<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN">
<html>
<head>
<title>Pexpect - a Pure Python Expect-like module</title>
<link rel="stylesheet" href="clean.css" type="text/css">
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<meta name="Author" content="Noah Spurrier">
<meta name="Keywords"
 content="pexpect, Noah Spurrier, pypect, Python, Libes, TCL, Expect, pipe, popen, pyExpect, expectpy, expect-like, expect-alike, expect like">
<meta name="Description"
 content="Pexpect is a pure Python Expect-like module. Pexpect makes Python a better tool for controlling other applications.">
</head>
<body bgcolor="#ffffff" text="#000000">
<div id="Header">
<h1>Pexpect version 2.4<br>
a Pure Python Expect-like module
</h1>
</div>
<div id="Content">
<p>Pexpect makes Python a better tool for controlling other
applications.</p>
<p>Pexpect is a pure Python module for spawning child applications;
controlling them; and responding to expected patterns in their output.
Pexpect works like Don Libes' Expect. Pexpect allows your script to
spawn a child application and control it as if a human were typing
commands.</p>
<p>Pexpect can be used for automating interactive applications such as
ssh, ftp, passwd, telnet, etc. It can be used to a automate setup
scripts for duplicating software package installations on different
servers. It can be used for automated software testing. Pexpect is in
the spirit of Don Libes' Expect, but Pexpect is pure Python. Unlike
other Expect-like modules for Python, Pexpect does not require TCL or
Expect nor does it require C extensions to be compiled. It should work
on any platform that supports the standard Python pty module. The
Pexpect interface was designed to be easy to use.</p>
<table border="0">
  <tbody>
    <tr>
      <td align="right" valign="top">Send questions to:</td>
      <td align="left"><a href="http://www.noah.org/email/"><img
 src="email.png" alt="Click to send email." border="0" height="16"
 width="100"></a></td>
    </tr>
  </tbody>
</table>
<hr noshade="noshade" size="1">
<h1><a name="license"></a>License: MIT style</h1>
<p>
Free, open source, and all that good stuff.<br>
<br>
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:<br>
<br>
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.<br>
<br>
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM,
DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
USE OR OTHER DEALINGS IN THE SOFTWARE.<br>
<br>
Pexpect Copyright (c) 2008 Noah Spurrier<br>
http://pexpect.sourceforge.net/
</p>

<hr noshade="noshade" size="1">
<h1><a name="download"></a><a
 href="http://sourceforge.net/project/showfiles.php?group_id=59762">Download</a></h1>
<p>Download the <a
 href="http://sourceforge.net/project/showfiles.php?group_id=59762">
current version here</a> from the SourceForge site. Grab the current Pexpect tarball.
</p>
<h2>Installing Pexpect</h2>
<p>The Pexpect tarball is a standard Python Distutil distribution.</p>
<ol>
  <li>download <span class="code">pexpect-2.4.tar.gz</span></li>
  <li><span class="code">tar zxf pexpect-2.4.tar.gz</span></li>
  <li><span class="code">cd pexpect-2.4</span></li>
  <li><span class="code">python setup.py install</span> <i>do this as root</i></li>
</ol>
<h2>Examples</h2>
<p>
Under the <span class="code">pexpect-2.4</span> directory you should find
the <span class="code">examples</span> directory.
This is the best way to learn to use Pexpect.
See the descriptions of <a href="examples.html">Pexpect Examples</a>.
</p>
<h2><a name="doc"></a>API Documentation</h2>
<p>
<blockquote>
<a href="pexpect.html">pexpect</a> This is the main module that you want.<br>
<a href="pxssh.html">pxssh</a> Pexpect SSH is an extension of 'pexpect.spawn' that specializes in SSH.<br>
</blockquote>
the following are experimental extensions to Pexpect<br>
<blockquote>
<a href="fdpexpect.html">fdpexpect</a> fdpexpect extension of 'pexpect.spawn' that uses an open file descriptor.<br>
<a href="screen.html">SCREEN</a> This represents a virtual 'screen'.<br>
<a href="ANSI.html">ANSI</a> This parses ANSI/VT-100 terminal escape codes.<br>
<a href="FSM.html">FSM</a> This is a finite state machine used by ANSI.<br>
</blockquote>
</p>
<hr noshade="noshade" size="1">
<h1><a name="status"></a>Project Status</h1>
<p>Automated pyunit tests reach over 80%
code coverage on pexpect.py. I regularly test on Linux and BSD
platforms. I try to test on Solaris and Irix. 
</p>
<hr noshade="noshade" size="1">
<h1><a name="requirements"></a>Requirements for use of Pexpect</h1>
<h2>Python</h2>
<blockquote>
  <p>Pexpect was written and tested with Python 2.4. It should work on
earlier versions that have the <span class="code">pty</span> module. I
sometimes even manually test it with Python 1.5.2, but I can't easily
run the PyUnit test framework against Python 1.5.2, so I have less
confidence in Pexpect on Python 1.5.2.</p>
</blockquote>
<h2>pty module</h2>
<blockquote>
  <p>Any POSIX system (UNIX) with a working <span class="code">pty</span>
module should be able to run Pexpect. The <span class="code">pty</span>
module is part of the Standard Python Library, so if you are running on
a POSIX system you should have it. The <span class="code">pty</span>
module does not run the same on all platforms. It should be solid on Linux
and BSD systems. I have taken effort to try to smooth the wrinkles out of the different platforms. To learn more
about the wrinkles see <a href="#bugs">Bugs</a> and <a href="#testing">Testing</a>.</p>
</blockquote>
<p>Pexpect does not currently work on the standard Windows Python (see
the pty requirement); however, it seems to work fine using <a
 href="http://www.cygwin.com/">Cygwin</a>. It is possible to build
something like a pty for Windows, but it would have to use a different
technique that I am still investigating. I know it's possible because
Libes' Expect was ported to Windows. <i>If you have any ideas or
skills to contribute in this area then I would really appreciate some
tips on how to approach this problem.</i> </p>
<hr noshade="noshade" size="1">
<h1><a name="overview"></a>Overview</h1>
<p>Pexpect can be used for automating interactive applications such as
ssh, ftp, mencoder, passwd, etc. The Pexpect interface was designed to be
easy to use. Here is an example of Pexpect in action:</p>
<blockquote>
  <pre class="code"># This connects to the openbsd ftp site and<br># downloads the recursive directory listing.<br>import pexpect<br>child = pexpect.spawn ('ftp ftp.openbsd.org')<br>child.expect ('Name .*: ')<br>child.sendline ('anonymous')<br>child.expect ('Password:')<br>child.sendline ('noah@example.com')<br>child.expect ('ftp&gt; ')<br>child.sendline ('cd pub')<br>child.expect('ftp&gt; ')<br>child.sendline ('get ls-lR.gz')<br>child.expect('ftp&gt; ')<br>child.sendline ('bye')<br></pre>
</blockquote>
<p> Obviously you could write an ftp client using Python's own <span
 class="code">ftplib</span> module, but this is just a demonstration.
You can use this technique with any application. This is especially
handy if you are writing automated test tools.</p>

