Subj : Re: I write a multithread program with pthread, and meet a program
To   : comp.programming.threads
From : David Butenhof
Date : Tue Apr 19 2005 07:32 pm

Giancarlo Niccolai wrote:
> Loic Domaigne wrote:
> 
>>It's only a theory at that point... But it looks like a bug in the
>>libstdc++ or the libpthread, when a thread is being cancelled while
>>writting on cout.
>>
> Uhm, AFAIK write() system call is a cancellation point, and cout << is
> supposed to write somewhere. So it *ought* to be a cancellation point
> unless posix or c++ standards states otherwise. There is not any statement
> in this sense to may knowledge, but I may be outdated...

There is no standard anywhere that says whether cout operations shall or 
shall not be cancellation points, because POSIX doesn't talk about C++ 
and C++ doesn't talk about threads. This is completely open and 
unspecified. Any assumptions about this sort of issue is completely 
nonstandard and nonportable.

The problem is that a C++ library that's not written with consideration 
for the fact that it might be running in a threaded process is that 
write() IS a cancellation point (on a conforming POSIX system) unless 
the caller disables cancelability around it. So this may well be what's 
happening. Whether it's a "bug" in C++ or just a fact of life you need 
to survive depends on the intent of the C++ developers with respect to 
thread compatibility. That is, they may intend that you disable 
cancelability around any area of code that might do this; and while you 
can argue which makes most sense there's no objective authority to say 
they're wrong.

-- 
Dave Butenhof, David.Butenhof@hp.com
HP Utility Pricing software, POSIX thread consultant
Manageability Solutions Lab (MSL), Hewlett-Packard Company
110 Spit Brook Road, ZK2/3-Q18, Nashua, NH 03062

.