1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01 Transitional//EN"> 2 <html> 3 <head> 4 <title>Pexpect - a Pure Python Expect-like module</title> 5 <link rel="stylesheet" href="clean.css" type="text/css"> 6 <meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1"> 7 <meta name="Author" content="Noah Spurrier"> 8 <meta name="Keywords" 9 content="pexpect, Noah Spurrier, pypect, Python, Libes, TCL, Expect, pipe, popen, pyExpect, expectpy, expect-like, expect-alike, expect like"> 10 <meta name="Description" 11 content="Pexpect is a pure Python Expect-like module. Pexpect makes Python a better tool for controlling other applications."> 12 </head> 13 <body bgcolor="#ffffff" text="#000000"> 14 <div id="Header"> 15 <h1>Pexpect version VERSION<br> 16 a Pure Python Expect-like module 17 </h1> 18 </div> 19 <div id="Content"> 20 <p>Pexpect makes Python a better tool for controlling other 21 applications.</p> 22 <p>Pexpect is a pure Python module for spawning child applications; 23 controlling them; and responding to expected patterns in their output. 24 Pexpect works like Don Libes' Expect. Pexpect allows your script to 25 spawn a child application and control it as if a human were typing 26 commands.</p> 27 <p>Pexpect can be used for automating interactive applications such as 28 ssh, ftp, passwd, telnet, etc. It can be used to a automate setup 29 scripts for duplicating software package installations on different 30 servers. It can be used for automated software testing. Pexpect is in 31 the spirit of Don Libes' Expect, but Pexpect is pure Python. Unlike 32 other Expect-like modules for Python, Pexpect does not require TCL or 33 Expect nor does it require C extensions to be compiled. It should work 34 on any platform that supports the standard Python pty module. The 35 Pexpect interface was designed to be easy to use.</p> 36 <table border="0"> 37 <tbody> 38 <tr> 39 <td align="right" valign="top">Send questions to:</td> 40 <td align="left"><a href="http://www.noah.org/email/"><img 41 src="email.png" alt="Click to send email." border="0" height="16" 42 width="100"></a></td> 43 </tr> 44 </tbody> 45 </table> 46 <hr noshade="noshade" size="1"> 47 <h1><a name="license"></a>License: MIT style</h1> 48 <p> 49 Free, open source, and all that good stuff.<br> 50 <br> 51 Permission is hereby granted, free of charge, to any person obtaining a copy 52 of this software and associated documentation files (the "Software"), to deal 53 in the Software without restriction, including without limitation the rights 54 to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 55 copies of the Software, and to permit persons to whom the Software is 56 furnished to do so, subject to the following conditions:<br> 57 <br> 58 The above copyright notice and this permission notice shall be included in all 59 copies or substantial portions of the Software.<br> 60 <br> 61 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 62 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 63 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 64 IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, 65 DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 66 OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 67 USE OR OTHER DEALINGS IN THE SOFTWARE.<br> 68 <br> 69 Pexpect Copyright (c) 2008 Noah Spurrier<br> 70 http://pexpect.sourceforge.net/ 71 </p> 72 73 <hr noshade="noshade" size="1"> 74 <h1><a name="download"></a><a 75 href="http://sourceforge.net/project/showfiles.php?group_id=59762">Download</a></h1> 76 <p>Download the <a 77 href="http://sourceforge.net/project/showfiles.php?group_id=59762"> 78 current version here</a> from the SourceForge site. Grab the current Pexpect tarball. 79 </p> 80 <h2>Installing Pexpect</h2> 81 <p>The Pexpect tarball is a standard Python Distutil distribution.</p> 82 <ol> 83 <li>download <span class="code">pexpect-VERSION.tar.gz</span></li> 84 <li><span class="code">tar zxf pexpect-VERSION.tar.gz</span></li> 85 <li><span class="code">cd pexpect-VERSION</span></li> 86 <li><span class="code">python setup.py install</span> <i>do this as root</i></li> 87 </ol> 88 <h2>Examples</h2> 89 <p> 90 Under the <span class="code">pexpect-VERSION</span> directory you should find 91 the <span class="code">examples</span> directory. 92 This is the best way to learn to use Pexpect. 93 See the descriptions of <a href="examples.html">Pexpect Examples</a>. 94 </p> 95 <h2><a name="doc"></a>API Documentation</h2> 96 <p> 97 <blockquote> 98 <a href="pexpect.html">pexpect</a> This is the main module that you want.<br> 99 <a href="pxssh.html">pxssh</a> Pexpect SSH is an extension of 'pexpect.spawn' that specializes in SSH.<br> 100 </blockquote> 101 the following are experimental extensions to Pexpect<br> 102 <blockquote> 103 <a href="fdpexpect.html">fdpexpect</a> fdpexpect extension of 'pexpect.spawn' that uses an open file descriptor.<br> 104 <a href="screen.html">SCREEN</a> This represents a virtual 'screen'.<br> 105 <a href="ANSI.html">ANSI</a> This parses ANSI/VT-100 terminal escape codes.<br> 106 <a href="FSM.html">FSM</a> This is a finite state machine used by ANSI.<br> 107 </blockquote> 108 </p> 109 <hr noshade="noshade" size="1"> 110 <h1><a name="status"></a>Project Status</h1> 111 <p>Automated pyunit tests reach over 80% 112 code coverage on pexpect.py. I regularly test on Linux and BSD 113 platforms. I try to test on Solaris and Irix. 114 </p> 115 <hr noshade="noshade" size="1"> 116 <h1><a name="requirements"></a>Requirements for use of Pexpect</h1> 117 <h2>Python</h2> 118 <blockquote> 119 <p>Pexpect was written and tested with Python 2.4. It should work on 120 earlier versions that have the <span class="code">pty</span> module. I 121 sometimes even manually test it with Python 1.5.2, but I can't easily 122 run the PyUnit test framework against Python 1.5.2, so I have less 123 confidence in Pexpect on Python 1.5.2.