<p>There are two important methods in Pexpect -- <span class="code"><b>expect()</b></span>
and <span class="code"><b>send()</b></span> (or <span class="code">sendline()</span>
which is like <span class="code">send()</span> with a linefeed). 
The <span class="code">expect()</span> method waits for the child application
to return a given string. The string you specify is a regular expression, so
you can match complicated patterns. The <span class="code"><b>send()</b></span> method
writes a string to the child application. From the child's point of
view it looks just like someone typed the text from a terminal. After
each call to <span class="code"><b>expect()</b></span> the <span
 class="code"><b>before</b></span> and <span class="code"><b>after</b></span>
properties will be set to the text printed by child application. The <span
 class="code"><b>before</b></span> property will contain all text up to
the expected string pattern. The <span class="code"><b>after</b></span> string
will contain the text that was matched by the expected pattern.
The <span class="code">match</span> property is set to the <span class="code">re MatchObject</span>.
</p>

<p>An example of Pexpect in action may make things more clear. This example uses
<span class="code">ftp</span> to login to the OpenBSD site; list files
in a directory; and then pass interactive control of the ftp session to
the human user.</p>
<blockquote>
  <pre class="code">import pexpect<br>child = pexpect.spawn ('ftp ftp.openbsd.org')<br>child.expect ('Name .*: ')<br>child.sendline ('anonymous')<br>child.expect ('Password:')<br>child.sendline ('noah@example.com')<br>child.expect ('ftp&gt; ')<br>child.sendline ('ls /pub/OpenBSD/')<br>child.expect ('ftp&gt; ')<br>print child.before   # Print the result of the ls command.<br>child.interact()     # Give control of the child to the user.<br></pre>
</blockquote>
<h2>Special EOF and TIMEOUT patterns</h2>
<p>
There are two special patterns to match the End Of File or a Timeout condition.
You you can pass these patterns to <span class="code">expect()</span>.
These patterns are not regular expressions. Use them like predefined constants.
</p>
<p>If the child has died and you have read all the child's output then ordinarily
<span class="code">expect()</span> will raise an <span class="code">EOF</span>
exception. You can read everything up to the EOF without generating an
exception by using the EOF pattern <span class="code">expect(pexpect.EOF)</span>.
In this case everything the child has output will be available in the <span
 class="code">before</span> property.</p>
<p>The pattern given to <span class="code">expect()</span> may be a
regular expression or it may also be a <b>list</b> of regular expressions.
This allows you to match multiple optional responses. The <span class="code">expect()</span>
method returns the index of the pattern that was matched. For example,
say you wanted to login to a server. After entering a password you
could get various responses from the server -- your password could be
rejected; or you could be allowed in and asked for your terminal type;
or you could be let right in and given a command prompt. The following
code fragment gives an example of this:</p>
<blockquote>
  <pre class="code">child.expect('password:')<br>child.sendline (my_secret_password)<br># We expect any of these three patterns...<br>i = child.expect (['Permission denied', 'Terminal type', '[#\$] '])<br>if i==0:<br>    print 'Permission denied on host. Can't login'<br>    child.kill(0)<br>elif i==2:<br>    print 'Login OK... need to send terminal type.'<br>    child.sendline('vt100')<br>    child.expect ('[#\$] ')<br>elif i==3:<br>    print 'Login OK.'<br>    print 'Shell command prompt', child.after</pre>
</blockquote>
<p>If nothing matches an expected pattern then expect will eventually
raise a TIMEOUT exception. The default time is 30 seconds, but you can
change this by passing a timeout argument to expect():</p>
<blockquote>
  <pre class="code"># Wait no more than 2 minutes (120 seconds) for password prompt.<br>child.expect('password:', timeout=120)</pre>
</blockquote>
<h2>Find the end of line -- CR/LF conventions<br>
Matching at the end of a line can be tricky<br>
$ regex pattern is useless.<br>
</h2>
<p>Pexpect matches regular expressions a little differently than what
you might be used to.
</p>
<p><i><b>The $ pattern for end of line match is useless</b></i>.
The $ matches the end of string, but Pexpect reads from the child
one character at a time, so each character looks like the end of a line.
Pexpect can't do a look-ahead into the child's output stream.
In general you would have this situation when using regular expressions
with any stream.<br>
<i>Note, pexpect does have an internal buffer, so reads are faster
than one character at a time, but from the user's perspective the regex
patterns test happens one character at a time.</i></p>
<p>The best way to match the end of a line is to look for the
newline: "\r\n" (CR/LF). Yes, that does appear to be DOS-style.
It may surprise some UNIX people to learn that terminal TTY device drivers
(dumb, vt100, ANSI, xterm, etc.) all use the CR/LF combination to signify
the end of line. Pexpect uses a Pseudo-TTY device to talk to the child application, so
when the child app prints "\n" you actually see "\r\n".
</p>
<p><b>UNIX uses just linefeeds to end lines of text, but not when it
comes to TTY devices!</b> TTY devices are more like the Windows world.
Each line of text end with a CR/LF combination. When you intercept data
from a UNIX command from a TTY device you will find that the TTY device
