Re: [Tsp-devel] EIF TSP: tsp_consumer_request_sample_destroy

[Eric Noulard]

2007/2/28, Stephane GALLES <address@hidden>:
> At the end of my Ruby consumer example I used to do :
>
> consumer.request_sample_destroy
> consumer.request_close
>
> But I had to comment out the last line (request_close), because,
> sometimes, during the request_close, the tsp_stub_server starts
> throwing ERROR traces in the terminal, saying something like
> "I can't find the session you're trying to close !" and...coredump !

This core dump is a bug of the stub server ...

> Now, I didn't investigate at the time, coz I really didn't think
> that it could be in the core code of the provider (I really thought
> it could be my Ruby lib, or maybe the xmlrpc layer).
>
> I'll try to reproduce this behaviour. It really looks like
> ->sometimes<-, the request_sample_destroy get ride of the session

> before the request_close. Just an hypothesis, but it may have
> something to do with the session garbage collector when a socket
> is broken (I don't even know if this part of the code still exists though).

Yes I confirm :((
The garbage collector thread wrongly send request_close
on a "sampled_destroyed" session.

I am the guilty guy :[[

I added some time ago a automaton for "session state handling"
mainly to verify that you cannot jump from request_open
to request_sample_init or whatever forbidden request sequence.

This is cleaner than before and correct some bug
but introduce a new one in this case.
I'll try a fix tomorrow and let you know.

--
Erk