</p> 124 </blockquote> 125 <h2>pty module</h2> 126 <blockquote> 127 <p>Any POSIX system (UNIX) with a working <span class="code">pty</span> 128 module should be able to run Pexpect. The <span class="code">pty</span> 129 module is part of the Standard Python Library, so if you are running on 130 a POSIX system you should have it. The <span class="code">pty</span> 131 module does not run the same on all platforms. It should be solid on Linux 132 and BSD systems. I have taken effort to try to smooth the wrinkles out of the different platforms. To learn more 133 about the wrinkles see <a href="#bugs">Bugs</a> and <a href="#testing">Testing</a>.</p> 134 </blockquote> 135 <p>Pexpect does not currently work on the standard Windows Python (see 136 the pty requirement); however, it seems to work fine using <a 137 href="http://www.cygwin.com/">Cygwin</a>. It is possible to build 138 something like a pty for Windows, but it would have to use a different 139 technique that I am still investigating. I know it's possible because 140 Libes' Expect was ported to Windows. <i>If you have any ideas or 141 skills to contribute in this area then I would really appreciate some 142 tips on how to approach this problem.</i> </p> 143 <hr noshade="noshade" size="1"> 144 <h1><a name="overview"></a>Overview</h1> 145 <p>Pexpect can be used for automating interactive applications such as 146 ssh, ftp, mencoder, passwd, etc. The Pexpect interface was designed to be 147 easy to use. Here is an example of Pexpect in action:</p> 148 <blockquote> 149 <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 (a] example.com')<br>child.expect ('ftp> ')<br>child.sendline ('cd pub')<br>child.expect('ftp> ')<br>child.sendline ('get ls-lR.gz')<br>child.expect('ftp> ')<br>child.sendline ('bye')<br></pre> 150 </blockquote> 151 <p> Obviously you could write an ftp client using Python's own <span 152 class="code">ftplib</span> module, but this is just a demonstration. 153 You can use this technique with any application. This is especially 154 handy if you are writing automated test tools.</p> 155 156 <p>There are two important methods in Pexpect -- <span class="code"><b>expect()</b></span> 157 and <span class="code"><b>send()</b></span> (or <span class="code">sendline()</span> 158 which is like <span class="code">send()</span> with a linefeed). 159 The <span class="code">expect()</span> method waits for the child application 160 to return a given string. The string you specify is a regular expression, so 161 you can match complicated patterns. The <span class="code"><b>send()</b></span> method 162 writes a string to the child application. From the child's point of 163 view it looks just like someone typed the text from a terminal. After 164 each call to <span class="code"><b>expect()</b></span> the <span 165 class="code"><b>before</b></span> and <span class="code"><b>after</b></span> 166 properties will be set to the text printed by child application. The <span 167 class="code"><b>before</b></span> property will contain all text up to 168 the expected string pattern. The <span class="code"><b>after</b></span> string 169 will contain the text that was matched by the expected pattern. 170 The <span class="code">match</span> property is set to the <span class="code">re MatchObject</span>. 171 </p> 172 173 <p>An example of Pexpect in action may make things more clear. This example uses 174 <span class="code">ftp</span> to login to the OpenBSD site; list files 175 in a directory; and then pass interactive control of the ftp session to 176 the human user.</p> 177 <blockquote> 178 <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 (a] example.com')<br>child.expect ('ftp> ')<br>child.sendline ('ls /pub/OpenBSD/')<br>child.expect ('ftp> ')<br>print child.before # Print the result of the ls command.<br>child.interact() # Give control of the child to the user.<br></pre> 179 </blockquote> 180 <h2>Special EOF and TIMEOUT patterns</h2> 181 <p> 182 There are two special patterns to match the End Of File or a Timeout condition. 183 You you can pass these patterns to <span class="code">expect()</span>. 184 These patterns are not regular expressions. Use them like predefined constants. 185 </p> 186 <p>If the child has died and you have read all the child's output then ordinarily 187 <span class="code">expect()</span> will raise an <span class="code">EOF</span> 188 exception. You can read everything up to the EOF without generating an 189 exception by using the EOF pattern <span class="code">expect(pexpect.EOF)</span>. 190 In this case everything the child has output will be available in the <span 191 class="code">before</span> property.</p> 192 <p>The pattern given to <span class="code">expect()</span> may be a 193 regular expression or it may also be a <b>list</b> of regular expressions. 194 This allows you to match multiple optional responses. The <span class="code">expect()</span> 195 method returns the index of the pattern that was matched. For example, 196 say you wanted to login to a server. After entering a password you 197 could get various responses from the server -- your password could be 198 rejected; or you could be allowed in and asked for your terminal type; 199 or you could be let right in and given a command prompt. The following 200 code fragment gives an example of this:</p> 201 <blockquote> 202 <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> 203 </blockquote> 204 <p>If nothing matches an expected pattern then expect will eventually 205 raise a TIMEOUT exception. The default time is 30 seconds, but you can 206 change this by passing a timeout argument to expect():</p> 207 <blockquote> 208 <pre class="code"># Wait no more than 2 minutes (120 seconds) for password prompt.