outputs a CR/LF combination. A UNIX command may only write a linefeed
(\n), but the TTY device driver converts it to CR/LF. This means that
your terminal will see lines end with CR/LF (hex&nbsp;<span class="code">0D&nbsp;0A</span>).
Since Pexpect emulates a terminal, to match ends of lines you have to
expect the CR/LF combination.</p>
<blockquote>
  <p class="code">child.expect ('\r\n')</p>
</blockquote>
<p>If you just need to skip past a new line then <span class="code">expect
('\n')</span> by itself will work, but if you are expecting a specific
pattern before the end of line then you need to explicitly look for the
\r. For example the following expects a word at the end of a line:</p>
<blockquote>
  <p class="code">child.expect ('\w+\r\n')</p>
</blockquote>
<p>But the following would both fail:</p>
<blockquote>
  <p class="code">child.expect ('\w+\n')</p>
</blockquote>
<p>And as explained before, trying to use '$' to match the end of line
would not work either:</p>
<blockquote>
  <p class="code">child.expect ('\w+$')</p>
</blockquote>
<p>So if you need to explicitly look for the END OF LINE, you want to
look for the CR/LF combination -- not just the LF and not the $ pattern.</p>
<p>This problem is not limited to Pexpect. This problem happens any
time you try to perform a regular expression match on a stream. Regular
expressions need to look ahead. With a stream it is hard to look ahead
because the process generating the stream may not be finished. There is no
way to know if the process has paused momentarily or is finished and
waiting for you. <font color="#cc0000">Pexpect must implicitly always
do a NON greedy match (minimal) at the end of a input {### already said
this}.</font> </p>
<p>Pexpect compiles all regular expressions with the DOTALL flag. With
the DOTALL flag a "." will match a newline. See the Python <a
 href="http://www.python.org/doc/current/lib/node115.html#l2h-733">documentation</a></p>
<h2>Beware of + and * at the end of input.</h2>
<p>Remember that any time you try to match a pattern that needs
look-ahead that you will always get a minimal match (non greedy). For
example, the following will always return just one character:</p>
<blockquote>
  <p class="code">child.expect ('.+')</p>
</blockquote>
<p>This example will match successfully, but will always return no
characters:</p>
<blockquote>
  <p class="code">child.expect ('.*')</p>
</blockquote>
<p>Generally any star * expression will match as little as possible</p>
<p>One thing you can do is to try to force a non-ambiguous character at
the end of your <span class="code">\d+</span> pattern. Expect that
character to delimit the string. For example, you might try making the
end of your pattrn be <span class="code">\D+</span> instead of <span
 class="code">\D*</span>. That means number digits alone would not
satisfy the (<span class="code">\d+</span>) pattern. You would need
some number(s) and at least one <span class="code">\D</span> at the
end. </p>
<h2>Matching groups</h2>
<p>You can group regular expression using parenthesis. After a match,
the <span class="code">match</span> parameter of the spawn object will
contain the Python Match object. </p>
<h2>Examples</h2>
<p>Using "match" and groups...</p>
<h2>Debugging</h2>
<p>If you get the string value of a pexpect.spawn object you will get
lots of useful debugging information. For debugging it's very useful to
use the following pattern:</p>
<p>try:<br>
&nbsp;&nbsp;&nbsp; i = child.expect ([pattern1, pattern2, pattern3,
etc])<br>
except:<br>
&nbsp;&nbsp;&nbsp; print "Exception was thrown"<br>
&nbsp;&nbsp;&nbsp; print "debug information:"<br>
&nbsp;&nbsp;&nbsp; print str(child)<br>
</p>
<p>It is also useful to log the child's input and out to a file or the
screen. The following will turn on logging and send output to stdout
(the screen).<br>
</p>
<p>&nbsp;&nbsp;&nbsp; child = pexpect.spawn (foo)<br>
&nbsp;&nbsp;&nbsp; child.logfile = sys.stdout<br>
<br>
</p>
<hr noshade="noshade" size="1">
<h1>Exceptions</h1>
<p><b>EOF</b></p>
<p>Note that two flavors of EOF Exception may be thrown. They are
virtually identical except for the message string. For practical
purposes you should have no need to distinguish between them, but they
do give a little extra information about what type of platform you are
running. The two messages are:</p>
<blockquote>
  <p class="code">End Of File (EOF) in read(). Exception style platform.</p>
  <p class="code">End Of File (EOF) in read(). Empty string style
platform.</p>
</blockquote>
<p>Some UNIX platforms will throw an exception when you try to read
from a file descriptor in the EOF state. Other UNIX platforms instead
quietly return an empty string to indicate that the EOF state has been
reached.</p>
<p><b>Expecting EOF</b></p>
<p>If you wish to read up to the end of the child's output without
generating an <span class="code">EOF</span> exception then use the <span
 class="code">expect(pexpect.EOF)</span> method.</p>
<p><b>TIMEOUT</b></p>
<p>The <span class="code">expect()</span> and <span class="code">read()</span>
methods will also timeout if the child does not generate any output for
a given amount of time. If this happens they will raise a <span
 class="code">TIMEOUT</span> exception. You can have these method
ignore a timeout and block indefinitely by passing None for the timeout
parameter.</p>
<blockquote>
  <p class="code">child.expect(pexpect.EOF, timeout=None)</p>
</blockquote>
<hr noshade="noshade" size="1">
<h1><a name="faq"></a>FAQ</h1>
<p><b>Q: Why don't shell pipe and redirect (| and >) work when I
spawn a command?</b></p>
<p>

