[zeromq-dev] need advice to debug an alternative to CURVE in a multithread test

[Laurent Alebarde]

Hi Devs,

As some of you know it, I am working on an experimental alternative to 
CURVE: PARANO (because I am paranoïd of course ;-) ). Typically, I 
started by dupplicating the CURVE frame as a clone, check everything was 
right, and then started step by step to change produce/process methods. 
The equivalent of test_security_curve.cpp, test_security_parano.cpp is 
working. Then I made a multithread test based on test_proxy.cpp for 
curve: test_concurrency_curve.cpp which works well, and then port it to 
PARANO: test_concurrency_parano.cpp.

With one client it is ok, but with several ones, I have some 
unreproduceable destruction/re-construction of the mechanism. For 
example, with 3 clients, It can be once the mechanism for client n°2, 
and the other time no one, and another time two of them, or twice the 
one of the same client. When the mechanism_t object is destructed 
(destructor method called), the handcheck is process again of course, 
and finally, the messages are sent normally. So if I don't care, it 
delivers what it has to deliver. But it is not clean.

When I use the multithreaded debugger, it seems to be worse, with more 
mechanisms destruction/re-construction. My question is:
*Considering CURVE, Is there some time delay I should deactivate when 
some threads are paused ?*


Cheers,


Laurent.


P.S.: I plan to publish and propose PARANO when I will create my 
company. The RFC is in progress.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.zeromq.org/pipermail/zeromq-dev/attachments/20131025/86ada877/attachment.htm>