Here is a patch that implements correct __sizeof__ for C implementation of buffered I/O files: io.BufferedReader, io.BufferedWriter and io.BufferedRandom.

There is no patch.

Instead of putting this in MiscIOTest, you could use the relevant class-specific test cases.

Serhiy, I didn't analyze it too in depth, but why aren't the test cases using the __sizeof__ support work you implemented for issue15467?  I think these tests should be using the more exact method like your other '__sizeof__' patches.

> Instead of putting this in MiscIOTest, you could use the relevant
> class-specific test cases.

Thank you, good advice. Here is updated patch.

> Serhiy, I didn't analyze it too in depth, but why aren't the test cases
> using the __sizeof__ support work you implemented for issue15467?  I think
> these tests should be using the more exact method like your other
> '__sizeof__' patches.

Because struct has not codes for Py_off_t and PyThread_type_lock.

New changeset 8a02a93c803a by Antoine Pitrou in branch '3.2':
Issue #15487: Add a __sizeof__ implementation for buffered I/O objects.
http://hg.python.org/cpython/rev/8a02a93c803a

New changeset 1d811e1097ed by Antoine Pitrou in branch 'default':
Issue #15487: Add a __sizeof__ implementation for buffered I/O objects.
http://hg.python.org/cpython/rev/1d811e1097ed

New changeset 1102e86b8739 by Antoine Pitrou in branch '2.7':
Issue #15487: Add a __sizeof__ implementation for buffered I/O objects.
http://hg.python.org/cpython/rev/1102e86b8739

Thanks for the patch :)

On Sun, Jul 29, 2012 at 11:50 AM, Serhiy Storchaka <report@bugs.python.org> wrote:

>> Serhiy, I didn't analyze it too in depth, but why aren't the test cases
>> using the __sizeof__ support work you implemented for issue15467?  I think
>> these tests should be using the more exact method like your other
>> '__sizeof__' patches.
>
> Because struct has not codes for Py_off_t and PyThread_type_lock.

Of course it doesn't -- those are Python specific typedefs.  'PyThread_type_lock'
is just a typedef to 'void *' and something could be figured out for 'Py_off_t'
in the support code.

Anyway, the way you are implementing the tests has the same issue as Martin pointed
out for the 'object.__sizeof__' method in issue15402.  I could replace the 
'buffered_sizeof' implementation with:

   static PyObject *
   buffered_sizeof(buffered *self, void *unused)
   {
       Py_ssize_t res;

       res = 1;
       if (self->buffer)
           res += self->buffer_size;
       return PyLong_FromSsize_t(res);
   }

and the tests will still pass.

Le dimanche 29 juillet 2012 à 17:45 +0000, Meador Inge a écrit :
> Anyway, the way you are implementing the tests has the same issue as Martin pointed
> out for the 'object.__sizeof__' method in issue15402.  I could replace the 
> 'buffered_sizeof' implementation with:
> 
>    static PyObject *
>    buffered_sizeof(buffered *self, void *unused)
>    {
>        Py_ssize_t res;
> 
>        res = 1;
>        if (self->buffer)
>            res += self->buffer_size;
>        return PyLong_FromSsize_t(res);
>    }
> 
> and the tests will still pass.

That's true but, OTOH, I don't think it's very important. First, because
it is a rather unlikely mistake. Second, because the important thing is
to check that the internal buffer is part of the returned size, which is
what Serhiy's test checks for.