<br>child.expect('password:', timeout=120)</pre> 209 </blockquote> 210 <h2>Find the end of line -- CR/LF conventions<br> 211 Matching at the end of a line can be tricky<br> 212 $ regex pattern is useless.<br> 213 </h2> 214 <p>Pexpect matches regular expressions a little differently than what 215 you might be used to. 216 </p> 217 <p><i><b>The $ pattern for end of line match is useless</b></i>. 218 The $ matches the end of string, but Pexpect reads from the child 219 one character at a time, so each character looks like the end of a line. 220 Pexpect can't do a look-ahead into the child's output stream. 221 In general you would have this situation when using regular expressions 222 with any stream.<br> 223 <i>Note, pexpect does have an internal buffer, so reads are faster 224 than one character at a time, but from the user's perspective the regex 225 patterns test happens one character at a time.</i></p> 226 <p>The best way to match the end of a line is to look for the 227 newline: "\r\n" (CR/LF). Yes, that does appear to be DOS-style. 228 It may surprise some UNIX people to learn that terminal TTY device drivers 229 (dumb, vt100, ANSI, xterm, etc.) all use the CR/LF combination to signify 230 the end of line. Pexpect uses a Pseudo-TTY device to talk to the child application, so 231 when the child app prints "\n" you actually see "\r\n". 232 </p> 233 <p><b>UNIX uses just linefeeds to end lines of text, but not when it 234 comes to TTY devices!</b> TTY devices are more like the Windows world. 235 Each line of text end with a CR/LF combination. When you intercept data 236 from a UNIX command from a TTY device you will find that the TTY device 237 outputs a CR/LF combination. A UNIX command may only write a linefeed 238 (\n), but the TTY device driver converts it to CR/LF. This means that 239 your terminal will see lines end with CR/LF (hex <span class="code">0D 0A</span>). 240 Since Pexpect emulates a terminal, to match ends of lines you have to 241 expect the CR/LF combination.</p> 242 <blockquote> 243 <p class="code">child.expect ('\r\n')</p> 244 </blockquote> 245 <p>If you just need to skip past a new line then <span class="code">expect 246 ('\n')</span> by itself will work, but if you are expecting a specific 247 pattern before the end of line then you need to explicitly look for the 248 \r. For example the following expects a word at the end of a line:</p> 249 <blockquote> 250 <p class="code">child.expect ('\w+\r\n')</p> 251 </blockquote> 252 <p>But the following would both fail:</p> 253 <blockquote> 254 <p class="code">child.expect ('\w+\n')</p> 255 </blockquote> 256 <p>And as explained before, trying to use '$' to match the end of line 257 would not work either:</p> 258 <blockquote> 259 <p class="code">child.expect ('\w+$')</p> 260 </blockquote> 261 <p>So if you need to explicitly look for the END OF LINE, you want to 262 look for the CR/LF combination -- not just the LF and not the $ pattern.</p> 263 <p>This problem is not limited to Pexpect. This problem happens any 264 time you try to perform a regular expression match on a stream. Regular 265 expressions need to look ahead. With a stream it is hard to look ahead 266 because the process generating the stream may not be finished. There is no 267 way to know if the process has paused momentarily or is finished and 268 waiting for you. <font color="#cc0000">Pexpect must implicitly always 269 do a NON greedy match (minimal) at the end of a input {### already said 270 this}.</font> </p> 271 <p>Pexpect compiles all regular expressions with the DOTALL flag. With 272 the DOTALL flag a "." will match a newline. See the Python <a 273 href="http://www.python.org/doc/current/lib/node115.html#l2h-733">documentation</a></p> 274 <h2>Beware of + and * at the end of input.</h2> 275 <p>Remember that any time you try to match a pattern that needs 276 look-ahead that you will always get a minimal match (non greedy). For 277 example, the following will always return just one character:</p> 278 <blockquote> 279 <p class="code">child.expect ('.+')</p> 280 </blockquote> 281 <p>This example will match successfully, but will always return no 282 characters:</p> 283 <blockquote> 284 <p class="code">child.expect ('.*')</p> 285 </blockquote> 286 <p>Generally any star * expression will match as little as possible</p> 287 <p>One thing you can do is to try to force a non-ambiguous character at 288 the end of your <span class="code">\d+</span> pattern. Expect that 289 character to delimit the string. For example, you might try making the 290 end of your pattrn be <span class="code">\D+</span> instead of <span 291 class="code">\D*</span>. That means number digits alone would not 292 satisfy the (<span class="code">\d+</span>) pattern. You would need 293 some number(s) and at least one <span class="code">\D</span> at the 294 end. </p> 295 <h2>Matching groups</h2> 296 <p>You can group regular expression using parenthesis. After a match, 297 the <span class="code">match</span> parameter of the spawn object will 298 contain the Python Match object. </p> 299 <h2>Examples</h2> 300 <p>Using "match" and groups...</p> 301 <h2>Debugging</h2> 302 <p>If you get the string value of a pexpect.spawn object you will get 303 lots of useful debugging information. For debugging it's very useful to 304 use the following pattern:</p> 305 <p>try:<br> 306 i = child.expect ([pattern1, pattern2, pattern3, 307 etc])<br> 308 except:<br> 309 print "Exception was thrown"<br> 310 print "debug information:"<br> 311 print str(child)<br> 312 </p> 313 <p>It is also useful to log the child's input and out to a file or the 314 screen. The following will turn on logging and send output to stdout 315 (the screen).