A: Remember that Pexpect does NOT interpret shell meta characters such as
redirect, pipe, or wild cards (&gt;, |, or *). That's done by a shell not the
command you are spawning. This is a common mistake. If you want to run a
command and pipe it through another command then you must also start a shell.
For example:

<pre>
    child = pexpect.spawn('/bin/sh -c "ls -l | grep LOG &gt; log_list.txt"')
    child.expect(pexpect.EOF)
</pre>

The second form of spawn (where you pass a list of arguments) is useful in
situations where you wish to spawn a command and pass it its own argument list.
This can make syntax more clear. For example, the following is equivalent to
the previous example:

<pre>
    shell_cmd = 'ls -l | grep LOG &gt; log_list.txt'
    child = pexpect.spawn ('/bin/sh', ['-c', shell_cmd])
    child.expect (pexpect.EOF)
</pre>

</p>
<p><b>Q: Isn't there already a Python Expect?</b></p>
<p>A: Yes, there are several of them. They usually require you to
compile C. I wanted something that was pure Python and preferably a
single module that was simple to install. I also wanted something that
was easy to use. This pure Python expect only recently became possible
with the introduction of the pty module in the standard Python library.
Previously C extensions were required.</p>

<p><strong>Q: The before and after properties sound weird.</strong></p>
<p>Originally I was going to model Pexpect more after Expect, but then
I found that I could never remember how to get the context of the stuff
I was trying to parse. I hate having to read my own documentation. I
decided that it was easier for me to remember what before and after
was. It just so happens that this is how the -B and -A options in grep
works, so that made it even easier for me to remember. Whatever makes
my life easier is what's best.</p>

<p><b>Q: Why not just use Expect?</b></p>
<p>A: I love it. It's great. I has bailed me out of some real jams, but
I wanted something that would do 90% of what I need from Expect; be 10%
of the size; and allow me to write my code in Python instead of TCL.
Pexpect is not nearly as big as Expect, but Pexpect does everything I
have ever used Expect for.
<!-- :-P If I liked TCL then you wouldn't be reading this. My appologies to Don Libes -- Expect is cool, TK is cool, but TCL is only slightly better than Perl in my book. Hopefully after Expyct is done I will not need to use Expect anymore -- except for that lovely autoexpect tool. Damn, I wish I had that! --> </p>

