[zeromq-dev] Problems with errno on Win32

[Brian Granger]

Martin,

> One of the ideas I have is this: Python is single-threaded, right?

No, not really.  It really does have full blown native threading
support.  But it also has the global
interpreter lock, which restricts only 1 thread to executing Python
code at a time.  But other
threads that don't call into Python itself can run concurrently.
Thus, I am guessing that Python
does link against the multi-threaded CRT.

Cheers,

Brian

> Thus it
> uses single-threaded CRT. Then it gets linked dynamically with libzmq.dll
> whitch uses multi-threaded CRT. Now, when libzmq uses errno, which one it
> is? One defined in ST CRT or the one from MT CRT?
>
> Just an idea though.
>
> Martin
>
> Brian Granger wrote:
>>
>> Martin,
>>
>> I am in class most of the day today, but will try this later.  One
>> comment though: my feeling is that
>> if the problem I am seeing is related to thread safety, the conflict
>> is between the IO thread
>> and the app thread.
>>
>> Cheers,
>>
>> Brian
>>
>> On Thu, Feb 18, 2010 at 4:36 AM, Martin Sustrik <sustrik at 250bpm.com>
>> wrote:
>>>
>>> Hi Brian,
>>>
>>> Regarding the discussion we had yesterday about thread-safeness of errno
>>> on Win32: I've did a check today and it seems that address of errno is
>>> different for each thread - as it should be. I've run the test on WinXP.
>>> How does it look like on your system?
>>>
>>> To check it, just open thread.cpp and following piece of code on line 53:
>>>
>>> printf ("%p\n", (void*) &errno);
>>>
>>> Then run any 0MQ test with multiple I/O threads. Each I/O thread will
>>> print out the address of it's errno. Are they all the same?
>>>
>>> Thanks.
>>> Martin
>>> _______________________________________________
>>> zeromq-dev mailing list
>>> zeromq-dev at lists.zeromq.org
>>> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>>>
>>
>>
>>
>
>



-- 
Brian E. Granger, Ph.D.
Assistant Professor of Physics
Cal Poly State University, San Luis Obispo
bgranger at calpoly.edu
ellisonbg at gmail.com