<br> 316 </p> 317 <p> child = pexpect.spawn (foo)<br> 318 child.logfile = sys.stdout<br> 319 <br> 320 </p> 321 <hr noshade="noshade" size="1"> 322 <h1>Exceptions</h1> 323 <p><b>EOF</b></p> 324 <p>Note that two flavors of EOF Exception may be thrown. They are 325 virtually identical except for the message string. For practical 326 purposes you should have no need to distinguish between them, but they 327 do give a little extra information about what type of platform you are 328 running. The two messages are:</p> 329 <blockquote> 330 <p class="code">End Of File (EOF) in read(). Exception style platform.</p> 331 <p class="code">End Of File (EOF) in read(). Empty string style 332 platform.</p> 333 </blockquote> 334 <p>Some UNIX platforms will throw an exception when you try to read 335 from a file descriptor in the EOF state. Other UNIX platforms instead 336 quietly return an empty string to indicate that the EOF state has been 337 reached.</p> 338 <p><b>Expecting EOF</b></p> 339 <p>If you wish to read up to the end of the child's output without 340 generating an <span class="code">EOF</span> exception then use the <span 341 class="code">expect(pexpect.EOF)</span> method.</p> 342 <p><b>TIMEOUT</b></p> 343 <p>The <span class="code">expect()</span> and <span class="code">read()</span> 344 methods will also timeout if the child does not generate any output for 345 a given amount of time. If this happens they will raise a <span 346 class="code">TIMEOUT</span> exception. You can have these method 347 ignore a timeout and block indefinitely by passing None for the timeout 348 parameter.</p> 349 <blockquote> 350 <p class="code">child.expect(pexpect.EOF, timeout=None)</p> 351 </blockquote> 352 <hr noshade="noshade" size="1"> 353 <h1><a name="faq"></a>FAQ</h1> 354 <p><b>Q: Why don't shell pipe and redirect (| and >) work when I 355 spawn a command?</b></p> 356 <p> 357 358 A: Remember that Pexpect does NOT interpret shell meta characters such as 359 redirect, pipe, or wild cards (>, |, or *). That's done by a shell not the 360 command you are spawning. This is a common mistake. If you want to run a 361 command and pipe it through another command then you must also start a shell. 362 For example: 363 364 <pre> 365 child = pexpect.spawn('/bin/bash -c "ls -l | grep LOG > log_list.txt"') 366 child.expect(pexpect.EOF) 367 </pre> 368 369 The second form of spawn (where you pass a list of arguments) is useful in 370 situations where you wish to spawn a command and pass it its own argument list. 371 This can make syntax more clear. For example, the following is equivalent to 372 the previous example: 373 374 <pre> 375 shell_cmd = 'ls -l | grep LOG > log_list.txt' 376 child = pexpect.spawn ('/bin/bash', ['-c', shell_cmd]) 377 child.expect (pexpect.EOF) 378 </pre> 379 380 </p> 381 <p><b>Q: Isn't there already a Python Expect?</b></p> 382 <p>A: Yes, there are several of them. They usually require you to 383 compile C. I wanted something that was pure Python and preferably a 384 single module that was simple to install. I also wanted something that 385 was easy to use. This pure Python expect only recently became possible 386 with the introduction of the pty module in the standard Python library. 387 Previously C extensions were required.</p> 388 389 <p><strong>Q: The before and after properties sound weird.</strong></p> 390 <p>Originally I was going to model Pexpect more after Expect, but then 391 I found that I could never remember how to get the context of the stuff 392 I was trying to parse. I hate having to read my own documentation. I 393 decided that it was easier for me to remember what before and after 394 was. It just so happens that this is how the -B and -A options in grep 395 works, so that made it even easier for me to remember. Whatever makes 396 my life easier is what's best.</p> 397 398 <p><b>Q: Why not just use Expect?</b></p> 399 <p>A: I love it. It's great. I has bailed me out of some real jams, but 400 I wanted something that would do 90% of what I need from Expect; be 10% 401 of the size; and allow me to write my code in Python instead of TCL. 402 Pexpect is not nearly as big as Expect, but Pexpect does everything I 403 have ever used Expect for. 404 <!-- :-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> 405 406 <p><b>Q: Why not just use a pipe (popen())?</b></p> 407 <p>A: A pipe works fine for getting the output to non-interactive 408 programs. If you just want to get the output from <span class="code">ls</span>, 409 <span class="code">uname</span>, or <span class="code">ping</span> 410 then this works. Pipes do not work very well for interactive programs 411 and pipes will almost certainly fail for most applications that ask for 412 passwords such as telnet, ftp, or ssh.</p> 413 <p>There are two reasons for this. </p> 414 <p>First an application may bypass stdout and print directly to its 415 controlling TTY. Something like SSH will do this when it asks you for a 416 password. This is why you cannot redirect the password prompt because 417 it does not go through stdout or stderr.