<p><b>Q: Why not just use a pipe (popen())?</b></p>
<p>A: A pipe works fine for getting the output to non-interactive
programs. If you just want to get the output from <span class="code">ls</span>,
<span class="code">uname</span>, or <span class="code">ping</span>
then this works. Pipes do not work very well for interactive programs
and pipes will almost certainly fail for most applications that ask for
passwords such as telnet, ftp, or ssh.</p>
<p>There are two reasons for this. </p>
<p>First an application may bypass stdout and print directly to its
controlling TTY. Something like SSH will do this when it asks you for a
password. This is why you cannot redirect the password prompt because
it does not go through stdout or stderr.</p>
<p>The second reason is because most applications are built using the C
Standard IO Library (anything that uses <span class="code">#include
&lt;stdio.h&gt;</span>). One of the features of the stdio library is
that it buffers all input and output. Normally output is <b><i>line
buffered</i></b> when a program is printing to a TTY (your terminal
screen). Every time the program prints a line-feed the currently
buffered data will get printed to your screen. The problem comes when
you connect a pipe. The stdio library is smart and can tell that it is
printing to a pipe instead of a TTY. In that case it switches from line
buffer mode to <i><b>block buffered</b></i>. In this mode the
currently buffered data is flushed when the buffer is full. This causes
most interactive programs to deadlock. Block buffering is more
efficient when writing to disks and pipes. Take the situation where a
program prints a message "Enter your user name:\n" and then waits for
you type type something. In block buffered mode, the stdio library will
not put the message into the pipe even though a linefeed is printed.
The result is that you never receive the message, yet the child
application will sit and wait for you to type a response. Don't confuse
the stdio lib's buffer with the pipe's buffer. The pipe buffer is
another area that can cause problems. You could flush the input side of
a pipe, whereas you have no control over the stdio library buffer. </p>
<p>More information: the Standard IO library has three states for a
FILE *. These are: _IOFBF for block buffered; _IOLBF for line buffered;
and _IONBF for unbuffered. The STDIO lib will use block buffering when
talking to a block file descriptor such as a pipe. This is usually not
helpful for interactive programs. Short of recompiling your program to
include fflush() everywhere or recompiling a custom stdio library there
is not much a controlling application can do about this if talking over
a pipe.</p>
<p> The program may have put data in its output that remains unflushed
because the output buffer is not full; then the program will go and
deadlock while waiting for input -- because you never send it any
because you are still waiting for its output (still stuck in the
STDIO's output buffer).</p>
<p>The answer is to use a pseudo-tty. A TTY device will force <i><b>line</b></i>
buffering (as opposed to block buffering). Line buffering means that
you will get each line when the child program sends a line feed. This
corresponds to the way most interactive programs operate -- send a line
of output then wait for a line of input.</p>
<p>I put "answer" in quotes because it's ugly solution and because
there is no POSIX standard for pseudo-TTY devices (even though they
have a TTY standard...). What would make more sense to me would be to
have some way to set a mode on a file descriptor so that it will tell
the STDIO to be line-buffered. I have investigated, and I don't think
there is a way to set the buffered state of a child process. The STDIO
Library does not maintain any external state in the kernel or whatnot,
so I don't think there is any way for you to alter it. I'm not quite
sure how this line-buffered/block-buffered state change happens
internally in the STDIO library. I think the STDIO lib looks at the
file descriptor and decides to change behavior based on whether it's a
TTY or a block file (see isatty()).</p>
<p>I hope that this qualifies as helpful.</p>

<h1>Don't use a pipe to control another application...</h1>
<p>Pexpect may seem similar to <span class="code">os.popen()</span> or
<span class="code">commands</span> module. The main difference is that
Pexpect (like Expect) uses a pseudo-TTY to talk to the child
application. Most applications do no work well through the system()
call or through pipes. And probably all applications that ask a user to
type in a password will fail. These applications bypass the stdin and
read directly from the TTY device. Many applications do not explicitly
flush their output buffers. This causes deadlocks if you try to control
an interactive application using a pipe. What happens is that most UNIX
applications use the stdio (#include &lt;stdio.h&gt;) for input and
output. The stdio library behaves differently depending on where the
output is going. There is no way to control this behavior from the
client end.<br>
</p>

