This is the mail archive of the mailing list for the libstdc++ project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

reentrancy and cout/stdout


I am tracking down a problem in a multi-threaded run-time in which it 
appears that
the FILE * associated with cout is cached internally to libio.  This is 
with the
libio shipped with gcc 2.95.2.  I do not have a working gcc 3.0 to 
compare behaviors.

RTEMS uses the newlib reentrancy structure on a per-thread basis. 
Each thread gets its own stdin, out, and err structures.  The symptom is
that "task2" is using the correct reentrancy structure but tries to fwrite()
using the FILE * associated "task1".  Since the reentrancy structure and 
belong to different tasks, there is an accounting conflict.

I think the underlying cause is that somewhere in the bowels, libio is
grabbing a FILE * pointer and using its own local copy rather than
repeatedly using "stdout".  In this system, stdout is a macro defined as:

stdio.h:#define stdout  (_impure_ptr->_stdout)

So if libio assigns that to an internal variable, then that stdout ends up
being used by all tasks.

Does this description make sense?   Is this the behavior in v3?  Is there
a way to fix this on either side?


--joel sherrill

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]