</p> 418 <p>The second reason is because most applications are built using the C 419 Standard IO Library (anything that uses <span class="code">#include 420 <stdio.h></span>). One of the features of the stdio library is 421 that it buffers all input and output. Normally output is <b><i>line 422 buffered</i></b> when a program is printing to a TTY (your terminal 423 screen). Every time the program prints a line-feed the currently 424 buffered data will get printed to your screen. The problem comes when 425 you connect a pipe. The stdio library is smart and can tell that it is 426 printing to a pipe instead of a TTY. In that case it switches from line 427 buffer mode to <i><b>block buffered</b></i>. In this mode the 428 currently buffered data is flushed when the buffer is full. This causes 429 most interactive programs to deadlock. Block buffering is more 430 efficient when writing to disks and pipes. Take the situation where a 431 program prints a message "Enter your user name:\n" and then waits for 432 you type type something. In block buffered mode, the stdio library will 433 not put the message into the pipe even though a linefeed is printed. 434 The result is that you never receive the message, yet the child 435 application will sit and wait for you to type a response. Don't confuse 436 the stdio lib's buffer with the pipe's buffer. The pipe buffer is 437 another area that can cause problems. You could flush the input side of 438 a pipe, whereas you have no control over the stdio library buffer. </p> 439 <p>More information: the Standard IO library has three states for a 440 FILE *. These are: _IOFBF for block buffered; _IOLBF for line buffered; 441 and _IONBF for unbuffered. The STDIO lib will use block buffering when 442 talking to a block file descriptor such as a pipe. This is usually not 443 helpful for interactive programs. Short of recompiling your program to 444 include fflush() everywhere or recompiling a custom stdio library there 445 is not much a controlling application can do about this if talking over 446 a pipe.</p> 447 <p> The program may have put data in its output that remains unflushed 448 because the output buffer is not full; then the program will go and 449 deadlock while waiting for input -- because you never send it any 450 because you are still waiting for its output (still stuck in the 451 STDIO's output buffer).</p> 452 <p>The answer is to use a pseudo-tty. A TTY device will force <i><b>line</b></i> 453 buffering (as opposed to block buffering). Line buffering means that 454 you will get each line when the child program sends a line feed. This 455 corresponds to the way most interactive programs operate -- send a line 456 of output then wait for a line of input.</p> 457 <p>I put "answer" in quotes because it's ugly solution and because 458 there is no POSIX standard for pseudo-TTY devices (even though they 459 have a TTY standard...). What would make more sense to me would be to 460 have some way to set a mode on a file descriptor so that it will tell 461 the STDIO to be line-buffered. I have investigated, and I don't think 462 there is a way to set the buffered state of a child process. The STDIO 463 Library does not maintain any external state in the kernel or whatnot, 464 so I don't think there is any way for you to alter it. I'm not quite 465 sure how this line-buffered/block-buffered state change happens 466 internally in the STDIO library. I think the STDIO lib looks at the 467 file descriptor and decides to change behavior based on whether it's a 468 TTY or a block file (see isatty()).</p> 469 <p>I hope that this qualifies as helpful.</p> 470 471 <h1>Don't use a pipe to control another application...</h1> 472 <p>Pexpect may seem similar to <span class="code">os.popen()</span> or 473 <span class="code">commands</span> module. The main difference is that 474 Pexpect (like Expect) uses a pseudo-TTY to talk to the child 475 application. Most applications do no work well through the system() 476 call or through pipes. And probably all applications that ask a user to 477 type in a password will fail. These applications bypass the stdin and 478 read directly from the TTY device. Many applications do not explicitly 479 flush their output buffers. This causes deadlocks if you try to control 480 an interactive application using a pipe. What happens is that most UNIX 481 applications use the stdio (#include <stdio.h>) for input and 482 output. The stdio library behaves differently depending on where the 483 output is going. There is no way to control this behavior from the 484 client end.<br> 485 </p> 486 487 <p><b>Q: Can I do screen scraping with this thing?</b></p> 488 <p>A: That depends. If your application just does line-oriented output 489 then this is easy. If it does screen-oriented output then it may work, 490 but it could be hard. For example, trying to scrape data from the 'top' 491 command would be hard. The top command repaints the text window. </p> 492 <p>I am working on an ANSI / VT100 terminal emulator that will have 493 methods to get characters from an arbitrary X,Y coordinate of the 494 virtual screen. It works and you can play with it, but I have no 495 working examples at this time.</p> 496 <hr noshade="noshade" size="1"> 497 <h1><a name="bugs"></a>Bugs</h1> 498 <h2>Threads</h2> 499 <p>On Linux (RH 8) you cannot spawn a child from a different thread and 500 pass the handle back to a worker thread. The child is successfully 501 spawned but you can't interact with it. The only way to make it work is 502 to spawn and interact with the child all in the same thread. [Adam 503 Kerrison] </p> 504 <h2><a name="echo_bug"></a>Timing issue with send() and sendline()</h2> 505 <p>This problem has been addressed and should not effect most users.</p> 506 <p>It is sometimes possible to read an echo of the string sent with <span 507 class="code">send()</span> and <span class="code">sendline()</span>. 508 If you call <span class="code">sendline()</span> and then immediately 509 call <span class="code">readline()</span> you may get part of your 510 output echoed back. You may read back what you just wrote even if the 511 child application does not explicitly echo it. Timing is critical. This 512 could be a security issue when talking to an application that asks for 513 a password; otherwise, this does not seem like a big deal. <i>But why 514 do TTYs do this</i>?