<p><b>Q: Can I do screen scraping with this thing?</b></p>
<p>A: That depends. If your application just does line-oriented output
then this is easy. If it does screen-oriented output then it may work,
but it could be hard. For example, trying to scrape data from the 'top'
command would be hard. The top command repaints the text window. </p>
<p>I am working on an ANSI / VT100 terminal emulator that will have
methods to get characters from an arbitrary X,Y coordinate of the
virtual screen. It works and you can play with it, but I have no
working examples at this time.</p>
<hr noshade="noshade" size="1">
<h1><a name="bugs"></a>Bugs</h1>
<h2>Threads</h2>
<p>On Linux (RH 8) you cannot spawn a child from a different thread and
pass the handle back to a worker thread. The child is successfully
spawned but you can't interact with it. The only way to make it work is
to spawn and interact with the child all in the same thread. [Adam
Kerrison] </p>
<h2><a name="echo_bug"></a>Timing issue with send() and sendline()</h2>
<p>This problem has been addressed and should not effect most users.</p>
<p>It is sometimes possible to read an echo of the string sent with <span
 class="code">send()</span> and <span class="code">sendline()</span>.
If you call <span class="code">sendline()</span> and then immediately
call <span class="code">readline()</span> you may get part of your
output echoed back. You may read back what you just wrote even if the
child application does not explicitly echo it. Timing is critical. This
could be a security issue when talking to an application that asks for
a password; otherwise, this does not seem like a big deal. <i>But why
do TTYs do this</i>?</p>
<p>People usually report this when they are trying to control SSH or
some other login. For example, if your code looks something like this: </p>
<pre class="code">child.expect ('[pP]assword:')<br>child.sendline (my_password)</pre>
<p><br>
<blockquote>
1. SSH prints "password:" prompt to the user.<br>
2. SSH turns off echo on the TTY device.<br>
3. SSH waits for user to enter a password.<br>
</blockquote>
When scripting with Pexpect what can happen is that Pexpect will response to the "password:" prompt
before SSH has had time to turn off TTY echo. In other words, Pexpect sends the password between
steps 1. and 2., so the password gets echoed back to the TTY. I would call this an SSH bug.
</p>
<p>
Pexpect now automatically adds a short delay before sending data to a child process.
This more closely mimics what happens in the usual human-to-app interaction.
The delay can be tuned with the 'delaybeforesend' attribute of the spawn class.
In general, this fixes the problem for everyone and so this should not be an issue
for most users. For some applications you might with to turn it off.
    child = pexpect.spawn ("ssh user@example.com")
    child.delaybeforesend = 0
</p>
<p><br>
</p>
<p>Try changing it to look like the following. I know that this fix
does not look correct, but it works. I have not figured out exactly
what is happening. You would think that the sleep should be after the
sendline(). The fact that the sleep helps when it's between the
expect() and the sendline() must be a clue.</p>
<pre class="code">child.expect ('[pP]assword:')<br>child.sendline (my_password)</pre>
<h2>Timing issue with isalive()</h2>
<p>Reading the state of isalive() immediately after a child exits may
sometimes return 1. This is a race condition. The child has closed its
file descriptor, but has not yet fully exited before Pexpect's
isalive() executes. Addings a slight delay before the isalive() will
help. In the following example <span class="code">isalive()</span>
sometimes returns 1:</p>
<blockquote>
  <pre class="code">child = pexpect.spawn('ls')<br>child.expect(pexpect.EOF)<br>print child.isalive()</pre>
</blockquote>
<p>But if there is any delay before the call to <span class="code">isalive()</span>
then it will always return 0 as expected.</p>
<blockquote>
  <pre class="code">child = pexpect.spawn('ls')<br>child.expect(pexpect.EOF)<br>time.sleep(0.1)<br>print child.isalive()</pre>
</blockquote>

<h2>Truncated output just before child exits</h2>
<p><i>So far I have seen this only on older versions of <b>Apple's MacOS X</b>.</i>
If the child application quits it may not flush its output buffer. This
means that your Pexpect application will receive an EOF even though it
should have received a little more data before the child died. This is
not generally a problem when talking to interactive child applications.
One example where it is a problem is when trying to read output from a
program like '<span class="code">ls</span>'. You may receive most of
the directory listing, but the last few lines will get lost before you
receive an EOF. The reason for this is that '<span class="code">ls</span>'
runs; completes its task; and then exits. The buffer is not flushed
before exit so the last few lines are lost. The following example
demonstrates the problem:</p>
<p> </p>
<blockquote>
  <pre class="code">child = pexpect.spawn ('ls -l')<br>child.expect (pexpect.EOF)<br>print child.before <br>  </pre>
</blockquote>
<p></p>

