D7net
Home
Console
Upload
information
Create File
Create Folder
About
Tools
:
/
usr
/
share
/
doc
/
expect-5.45
/
Filename :
NEWS
back
Copy
This file is the NEWS file from the Expect distribution. ====================================================================== ====================================================================== Date: 08/01/00 ** SUMMARY Expect 5.32 is being released in conjuction with Tcl 8.3.2. This is a fairly minor update with no feature changes but with a number of useful bug fixes in the way expects uses the new regular expression engine and the UTF-8 features of Tcl. Details are in the HISTORY and ChangeLog files. ====================================================================== ====================================================================== Date: 10/22/99 ** SUMMARY Expect 5.31 now works with Tcl 8.2. Expect 5.31 does NOT work with prior releases of Tcl. Thanks to an incredible amount of work by Scott Stanton, Henry Spencer, Melissa Hirschl, and funding from Scriptics for making this possible. ** NEW FEATURES What? You mean that working with Tcl 8.2 isn't enough????? Expect supports Tcl's new regexp engine. Expect supports null bytes in strings directly. (You no longer have to use the "null" keyword to send or match nulls. Of course, the "null" keyword will continue to be supported.) Null removal (on input) is still enabled by default since nulls are almost never intended for end-user consumption in interactive dialogues. ** CHANGES IN BEHAVIOR (POTENTIAL INCOMPATIBILITIES) The interpreter command used to exit upon eof. Now it uses "-eof script" to control this behavior. The default behavior is to return. (This change was required because Expect now gives control to Tcl upon exit and Tcl waits (potentially forever) for all processes to die on exit.) Explicit calls to interpreter are almost non-existent. However, you should look for *implicit* calls via interact commands with a pattern but no action. This required changes in the examples: dislocate, dvorak, kibitz, and xkibitz. Indirect variables can no longer start with "exp". Such variables will be interpreted as channel names. Old-style regexps may need conversion. If you have been protecting regexps containing backslashes with {}, then you need to examine all your backslashes since the new regexp engine interprets backslash sequences (such as \n) itself. For example: expect "\n" (works the same in Tcl 8.0 and 8.1) expect {\n} (works differently in Tcl 8.0 and 8.1) Scriptics has also created a new-regexp-features page which you should read: http://www.scriptics.com/support/howto/regexp81.html. Some of the new features allow much more efficient regexps than before. For example, non-greedy quantifiers can replace [split] looping constructions with a single regexp, enabling Tcl to parse very efficiently. For the whole story, read the re_syntax man page. The interact command's regexp matching no longer skips nulls. (I'd be surprised if anyone misses this. At least I hope ....) Expect's C library now reports failures in spawn's underlying exec directly (by returning -1) rather than the way it used to (as data in the pty). This makes user code more robust. However, it requires you to simplify your code, alas. See the chesslib.c example. Linking with Expect's C library no longer requires the Tcl library (unless, of course, you make Tcl calls yourself). Tcl is still required to build the library in the first place, however. ** CHANGES IN BEHAVIOR (SHOULD NOT CAUSE INCOMPATIBILITIES) The match_max command now controls by bytes, not chars. This won't cause problems unless your existing scripts are interacting using sizeable chunks of multibyte characters. (If you don't know what I'm talking about, ignore this.) The Make/configure suite now corresponds to the TEA conventions (at least in theory; the conventions are changing regularly so it's hard to be less vague on this point). Significantly, this means that you should be able to use the same configure flags as when compiling Tcl or any other TEA-compatible extension. (See the INSTALL file.) The values of special variables such as exp_spawn_id_any have changed. (The values were never documented so you shouldn't have been using them anyway.) Spawn ids now appear as "exp...". (They used to be small integers.) Do not assume that spawn ids will continue to be represented in any particular way (other than unique strings). ** OTHER NOTES Expect uses channels. There is an Expect channel type. It is possible to use Tcl's channel commands, such as fconfigure, to change the encoding. However, Expect layers its own buffering system on top of Tcl's channel handler so don't expect intuitive behavior when using commands such as gets and puts. Unless you know what you're doing, I recommend manipulating the Expect channels only with the expect commands. Some effort was made to make Expect support threads, however it is not complete. You can compile Expect with threads enabled but don't run Expect in multiple threads just yet. So much code has changed, there are bound to be bugs in dark corners. Please let me know of such cases. The best response will come by supplying a simple test case that can be added to Expect's test suite. In places where the behavior of Expect was not precisely documented, full advantage was taken to do something different :-) Several esoteric bugs were fixed. Although Expect itself uses Henry Spencer's new regexp engine, Expect's C library still uses his original regexp engine. No testing has been done of the poll and non-event subsystems. (These are used on systems which don't support