# Copyright 2010-2011 Gentoo Foundation
# Distributed under the terms of the GNU General Public License v2
-import errno
-import fcntl
import platform
import pty
-import select
-import sys
import termios
-from portage import os, _unicode_decode, _unicode_encode
+from portage import os
from portage.output import get_term_size, set_term_size
-from portage.process import spawn_bash
from portage.util import writemsg
-def _can_test_pty_eof():
- """
- The _test_pty_eof() function seems to hang on most
- kernels other than Linux.
- This was reported for the following kernels which used to work fine
- without this EOF test: Darwin, AIX, FreeBSD. They seem to hang on
- the slave_file.close() call. Note that Python's implementation of
- openpty on Solaris already caused random hangs without this EOF test
- and hence is globally disabled.
- @rtype: bool
- @returns: True if _test_pty_eof() won't hang, False otherwise.
- """
- return platform.system() in ("Linux",)
-
-def _test_pty_eof():
- """
- Returns True if EOF appears to be handled correctly with pty
- devices. Raises an EnvironmentError from openpty() if it fails.
-
- This used to be used to detect if the following issue was fixed
- in the currently running version of python:
-
- http://bugs.python.org/issue5380
-
- However, array.fromfile() use has since been abandoned due to
- bugs that exist in all known versions of Python (including Python
- 2.7 and Python 3.2). See PipeReaderArrayTestCase, for example.
- This is somewhat unfortunate, since the combination of
- array.fromfile() and array.tofile() is approximately 10% faster
- than the combination of os.read() and os.write().
- """
-
- use_fork = False
-
- test_string = 2 * "blah blah blah\n"
- test_string = _unicode_decode(test_string,
- encoding='utf_8', errors='strict')
-
- # may raise EnvironmentError
- master_fd, slave_fd = pty.openpty()
-
- # Non-blocking mode is required for Darwin kernel.
- fcntl.fcntl(master_fd, fcntl.F_SETFL,
- fcntl.fcntl(master_fd, fcntl.F_GETFL) | os.O_NONBLOCK)
-
- # Disable post-processing of output since otherwise weird
- # things like \n -> \r\n transformations may occur.
- mode = termios.tcgetattr(slave_fd)
- mode[1] &= ~termios.OPOST
- termios.tcsetattr(slave_fd, termios.TCSANOW, mode)
-
- # Simulate a subprocess writing some data to the
- # slave end of the pipe, and then exiting.
- pid = None
- if use_fork:
- pids = spawn_bash(_unicode_encode("echo -n '%s'" % test_string,
- encoding='utf_8', errors='strict'), env=os.environ,
- fd_pipes={0:sys.stdin.fileno(), 1:slave_fd, 2:slave_fd},
- returnpid=True)
- if isinstance(pids, int):
- os.close(master_fd)
- os.close(slave_fd)
- raise EnvironmentError('spawn failed')
- pid = pids[0]
- else:
- os.write(slave_fd, _unicode_encode(test_string,
- encoding='utf_8', errors='strict'))
- os.close(slave_fd)
-
- # If using a fork, we must wait for the child here,
- # in order to avoid a race condition that would
- # lead to inconsistent results.
- if pid is not None:
- os.waitpid(pid, 0)
+# Disable the use of openpty on Solaris as it seems Python's openpty
+# implementation doesn't play nice on Solaris with Portage's
+# behaviour causing hangs/deadlocks.
+# Additional note for the future: on Interix, pipes do NOT work, so
+# _disable_openpty on Interix must *never* be True
+_disable_openpty = platform.system() in ("SunOS",)
- data = []
- iwtd = [master_fd]
- owtd = []
- ewtd = []
-
- while True:
-
- events = select.select(iwtd, owtd, ewtd)
- if not events[0]:
- # EOF
- break
-
- buf = None
- try:
- buf = os.read(master_fd, 1024)
- except OSError as e:
- # EIO happens with pty on Linux after the
- # slave end of the pty has been closed.
- if e.errno == errno.EIO:
- # EOF: return empty string of bytes
- buf = b''
- elif e.errno == errno.EAGAIN:
- # EAGAIN: return None
- buf = None
- else:
- raise
-
- if buf is None:
- pass
- elif not buf:
- # EOF
- break
- else:
- data.append(buf)
-
- os.close(master_fd)
-
- return test_string == _unicode_decode(b''.join(data), encoding='utf_8', errors='strict')
-
-# If _test_pty_eof() can't be used for runtime detection of
-# http://bugs.python.org/issue5380, openpty can't safely be used
-# unless we can guarantee that the current version of python has
-# been fixed (affects all current versions of python3). When
-# this issue is fixed in python3, we can add another sys.hexversion
-# conditional to enable openpty support in the fixed versions.
-if sys.hexversion >= 0x3000000 and not _can_test_pty_eof():
- _disable_openpty = True
-else:
- # Disable the use of openpty on Solaris as it seems Python's openpty
- # implementation doesn't play nice on Solaris with Portage's
- # behaviour causing hangs/deadlocks.
- # Additional note for the future: on Interix, pipes do NOT work, so
- # _disable_openpty on Interix must *never* be True
- _disable_openpty = platform.system() in ("SunOS",)
_tested_pty = False
-if not _can_test_pty_eof():
- # Skip _test_pty_eof() on systems where it hangs.
- _tested_pty = True
-
_fbsd_test_pty = platform.system() == 'FreeBSD'
def _create_pty_or_pipe(copy_term_size=None):
got_pty = False
- global _disable_openpty, _fbsd_test_pty, _tested_pty
- if not (_tested_pty or _disable_openpty):
- try:
- if not _test_pty_eof():
- _disable_openpty = True
- except EnvironmentError as e:
- _disable_openpty = True
- writemsg("openpty failed: '%s'\n" % str(e),
- noiselevel=-1)
- del e
- _tested_pty = True
+ global _disable_openpty, _fbsd_test_pty
if _fbsd_test_pty and not _disable_openpty:
# Test for python openpty breakage after freebsd7 to freebsd8
projects / portage.git / commitdiff