</p> 515 <p>People usually report this when they are trying to control SSH or 516 some other login. For example, if your code looks something like this: </p> 517 <pre class="code">child.expect ('[pP]assword:')<br>child.sendline (my_password)</pre> 518 <p><br> 519 <blockquote> 520 1. SSH prints "password:" prompt to the user.<br> 521 2. SSH turns off echo on the TTY device.<br> 522 3. SSH waits for user to enter a password.<br> 523 </blockquote> 524 When scripting with Pexpect what can happen is that Pexpect will response to the "password:" prompt 525 before SSH has had time to turn off TTY echo. In other words, Pexpect sends the password between 526 steps 1. and 2., so the password gets echoed back to the TTY. I would call this an SSH bug. 527 </p> 528 <p> 529 Pexpect now automatically adds a short delay before sending data to a child process. 530 This more closely mimics what happens in the usual human-to-app interaction. 531 The delay can be tuned with the 'delaybeforesend' attribute of the spawn class. 532 In general, this fixes the problem for everyone and so this should not be an issue 533 for most users. For some applications you might with to turn it off. 534 child = pexpect.spawn ("ssh user (a] example.com") 535 child.delaybeforesend = 0 536 </p> 537 <p><br> 538 </p> 539 <p>Try changing it to look like the following. I know that this fix 540 does not look correct, but it works. I have not figured out exactly 541 what is happening. You would think that the sleep should be after the 542 sendline(). The fact that the sleep helps when it's between the 543 expect() and the sendline() must be a clue.</p> 544 <pre class="code">child.expect ('[pP]assword:')<br>child.sendline (my_password)</pre> 545 <h2>Timing issue with isalive()</h2> 546 <p>Reading the state of isalive() immediately after a child exits may 547 sometimes return 1. This is a race condition. The child has closed its 548 file descriptor, but has not yet fully exited before Pexpect's 549 isalive() executes. Addings a slight delay before the isalive() will 550 help. In the following example <span class="code">isalive()</span> 551 sometimes returns 1:</p> 552 <blockquote> 553 <pre class="code">child = pexpect.spawn('ls')<br>child.expect(pexpect.EOF)<br>print child.isalive()</pre> 554 </blockquote> 555 <p>But if there is any delay before the call to <span class="code">isalive()</span> 556 then it will always return 0 as expected.</p> 557 <blockquote> 558 <pre class="code">child = pexpect.spawn('ls')<br>child.expect(pexpect.EOF)<br>time.sleep(0.1)<br>print child.isalive()</pre> 559 </blockquote> 560 561 <h2>Truncated output just before child exits</h2> 562 <p><i>So far I have seen this only on older versions of <b>Apple's MacOS X</b>.</i> 563 If the child application quits it may not flush its output buffer. This 564 means that your Pexpect application will receive an EOF even though it 565 should have received a little more data before the child died. This is 566 not generally a problem when talking to interactive child applications. 567 One example where it is a problem is when trying to read output from a 568 program like '<span class="code">ls</span>'. You may receive most of 569 the directory listing, but the last few lines will get lost before you 570 receive an EOF. The reason for this is that '<span class="code">ls</span>' 571 runs; completes its task; and then exits. The buffer is not flushed 572 before exit so the last few lines are lost. The following example 573 demonstrates the problem:</p> 574 <p> </p> 575 <blockquote> 576 <pre class="code">child = pexpect.spawn ('ls -l')<br>child.expect (pexpect.EOF)<br>print child.before <br> </pre> 577 </blockquote> 578 <p></p> 579 580 <h2>Controlling SSH on Solaris</h2> 581 <p>Pexpect does not yet work perfectly on Solaris. 582 One common problem is that SSH sometimes will not allow TTY password 583 authentication. For example, you may expect SSH to ask you for a 584 password using code like this: 585 </p> 586 <pre class="code">child = pexpect.spawn ('ssh user (a] example.com')<br>child.expect ('assword')<br>child.sendline ('mypassword')<br></pre> 587 You may see the following error come back from a spawned 588 child SSH: 589 <p></p> 590 <blockquote>Permission denied (publickey,keyboard-interactive). </blockquote> 591 <p> 592 This means that SSH thinks it can't access the TTY to ask you for your 593 password. 594 The only solution I have found is to use public key authentication with 595 SSH. 596 This bypasses the need for a password. I'm not happy with this 597 solution. 598 The problem is due to poor support for Solaris Pseudo TTYs in the 599 Python 600 Standard Library. </p> 601 <hr noshade="noshade" size="1"> 602 <h1><a name="changes"></a>CHANGES</h1> 603 <h2>Current Release</h2> 604 <p>Fixed OSError exception when a pexpect object is cleaned up. 605 Previously you might have seen this exception:</p> 606 <blockquote> 607 <pre class="code">Exception exceptions.OSError: (10, 'No child processes') <br>in <bound method spawn.__del__ of<br><pexpect.spawn instance at 0xd248c>> ignored</pre> 608 </blockquote> 609 <p>You should not see that anymore. Thanks to Michael Surette.</p> 610 <p>Added support for buffering reads. This greatly improves speed when 611 trying to match long output from a child process. When you create an 612 instance of the spawn object you can then set a buffer size. For now 613 you MUST do the following to turn on buffering -- it may be on by 614 default in future version.</p> 615 <blockquote> 616 <pre class="code">child = pexpect.spawn ('my_command')<br>child.maxread=1000 # Sets buffer to 1000 characters.