<h2>Controlling SSH on Solaris</h2>
<p>Pexpect does not yet work perfectly on Solaris.
One common problem is that SSH sometimes will not allow TTY password
authentication. For example, you may expect SSH to ask you for a
password using code like this:
</p>
<pre class="code">child = pexpect.spawn ('ssh user@example.com')<br>child.expect ('assword')<br>child.sendline ('mypassword')<br></pre>
You may see the following error come back from a spawned
child SSH:
<p></p>
<blockquote>Permission denied (publickey,keyboard-interactive). </blockquote>
<p>
This means that SSH thinks it can't access the TTY to ask you for your
password.
The only solution I have found is to use public key authentication with
SSH.
This bypasses the need for a password. I'm not happy with this
solution.
The problem is due to poor support for Solaris Pseudo TTYs in the
Python
Standard Library. </p>
<hr noshade="noshade" size="1">
<h1><a name="changes"></a>CHANGES</h1>
<h2>Current Release</h2>
<p>Fixed OSError exception when a pexpect object is cleaned up.
Previously you might have seen this exception:</p>
<blockquote>
  <pre class="code">Exception exceptions.OSError: (10, 'No child processes') <br>in &lt;bound method spawn.__del__ of<br>&lt;pexpect.spawn instance at 0xd248c&gt;&gt; ignored</pre>
</blockquote>
<p>You should not see that anymore. Thanks to Michael Surette.</p>
<p>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.</p>
<blockquote>
  <pre class="code">child = pexpect.spawn ('my_command')<br>child.maxread=1000 # Sets buffer to 1000 characters.</pre>
</blockquote>
<div>
<p>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 'before' property was set to everything before the
state was encountered, but if you let pexpect raise the exception then
'before' was not set. Now the 'before' property will get set either way
you choose to handle these states.</p>
<h2><i>Older changes...</i></h2>
<p>The spawn object now provides iterators for a <i>file-like interface</i>.
This makes Pexpect a more complete file-like object. You can now write
code like this:</p>
<blockquote>
  <pre class="code">child = pexpect.spawn ('ls -l')<br>for line in child:<br>    print line<br></pre>
</blockquote>
<p>I added the attribute <span class="code">exitstatus</span>. This
will give the exit code returned by the child process. This will be set
to <span class="code">None</span> while the child is still alive. When
<span class="code">isalive()</span> returns 0 then <span class="code">exitstatus</span>
will be set.</p>
<p>I made a few more tweaks to <span class="code">isalive()</span> so
that it will operate more consistently on different platforms. Solaris
is the most difficult to support.</p>
<p>&nbsp;</p>
<p>You can now put <span class="code">TIMEOUT</span> in a list of
expected patterns. This is just like putting <span class="code">EOF</span>
in the pattern list. Expecting for a <span class="code">TIMEOUT</span>
may not be used as often as <span class="code">EOF</span>, but this
makes Pexpect more consitent.</p>
<p>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 contsructing a spawn open. For
example on a Linux box with a modem on ttyS1:</p>
<blockquote>
  <pre class="code">fd = os.open("/dev/ttyS1", os.O_RDWR|os.O_NONBLOCK|os.O_NOCTTY)<br>m = pexpect.spawn(fd) # Note integer fd is used instead of usual string.<br>m.send("+++") # Escape sequence<br>m.send("ATZ0\r") # Reset modem to profile 0<br>rval = m.expect(["OK", "ERROR"])</pre>
</blockquote>
<h3>Pexpect now tests itself on Compile Farm!</h3>
<p>I wrote a nice script that uses ssh to connect to each machine on
Source Forge's Compile Farm and then run the testall.py script for each
platform. The result of the test is then recorded for each platform.
Now it's easy to run regression tests across multiple platforms.</p>
<h3>Pexpect is a file-like object</h3>
<p>The spawn object now provides a <i>file-like interface</i>. It
supports most of the methods and attributes defined for Python File
Objects. </p>
<p>I changed write and writelines() so that they 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.</p>
<p>read() was renamed to read_nonblocking(). I added a 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.</p>
<p>I changed the name of <span class="code">set_echo()</span> to <span
 class="code">setecho()</span>.</p>
<p>I changed the name of <span class="code">send_eof()</span> to <span
 class="code">sendeof()</span>.</p>
<p>I modified <span class="code">kill()</span> so that it checks to
make sure the pid isalive().</p>
<p>I modified <span class="code">spawn()</span> (really called from <span
 class="code">__spawn()</span>)so that it does not raise an expection
if <span class="code">setwinsize()</span> 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.</p>
<p>Added a method <span class="code">close()</span> 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. </p>
<p>Add variables <span class="code">__version__</span> and <span
 class="code">__revision__</span> (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.</p>
<h3>Logging changes</h3>
<blockquote>
  <p><span class="code">log_open()</span> and <span class="code">log_close()</span>
have been removed. Now use <span class="code">setlog()</span>. The <span
 class="code">setlog()</span> 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 <span
