# Bug Description In a multi-threaded environment, the Win32 Python3000 built-in function "print" may give the output several times. # How to Reproduce: import threading event = threading.Event() class Test(threading.Thread): def __init__(self, ord): super().__init__() self.ord = ord def run(self): event.wait() print('Hello, world!', self.ord) threads = tuple(map(Test, range(8))) tuple(map(lambda thread: thread.start(), threads)) event.set() tuple(map(lambda thread: thread.join(), threads)) # EOF # Problem Observed [The first run, 0 is doubled] Hello, world! 0 Hello, world! 0 Hello, world! 1 Hello, world! 2 Hello, world! 3 Hello, world! 4 Hello, world! 5 Hello, world! 6 Hello, world! 7 [the second run, 1 and 7 are doubled] Hello, world! 1 Hello, world! 1 Hello, world! 2 Hello, world! 3 Hello, world! 4 Hello, world! 5 Hello, world! 6 Hello, world! 7 Hello, world! 7 Hello, world! 0 # Expected Result Each thread gives ONE AND ONLY ONE output. OR State this as The Expected Behavior, document it and ask the user to write something such as critical section.

The TextIOWrapper class is not thread-safe, because it calls [the equivalent of] self.buffer.write(self.pending_bytes) before clearing self.pending_bytes. Of course the write() function will release the GIL, and another thread may send the same pending_bytes again. Patch is attached. Antoine, can you review it? (_pyio.TextIOWrapper does no buffering.)

The patch looks ok to me. It should be noted that TextIOWrapper was not designed to be thread-safe at all (neither the Python nor the C version); but admittedly it is more common to write() than to read() from multiple threads, so fixing this particular case makes sense.

fixed with r74581, r74582 and r74583 (in: trunk, py3k and release31-maint)