</pre> 617 </blockquote> 618 <div> 619 <p>I made a subtle change to the way TIMEOUT and EOF exceptions behave. 620 Previously you could either expect these states in which case pexpect 621 will not raise an exception, or you could just let pexpect raise an 622 exception when these states were encountered. If you expected the 623 states then the 'before' property was set to everything before the 624 state was encountered, but if you let pexpect raise the exception then 625 'before' was not set. Now the 'before' property will get set either way 626 you choose to handle these states.</p> 627 <h2><i>Older changes...</i></h2> 628 <p>The spawn object now provides iterators for a <i>file-like interface</i>. 629 This makes Pexpect a more complete file-like object. You can now write 630 code like this:</p> 631 <blockquote> 632 <pre class="code">child = pexpect.spawn ('ls -l')<br>for line in child:<br> print line<br></pre> 633 </blockquote> 634 <p>I added the attribute <span class="code">exitstatus</span>. This 635 will give the exit code returned by the child process. This will be set 636 to <span class="code">None</span> while the child is still alive. When 637 <span class="code">isalive()</span> returns 0 then <span class="code">exitstatus</span> 638 will be set.</p> 639 <p>I made a few more tweaks to <span class="code">isalive()</span> so 640 that it will operate more consistently on different platforms. Solaris 641 is the most difficult to support.</p> 642 <p> </p> 643 <p>You can now put <span class="code">TIMEOUT</span> in a list of 644 expected patterns. This is just like putting <span class="code">EOF</span> 645 in the pattern list. Expecting for a <span class="code">TIMEOUT</span> 646 may not be used as often as <span class="code">EOF</span>, but this 647 makes Pexpect more consitent.</p> 648 <p>Thanks to a suggestion and sample code from Chad J. Schroeder I 649 added the ability for Pexpect to operate on a file descriptor that is 650 already open. This means that Pexpect can be used to control streams 651 such as those from serial port devices. Now you just pass the integer 652 file descriptor as the "command" when contsructing a spawn open. For 653 example on a Linux box with a modem on ttyS1:</p> 654 <blockquote> 655 <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> 656 </blockquote> 657 <h3>Pexpect now tests itself on Compile Farm!</h3> 658 <p>I wrote a nice script that uses ssh to connect to each machine on 659 Source Forge's Compile Farm and then run the testall.py script for each 660 platform. The result of the test is then recorded for each platform. 661 Now it's easy to run regression tests across multiple platforms.</p> 662 <h3>Pexpect is a file-like object</h3> 663 <p>The spawn object now provides a <i>file-like interface</i>. It 664 supports most of the methods and attributes defined for Python File 665 Objects. </p> 666 <p>I changed write and writelines() so that they no longer return a 667 value. Use send() if you need that functionality. I did this to make 668 the Spawn object more closely match a file-like object.</p> 669 <p>read() was renamed to read_nonblocking(). I added a new read() 670 method that matches file-like object interface. In general, you should 671 not notice the difference except that read() no longer allows you to 672 directly set the timeout value. I hope this will not effect any 673 existing code. Switching to read_nonblocking() should fix existing code.</p> 674 <p>I changed the name of <span class="code">set_echo()</span> to <span 675 class="code">setecho()</span>.</p> 676 <p>I changed the name of <span class="code">send_eof()</span> to <span 677 class="code">sendeof()</span>.</p> 678 <p>I modified <span class="code">kill()</span> so that it checks to 679 make sure the pid isalive().</p> 680 <p>I modified <span class="code">spawn()</span> (really called from <span 681 class="code">__spawn()</span>)so that it does not raise an expection 682 if <span class="code">setwinsize()</span> fails. Some platforms such 683 as Cygwin do not like setwinsize. This was a constant problem and since 684 it is not a critical feature I decided to just silence the error. 685 Normally I don't like to do that, but in this case I'm making an 686 exception.</p> 687 <p>Added a method <span class="code">close()</span> that does what you 688 think. It closes the file descriptor of the child application. It makes 689 no attempt to actually kill the child or wait for its status. </p> 690 <p>Add variables <span class="code">__version__</span> and <span 691 class="code">__revision__</span> (from cvs) to the pexpect modules. 692 This is mainly helpful to me so that I can make sure that I'm testing 693 with the right version instead of one already installed.</p> 694 <h3>Logging changes</h3> 695 <blockquote> 696 <p><span class="code">log_open()</span> and <span class="code">log_close()</span> 697 have been removed. Now use <span class="code">setlog()</span>. The <span 698 class="code">setlog()</span> method takes a file object. This is far 699 more flexible than the previous log method. Each time data is written 700 to the file object it will be flushed. To turn logging off simply call <span 701 class="code">setlog()</span> with None.</p> 702 </blockquote> 703 <h2>isalive changes</h2> 704 <blockquote> 705 <p>I renamed the <span class="code">isAlive()</span> method to <span 706 class="code">isalive()</span> to match the more typical naming style 707 in Python. Also the technique used to detect child process status has 708 been drastically modified. Previously I did some funky stuff with 709 signals which caused indigestion in other Python modules on some 710 platforms. It's was a big headache. It still is, but I think it works 711 better now.