In 3.2.2 and 3.2.3 on linux64, when running the following code. 

    try:
        print("Writing to /dev/full")

        with open("/dev/full", "w") as f:
             f.write("Write to full device")
    except:
        print("Catch, file closed?")

Using

    strace -e close ~/usr/python3.2.3/bin/python3 test.py
    ...
    Writing to /dev/full
    Catch, file closed?
    close(3)                                = 0

The file descriptor being used (3) to attempt writing to /dev/full is not closed until the process exits.

I expected this to be closed when leaving the context manager.

Originally discovered on http://stackoverflow.com/questions/13649330/what-happens-to-file-descriptors-in-python-3-when-close-fails

>>> f = open("/dev/full", "wb", buffering=0)
>>> f.write(b"Write to full device")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OSError: [Errno 28] No space left on device
>>> f.close()
>>> f.closed
True
>>> f = open("/dev/full", "wb")
>>> f.write(b"Write to full device")
20
>>> f.close()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
OSError: [Errno 28] No space left on device
>>> f.closed
False


Python 2 has the same behavior using io.open.

The bug only in C implementation.

>>> import _pyio
>>> f = _pyio.open("/dev/full", "wb")
>>> f.write(b"Write to full device")
20
>>> f.close()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/home/serhiy/py/cpython/Lib/_pyio.py", line 732, in close
    self.flush()
  File "/home/serhiy/py/cpython/Lib/_pyio.py", line 1121, in flush
    self._flush_unlocked()
  File "/home/serhiy/py/cpython/Lib/_pyio.py", line 1128, in _flush_unlocked
    n = self.raw.write(self._write_buf)
OSError: [Errno 28] No space left on device
>>> f.closed
True

Here is a patch which calls close() on underlying stream even if flush() raises an exception.

I am not sure that I correctly set a context exception. There is no other examples in the code.

There are a lot of people in the nosy list already. Does anyone have enough experience with exception machinery to review the patch (especially C part)?

Patch updated with Benjamin's nit.

New changeset b0602a1eb3f6 by Benjamin Peterson in branch '3.3':
call close on the underlying stream even if flush raises (closes #16597)
http://hg.python.org/cpython/rev/b0602a1eb3f6

New changeset 142012e47c3b by Benjamin Peterson in branch 'default':
merge 3.3 (#16597)
http://hg.python.org/cpython/rev/142012e47c3b

Here is a patch for 3.2.

Here is a patch for 2.7.

I think we can leave 3.2 alone.

New changeset b6ff6ac1f049 by Benjamin Peterson in branch '2.7':
call close on the underlying stream even if flush raises (#16597)
http://hg.python.org/cpython/rev/b6ff6ac1f049

What is the peculiarity of 3.2?

About Misc/NEWS. Actually fixed close() methods of: Python implementation of BaseIO (C implementation already do right things), C implemettation of Buffered(Reader|Writer|Random) (Python implementation already do right things) and both implementations of TextIOWrapper.

New changeset 54372f38932e by Benjamin Peterson in branch '3.3':
improve message (#16597)
http://hg.python.org/cpython/rev/54372f38932e

New changeset faaac6ceffff by Benjamin Peterson in branch 'default':
merge 3.3 (#16597)
http://hg.python.org/cpython/rev/faaac6ceffff

New changeset a057d9985fb8 by Benjamin Peterson in branch '2.7':
add news note (#16597)
http://hg.python.org/cpython/rev/a057d9985fb8

Thank you for all commits, Benjamin.

What is the peculiarity of 3.2?

2012/12/20 Serhiy Storchaka <report@bugs.python.org>:
>
> Serhiy Storchaka added the comment:
>
> Thank you for all commits, Benjamin.
>
> What is the peculiarity of 3.2?

Applying bugfixes is optional. It should move to security-fix mode soon.