input() uses sys.__stdout__ instead of sys.stdout for prompt #68590
Comments
|
On Python 3.4, input() does not use wrapped sys.stdout. For example, the following script should print "Wrapped stdout" by print() and input() because sys.stdout is replaced with Writer object. -------- import io
import sys
class Writer:
def __getattr__(self, name):
return getattr(sys.__stdout__, name)
def write(self, data):
if data != '\n':
sys.__stdout__.write('Wrapped stdout\n')
def fileno():
raise OSError()
sys.stdout = Writer()
print('print')
input('input')But the script does not print "Wrapped stdout" as prompt of input(). The script ----- Although, when sys.stdin is also wrapped, input() will use sys.stdout for prompt. For example, the following script prints "Wrapped stdout" by print() and input(). ---- import io
import sys
class Writer:
def __getattr__(self, name):
return getattr(sys.__stdout__, name)
def write(self, data):
if data != '\n':
sys.__stdout__.write('Wrapped stdout\n')
def fileno():
raise OSError()
class Reader:
def __getattr__(self, name):
return getattr(sys.__stdin__, name)
def read(self, size):
return sys.__stdin__.read(size)
def fileno():
raise OSError()
sys.stdout = Writer()
sys.stdin = Reader()
print('print')
input('input')The script prints the following. ----- |
KeitaKita
mannequin
added
interpreter-core
|
This looks like an oversight. In the implementation in bltinmodule.c, if checking sys.stdout.fileno() fails, it clears the error without setting tty = 0. It's the same in 3.5. /* We should only use (GNU) readline if Python's sys.stdin and
sys.stdout are the same as C's stdin and stdout, because we
need to pass it those. */
tmp = _PyObject_CallMethodId(fin, &PyId_fileno, "");
if (tmp == NULL) {
PyErr_Clear();
tty = 0;
}
else {
fd = PyLong_AsLong(tmp);
Py_DECREF(tmp);
if (fd < 0 && PyErr_Occurred())
return NULL;
tty = fd == fileno(stdin) && isatty(fd);
}
if (tty) {
tmp = _PyObject_CallMethodId(fout, &PyId_fileno, "");
if (tmp == NULL)
PyErr_Clear();
else {
fd = PyLong_AsLong(tmp);
Py_DECREF(tmp);
if (fd < 0 && PyErr_Occurred())
return NULL;
tty = fd == fileno(stdout) && isatty(fd);
}
} |
|
Okay, so the change is straightforward. With this change the entire test suite passes on my machine. We'll need a test for this for the test suite, and that's far less straightforward. It seems that test.test_builtin.BuiltinTest.check_input_tty() could be a starting point. Could someone come up with a working test for this? |
|
Here's a patch with just the required fix (no tests). |
|
Patch itself looks good. |
|
This patch includes a test case using pty.fork() to ensure stdin is a terminal, so the bug would be triggered. It is a bit tricky, but hopefully works on Unix platforms. I have only tested it on Linux. |
|
LGTM. See also minor note on Rietveld. I think it is worth to extract all input() tests to separate class. |
|
Okay I plan to commit a modified version where I moved the input() test methods that use pseudo terminals to a new PtyTests class. The test_input() method stays in the original class, because it shares a helper method with other tests. I am resisting dropping that decode() step, because it makes any error message from the forked child hard to read: AssertionError: 256 != 0 : b'quux\r\nTraceback (most recent call last):\r\n File "/media/disk/home/proj/python/cpython/Lib/test/test_builtin.py", line 1593, in test_input_no_stdout_fileno\r\n input("prompt")\r\nTypeError: bad argument type for built-in operation\r\n' |
|
New changeset 425d81d10b13 by Martin Panter in branch '3.4': New changeset e97d940a6543 by Martin Panter in branch '3.5': New changeset bcc0f8eb6797 by Martin Panter in branch 'default': |
|
Seems my test hangs the Free BSD 9 buildbot, although Free BSD 10 apparently passed. [390/390] test_builtin |
|
This test also fails on Ubuntu. [1/1] test_builtin
test test_builtin failed -- Traceback (most recent call last):
File "/home/angwer/my-cpython/Lib/test/test_builtin.py", line 1588, in test_input_no_stdout_fileno
self.assertEqual(status, 0, output)
AssertionError: 256 != 0 : quux
Traceback (most recent call last):
File "/home/angwer/my-cpython/Lib/test/test_builtin.py", line 1593, in test_input_no_stdout_fileno
input("prompt")
TypeError: bad argument type for built-in operation1 test failed: |
|
I wrote this on Arch Linux, so I doubt your Ubuntu failure is valid. The error you see is what I would expect if the bug were not actually fixed. I suspect you need to recompile /Python/bltinmodule.c, which contains the bug fix. However, the test also hangs OS X buildbots. I can only guess at why it is hanging, either at the waitpid() line or the os.read() line. I am looking at refactoring my test to use the existing check_input_tty() infrastructure. It seems over the top by adding a separate pipe back to the parent process, but presumably will workaround whatever is causing the hang. |
|
It's my fault. I forget to recompile. Now it works. |
|
New changeset d8dd9015b086 by Martin Panter in branch '3.4': New changeset 6a8f96b46dce by Martin Panter in branch '3.5': New changeset cb574ee7231e by Martin Panter in branch 'default': |
|
My second attempt seems to have avoided the hang. For the record, the backtrace from 3.4.koobs-freebsd9 did say it was hanging at the os.waitpid() line, but I still don’t understand why. |
Note: these values reflect the state of the issue at the time it was migrated and might not reflect the current state.
Show more details
GitHub fields:
bugs.python.org fields:
The text was updated successfully, but these errors were encountered: