PopenSpawnnow has a
subprocess.Popen, for a function to be called in the child process before executing the new command. Like in
Popen, this works only in POSIX, and can cause issues if your application also uses threads (PR #460).
- Significant performance improvements when processing large amounts of data (PR #464).
- Ensure that
spawn.closedgets set by
close(), and improve an example for passing
SIGWINCHthrough to a child process (PR #466).
async=parameter to integrate with asyncio has become
async_=(PR #431), as async is becoming a Python keyword from Python 3.6. Pexpect will still recognise
asyncas an alternative spelling.
- Similarly, the module
pexpect._async(PR #450). This module is not part of the public API.
- Fix problems with asyncio objects closing file descriptors during garbage collection (#347, PR #376).
- Set the
.pidattribute of a
PopenSpawnobject (PR #417).
- Fix passing Windows paths to
PopenSpawnon Windows can pass string commands through to
Popenwithout splitting them into a list (PR #447).
shlextrying to read from stdin when
cmd=None(#433, PR #434).
- Ensure that an error closing a Pexpect spawn object raises a Pexpect error, rather than a Ptyprocess error (#383, PR #386).
- Cleaned up invalid backslash escape sequences in strings (PR #430, PR #445).
- The pattern for a password prompt in
- Correct docstring for using unicode with spawn (PR #395).
- Various other improvements to documentation.
- Fix to allow running
envin replwrap-ed bash.
- Raise more informative exception from pxssh if it fails to connect.
passmassexample to not log passwords entered.
- Change: When an
envparameter is specified to the
runfamily of calls containing a value for
PATH, its value is used to discover the target executable from a relative path, rather than the current process’s environment
PATH. This mirrors the behavior of
subprocess.Popen()in the standard library (#348).
- Regression: Re-introduce capability for
fdspawnas previously supported in version 3.3 (#359).
- Integration with
expect_list()will make them return a coroutine. You can get the result using
yield from, or wrap it in an
asyncio.Task. This allows the event loop to do other things while waiting for output that matches a pattern.
- Experimental support for Windows (with some caveats)—see Pexpect on Windows.
- Enhancement: allow method as callbacks of argument
- It is now possible to call
wait()multiple times, or after a process is already determined to be terminated without raising an exception (PR #211).
dimensions=(rows, columns)allows setting terminal screen dimensions before launching a program (#122).
- Fix regression that prevented executable, but unreadable files from being found when not specified by absolute path – such as /usr/bin/sudo (#104).
- Fixed regression when executing pexpect with some prior releases of the multiprocessing module where stdin has been closed (#86).
Backwards incompatible changes¶
pexpect.ANSI. Please use other packages such as pyte to emulate a terminal.
- Removed the independent top-level modules (
pxssh fdpexpect FSM screen ANSI) which were installed alongside Pexpect. These were moved into the Pexpect package in 3.0, but the old names were left as aliases.
- Child processes created by Pexpect no longer ignore SIGHUP by default: the
pexpect.spawndefaults to False. To get the old behaviour, pass
- Added a mechanism to wrap REPLs, or shells, in an object which can conveniently
be used to send commands and wait for the output (
- Fixed issue where pexpect would attempt to execute a directory because it has the ‘execute’ bit set (#37).
- Removed the
pexpect.pshmodule. This was never documented, and we found no evidence that people use it. The new
pexpect.replwrapmodule provides a more flexible alternative.
TypeError: got <type 'str'> ('\r\n') as patternin
- Fixed issue where EOF was not correctly detected in
interact(), causing a repeating loop of output on Linux, and blocking before EOF on BSD and Solaris (#49).
- Several Solaris (SmartOS) bugfixes, preventing
IOErrorexceptions, especially when used with cron(1) (#44).
- Added new keyword argument
spawn. On SVR4-like systems, the method
isatty()will always return False: the child pty does not appear as a terminal. Therefore,
waitnoecho()are not supported on those platforms.
After this, we intend to start working on a bigger refactoring of the code, to be released as Pexpect 4. There may be more bugfix 3.x releases, however.
- Fix exception handling from
select.select()on Python 2 (PR #38). This was accidentally broken in the previous release when it was fixed for Python 3.
- Removed a workaround for
TIOCSWINSZon very old systems, which was causing issues on some BSD systems (PR #40).
- Fixed an issue with exception handling in
The documentation for
pxssh was improved.
- Fix an issue that prevented importing pexpect on Python 3 when
sys.stdoutwas reassigned (#30).
- Improve prompt synchronisation in
- Fix pickling exception instances (PR #34).
- Fix handling exceptions from
select.select()on Python 3 (PR #33).
The examples have also been cleaned up somewhat - this will continue in future releases.
The new major version number doesn’t indicate any deliberate API incompatibility. We have endeavoured to avoid breaking existing APIs. However, pexpect is under new maintenance after a long dormancy, so some caution is warranted.
- A new unicode API was introduced.
- Python 3 is now supported, using a single codebase.
- Pexpect now requires at least Python 2.6 or 3.2.
- The modules other than pexpect, such as
pexpect.pxssh, were moved into the pexpect package. For now, wrapper modules are installed to the old locations for backwards compatibility (e.g.
import pxsshwill still work), but these will be removed at some point in the future.
SIGHUPis now optional - thanks to Kimmo Parviainen-Jalanko for the patch.
- Fix a bug regarding making the pty the controlling terminal when the process spawning it is not, actually, a terminal (such as from cron)
Fixed OSError exception when a pexpect object is cleaned up. Previously, you might have seen this exception:
Exception exceptions.OSError: (10, 'No child processes') in <bound method spawn.__del__ of <pexpect.spawn instance at 0xd248c>> ignored
You should not see that anymore. Thanks to Michael Surette.
Added support for buffering reads. This greatly improves speed when trying to match long output from a child process. When you create an instance of the spawn object you can then set a buffer size. For now you MUST do the following to turn on buffering – it may be on by default in future version:
child = pexpect.spawn ('my_command') child.maxread=1000 # Sets buffer to 1000 characters.
I made a subtle change to the way TIMEOUT and EOF exceptions behave. Previously you could either expect these states in which case pexpect will not raise an exception, or you could just let pexpect raise an exception when these states were encountered. If you expected the states then the
beforeproperty was set to everything before the state was encountered, but if you let pexpect raise the exception then
beforewas not set. Now, the
beforeproperty will get set either way you choose to handle these states.
The spawn object now provides iterators for a file-like interface. This makes Pexpect a more complete file-like object. You can now write code like this:
child = pexpect.spawn ('ls -l') for line in child: print line
write and writelines() no longer return a value. Use send() if you need that functionality. I did this to make the Spawn object more closely match a file-like object.
Added the attribute
exitstatus. This will give the exit code returned by the child process. This will be set to
Nonewhile the child is still alive. When
isalive()returns 0 then
exitstatuswill be set.
Made a few more tweaks to
isalive()so that it will operate more consistently on different platforms. Solaris is the most difficult to support.
You can now put
TIMEOUTin a list of expected patterns. This is just like putting
EOFin the pattern list. Expecting for a
TIMEOUTmay not be used as often as
EOF, but this makes Pexpect more consistent.
Thanks to a suggestion and sample code from Chad J. Schroeder I added the ability for Pexpect to operate on a file descriptor that is already open. This means that Pexpect can be used to control streams such as those from serial port devices. Now, you just pass the integer file descriptor as the “command” when constructing a spawn open. For example on a Linux box with a modem on ttyS1:
fd = os.open("/dev/ttyS1", os.O_RDWR|os.O_NONBLOCK|os.O_NOCTTY) m = pexpect.spawn(fd) # Note integer fd is used instead of usual string. m.send("+++") # Escape sequence m.send("ATZ0\r") # Reset modem to profile 0 rval = m.expect(["OK", "ERROR"])
read()was renamed to
read_nonblocking(). Added new
read()method that matches file-like object interface. In general, you should not notice the difference except that
read()no longer allows you to directly set the timeout value. I hope this will not effect any existing code. Switching to
read_nonblocking()should fix existing code.
Changed the name of
Changed the name of
kill()so that it checks to make sure the pid
spawn()(really called from
__spawn()) so that it does not raise an exception if
setwinsize()fails. Some platforms such as Cygwin do not like setwinsize. This was a constant problem and since it is not a critical feature I decided to just silence the error. Normally I don’t like to do that, but in this case I’m making an exception.
Added a method
close()that does what you think. It closes the file descriptor of the child application. It makes no attempt to actually kill the child or wait for its status.
__revision__(from cvs) to the pexpect modules. This is mainly helpful to me so that I can make sure that I’m testing with the right version instead of one already installed.
log_close(have been removed. Now use
setlog()method takes a file object. This is far more flexible than the previous log method. Each time data is written to the file object it will be flushed. To turn logging off simply call
isalive()to match the more typical naming style in Python. Also the technique used to detect child process status has been drastically modified. Previously I did some funky stuff with signals which caused indigestion in other Python modules on some platforms. It was a big headache. It still is, but I think it works better now.
expect_eof()method is gone. You can now simply use the
expect()method to look for EOF.
Pexpect works on OS X, but the nature of the quirks cause many of the tests to fail. See bugs. (Incomplete Child Output). The problem is more than minor, but Pexpect is still more than useful for most tasks.
Solaris: For some reason, the second time a pty file descriptor is created and deleted it never gets returned for use. It does not effect the first time or the third time or any time after that. It’s only the second time. This is weird… This could be a file descriptor leak, or it could be some peculiarity of how Solaris recycles them. I thought it was a UNIX requirement for the OS to give you the lowest available filedescriptor number. In any case, this should not be a problem unless you create hundreds of pexpect instances… It may also be a pty module bug.
Moves and forks¶
- Pexpect development used to be hosted on Sourceforge.
- In 2011, Thomas Kluyver forked pexpect as ‘pexpect-u’, to support Python 3. He later decided he had taken the wrong approach with this.
- In 2012, Noah Spurrier, the original author of Pexpect, moved the project to Github, but was still too busy to develop it much.
- In 2013, Thomas Kluyver and Jeff Quast forked Pexpect again, intending to call the new fork Pexpected. Noah Spurrier agreed to let them use the name Pexpect, so Pexpect versions 3 and above are based on this fork, which now lives here on Github.