?

Where is the dimension set to 0 ?
The only solution for setting encoding dimension to 0 would be to have
group->items[i].nelem == 0

at tsp_data_sender.c:332

which is weird?
There should be some memory corruption leading to that.
If someone gets its hands on reproducing this please valgrind it
in order to have more localized informations.

However since Olivier said:
>>>>
I have done the test again with 0.8.3 and it appeared that:
       - I didn't reproduced this bug with these softs (bb_simu,
tsp_bb_provider, tsp_stdout_client)
>>>>

I bet the memory corruption is not *always* hapening....
However since I don't think that the missing bb_lock case for SHADOW_BB
can explain this I did file a reminder bug
https://savannah.nongnu.org/bugs/index.php?26304


-- 
Erk