</p> 712 </blockquote> 713 <h3>attribute name changes</h3> 714 <blockquote> 715 <p>The names of some attributes have been changed. This effects the 716 names of the attributes that are set after called the <span 717 class="code">expect()</span> method.</p> 718 <table class="pymenu" border="0" cellpadding="5"> 719 <tbody> 720 <tr> 721 <th class="pymenu">NEW NAME</th> 722 <th class="pymenu">OLD NAME</th> 723 </tr> 724 <tr> 725 <td><span class="code">before</span><br> 726 <i>Everything before the match.</i></td> 727 <td><span class="code">before</span></td> 728 </tr> 729 <tr> 730 <td><span class="code">after</span><br> 731 <i>Everything after and including the first character of the 732 match</i></td> 733 <td><span class="code">matched</span></td> 734 </tr> 735 <tr> 736 <td><span class="code">match</span><br> 737 <i>This is the re MatchObject from the match.<br> 738 You can get groups() from this.<br> 739 See '<span class="code">uptime.py</span>' in the examples tar ball.</i></td> 740 <td><i>New -- Did not exist</i></td> 741 </tr> 742 </tbody> 743 </table> 744 </blockquote> 745 <h3>EOF changes</h3> 746 <blockquote> 747 <p>The <span class="code">expect_eof()</span> method is gone. You 748 can now simply use the <span class="code">expect()</span> method to 749 look for EOF.</p> 750 <p>Was:</p> 751 <blockquote> 752 <p><span class="code">p.expect_eof ()</span></p> 753 </blockquote> 754 <p>Now:</p> 755 <blockquote> 756 <p><span class="code">p.expect (pexpect.EOF)</span></p> 757 </blockquote> 758 </blockquote> 759 <hr noshade="noshade" size="1"> 760 <h1><a name="testing"></a>TESTING</h1> 761 <p>The following platforms have been tested:</p> 762 <!-- 763 <table class="pymenu" border="0" cellpadding="5"> 764 <tbody> 765 <tr> 766 <th class="pymenu">PLATFORM</th> 767 <th class="pymenu">RESULTS</th> 768 </tr> 769 <tr> 770 <td>Linux 2.4.9-ac10-rmk2-np1-cerf2<br> 771 armv4l</td> 772 <td><b><i>all tests passed</i></b></td> 773 </tr> 774 <tr> 775 <td>Linux 2.4.18 #2<br> 776 sparc64</td> 777 <td><b><i>all tests passed</i></b></td> 778 </tr> 779 <tr> 780 <td>MacOS X Darwin Kernel Version 5.5<br> 781 powerpc</td> 782 <td> 783 <p>failed more than one test.</p> 784 <p>Generally Pexpect works on OS X, but the nature of the quirks 785 cause a many of the tests to fail. See <a href="#bugs">bugs</a> 786 (Incomplete Child Output). The problem is more than minor, but Pexpect 787 is still more than useful for most tasks. The problem is an edge case.</p> 788 </td> 789 </tr> 790 <tr> 791 <td>Linux 2.2.20<br> 792 alpha<br> 793 </td> 794 <td><b><i>all tests passed</i></b></td> 795 </tr> 796 <tr> 797 <td>Linux 2.4.18-5smp<br> 798 i686</td> 799 <td><b><i>all tests passed</i></b></td> 800 </tr> 801 <tr> 802 <td>OpenBSD 2.9 GENERIC#653<br> 803 i386</td> 804 <td><b><i>all tests passed</i></b></td> 805 </tr> 806 <tr> 807 <td>Solaris</td> 808 <td> 809 <p>failed <span class="code">test_destructor</span></p> 810 <p>Otherwise, this is working pretty well. The destructor problem 811 is minor. For some reason, the <i>second</i> time a pty file 812 descriptor is created and deleted it never gets returned for use. It 813 does not effect the first time or the third time or any time after 814 that. It's only the second time. This is weird... This could be a file 815 descriptor leak, or it could be some peculiarity of how Solaris 816 recycles them. I thought it was a UNIX requirement for the OS to give 817 you the lowest available filedescriptor number. In any case, this 818 should not be a problem unless you create hundreds of pexpect 819 instances... It may also be a pty module bug. </p> 820 </td> 821 </tr> 822 <tr> 823 <td>Windows XP Cygwin</td> 824 <td>failed <span class="code">test_destructor</span>. That it 825 works at all is amazing to me. Cygwin rules!</td> 826 </tr> 827 </tbody> 828 </table> 829 --> 830 <h1> </h1> 831 <h1><a name="todo">TO DO</a></h1> 832 <p>Add an option to add a delay after each expect() or before each 833 read()/readline() call to automatically avoid the <a href="#echo_bug">echo 834 bug</a>.</p> 835 <p> </p> 836 </div> 837 <hr noshade="noshade" size="1"> 838 <table border="0"> 839 <tbody> 840 <tr> 841 <td> <a href="http://www.noah.org/email/"><img src="email.png" 842 alt="Click to send email." border="0" height="16" width="100"></a> </td> 843 </tr> 844 </tbody> 845 </table> 846 </div> 847 <div id="Menu"><b>INDEX</b><br> 848 <hr noshade="noshade" size="1"> <a href="#license" 849 title="Python Software Foundation License">License</a><br> 850 <a href="#download" title="Download and setup instructions">Download</a><br> 851 <a href="#doc" title="Documentation and overview">Documentation</a><br> 852 <a href="#status" title="Project Status">Project Status</a><br> 853 <a href="#requirements" title="System requirements to use Pexpect">Requirements</a><br> 854 <a href="#overview" title="Overview of what Pexpect does">Overview</a><br> 855 <a href="#faq" title="FAQ">FAQ</a><br> 856 <a href="#bugs" title="Bugs and work-arounds">Known Bugs</a><br> 857 <a href="#changes" title="What's new with Pexpect">Recent Changes</a><br> 858 <a href="#testing" title="Test results on various platforms">Testing</a><br> 859 <a href="#todo" title="What to do next">To do</a><br> 860 <a href="http://pexpect.svn.sourceforge.net/viewvc/pexpect/trunk/pexpect/" title="browse SVN">Browse SVN</a><br> 861 <br> 862 <a href="http://sourceforge.net/projects/pexpect/" 863 title="The Pexpect project page on SourceForge.net"> <img 864 src="http://sourceforge.net/sflogo.php?group_id=59762&type=5" 865 alt="The Pexpect project page on SourceForge.net" border="0" 866 height="31" width="105"> </a> </div> 867 </body> 868 </html> 869