 class="code">setlog()</span> with None.</p>
</blockquote>
<h2>isalive changes</h2>
<blockquote>
  <p>I renamed the <span class="code">isAlive()</span> method to <span
 class="code">isalive()</span> 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's was a big headache. It still is, but I think it works
better now.</p>
</blockquote>
<h3>attribute name changes</h3>
<blockquote>
  <p>The names of some attributes have been changed. This effects the
names of the attributes that are set after called the <span
 class="code">expect()</span> method.</p>
  <table class="pymenu" border="0" cellpadding="5">
    <tbody>
      <tr>
        <th class="pymenu">NEW NAME</th>
        <th class="pymenu">OLD NAME</th>
      </tr>
      <tr>
        <td><span class="code">before</span><br>
        <i>Everything before the match.</i></td>
        <td><span class="code">before</span></td>
      </tr>
      <tr>
        <td><span class="code">after</span><br>
        <i>Everything after and including the first character of the
match</i></td>
        <td><span class="code">matched</span></td>
      </tr>
      <tr>
        <td><span class="code">match</span><br>
        <i>This is the re MatchObject from the match.<br>
You can get groups() from this.<br>
See '<span class="code">uptime.py</span>' in the examples tar ball.</i></td>
        <td><i>New -- Did not exist</i></td>
      </tr>
    </tbody>
  </table>
</blockquote>
<h3>EOF changes</h3>
<blockquote>
  <p>The <span class="code">expect_eof()</span> method is gone. You
can now simply use the <span class="code">expect()</span> method to
look for EOF.</p>
  <p>Was:</p>
  <blockquote>
    <p><span class="code">p.expect_eof ()</span></p>
  </blockquote>
  <p>Now:</p>
  <blockquote>
    <p><span class="code">p.expect (pexpect.EOF)</span></p>
  </blockquote>
</blockquote>
<hr noshade="noshade" size="1">
<h1><a name="testing"></a>TESTING</h1>
<p>The following platforms have been tested:</p>
<!--
<table class="pymenu" border="0" cellpadding="5">
  <tbody>
    <tr>
      <th class="pymenu">PLATFORM</th>
      <th class="pymenu">RESULTS</th>
    </tr>
    <tr>
      <td>Linux 2.4.9-ac10-rmk2-np1-cerf2<br>
armv4l</td>
      <td><b><i>all tests passed</i></b></td>
    </tr>
    <tr>
      <td>Linux 2.4.18 #2<br>
sparc64</td>
      <td><b><i>all tests passed</i></b></td>
    </tr>
    <tr>
      <td>MacOS X Darwin Kernel Version 5.5<br>
powerpc</td>
      <td>
      <p>failed more than one test.</p>
      <p>Generally Pexpect works on OS X, but the nature of the quirks
cause a many of the tests to fail. See <a href="#bugs">bugs</a>
(Incomplete Child Output). The problem is more than minor, but Pexpect
is still more than useful for most tasks. The problem is an edge case.</p>
      </td>
    </tr>
    <tr>
      <td>Linux 2.2.20<br>
alpha<br>
      </td>
      <td><b><i>all tests passed</i></b></td>
    </tr>
    <tr>
      <td>Linux 2.4.18-5smp<br>
i686</td>
      <td><b><i>all tests passed</i></b></td>
    </tr>
    <tr>
      <td>OpenBSD 2.9 GENERIC#653<br>
i386</td>
      <td><b><i>all tests passed</i></b></td>
    </tr>
    <tr>
      <td>Solaris</td>
      <td>
      <p>failed <span class="code">test_destructor</span></p>
      <p>Otherwise, this is working pretty well. The destructor problem
is minor. For some reason, the <i>second</i> 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. </p>
      </td>
    </tr>
    <tr>
      <td>Windows XP Cygwin</td>
      <td>failed <span class="code">test_destructor</span>. That it
works at all is amazing to me. Cygwin rules!</td>
    </tr>
  </tbody>
</table>
-->
<h1>&nbsp;</h1>
<h1><a name="todo">TO DO</a></h1>
<p>Add an option to add a delay after each expect() or before each
read()/readline() call to automatically avoid the <a href="#echo_bug">echo
bug</a>.</p>
<p>&nbsp;</p>
</div>
<hr noshade="noshade" size="1">
<table border="0">
  <tbody>
    <tr>
      <td> <a href="http://www.noah.org/email/"><img src="email.png"
 alt="Click to send email." border="0" height="16" width="100"></a> </td>
    </tr>
  </tbody>
</table>
</div>
<div id="Menu"><b>INDEX</b><br>
<hr noshade="noshade" size="1"> <a href="#license"
 title="Python Software Foundation License">License</a><br>
<a href="#download" title="Download and setup instructions">Download</a><br>
<a href="#doc" title="Documentation and overview">Documentation</a><br>
<a href="#status" title="Project Status">Project Status</a><br>
<a href="#requirements" title="System requirements to use Pexpect">Requirements</a><br>
<a href="#overview" title="Overview of what Pexpect does">Overview</a><br>
<a href="#faq" title="FAQ">FAQ</a><br>
<a href="#bugs" title="Bugs and work-arounds">Known Bugs</a><br>
<a href="#changes" title="What's new with Pexpect">Recent Changes</a><br>
<a href="#testing" title="Test results on various platforms">Testing</a><br>
<a href="#todo" title="What to do next">To do</a><br>
<a href="http://pexpect.svn.sourceforge.net/viewvc/pexpect/trunk/pexpect/" title="browse SVN">Browse SVN</a><br>
<br>
<a href="http://sourceforge.net/projects/pexpect/"
 title="The Pexpect project page on SourceForge.net"> <img
 src="http://sourceforge.net/sflogo.php?group_id=59762&amp;type=5"
 alt="The Pexpect project page on SourceForge.net" border="0"
 height="31" width="105"> </a> </div>
</body>
</html>