select on ptys or ttys. Some minor work needs to be done on them (because the event subsystem was rewritten) which I'll probably do only if anyone requests it. Many deprecated features (deprecated for many years!) have been removed. All such features were deprecated prior to Exploring Expect so if that's how you learned Expect, you have nothing to worry about. For example, Expect's getpid command predates Tcl's pid command and it's been deprecated for, oh.... 6 years - wow! Other deprecated features include: expect -timestamp (flag only; behavior itself was removed years ago) expect -iwrite (flag only; behavior occurs all the time) expect_version (use "exp_version" command) expect_library (use "exp_library" global variable) interact -eof (use "eof" keyword) interact -timeout (use "timeout" keyword) interact -timestamp (use "clock" command) getpid (use "pid" command) system stty (use "stty" command) With this release, the following are deprecated: timestamp (use "clock" command) debugger (use a different one; there are very nice replacements around. Fortunately the Expect debugger is not something anyone is wiring into their scripts, so for now, consider it on the endangered species list. Anyone still want this debugger?) From now on, the most current snapshots of Expect will be found in the Scriptics CVS repository. Not all snapshots are official releases. ====================================================================== ====================================================================== Date: 8/18/96 Expect now works with Tcl 8.0. No changes were made to take advantage of 8.0 features such as namespaces. (If you want to put the Expect commands in a namespace, declare a namespace before loading them in.) Even though Tcl allows embedded nulls in commands, Expect still does not. Tcl still doesn't support embedded in patterns and regexps. I'll wait til Tcl supports that before rewriting Expect's null support. ====================================================================== ====================================================================== Date: 9/28/96 There is now an Expect FAQ and home page. Feedback is encouraged. You can get to the FAQ from the home page and vice versa, so you only need to remember one of the links, but here are both for completeness: home page: http://expect.nist.gov FAQ: http://expect.nist.gov/FAQ.html ====================================================================== ====================================================================== This section describes the changes to Expect 5, many due to changes from Tcl 7.4 to 7.5. Improvements ============================== You can now use expect_background from Tcl along with all the Tcl-based event stuff such as "after". You can also do fun things such as have "after" actions run while expect is waiting for input. It's worth comparing the difference between commands such as expect (which waits in the event loop) and gets (which doesn't). Incompatibilities ============================== libexpectk is gone. Because of the event loop was moved into Tcl, this is no longer necessary. Just use the regular Expect library. This only affects you if are hand-linking. The name of the static C library now has the extension on the end - just like Tcl. This only affects you if are hand-linking. ============================== Some obvious questions and answers Nothing in the user interface has changed. All your old scripts should run. Tcl now has commands that replace Expect functionality in the area of version control and date handling. I encourage you to use Tcl's support rather than Expect's for any new scripts that you write. However, I won't be removing Expect's commands, so don't worry about having to convert old scripts. It is my understand that shared/dl libs work. (I say "it is my understanding", because my own environment doesn't handle it, sigh. My system admins tell me that they're working on it.) So I've had to guess on some things - in fact, I've talked to a number of people and I get the feeling that a lot of people are guessing on shared/dl libs. I have yet to talk to anyone that REALLY understands this stuff (by "understand", I mean "can write the configure/Makefile portably and correctly". So much for my griping. In theory, the shared/dl support is pretty much a freebie because Tcl itself provides all the support for this. (There is some reorganization that could be done to improve shared library memory use - I'll get to it eventually - it shouldn't affect most people.) Don't send me complaints about shared/dl libs unless you are *positive* it is something that I am responsible for. Even if Tcl works and Expect fails, it is likely to be a Tcl error (or more precisely, a configuration problem that is more appropriately fixed from the Tcl distribution). For Tcl-package purposes, Expect is called "Expect". (Duh...) Expect's multiple interpreter support is minimal. It should work for most things, serendipitously. I haven't spent any time making this perfect. No, this release isn't portable to Windows or Mac. Let me know if you're seriously interested in a lot of work. I'm not saying it's not possible. It's definitely possible and the porting working at Sun has made it easier than before. But it's still not a weekend hack. ============================== Building the code ============================== Expect builds as usual. (See the INSTALL file for details.) The only significant change is that Expect now has to find the tclConfig.sh file (and tkConfig.sh if you want). So if you like to store Tcl where Expect can't find it, you'll need to use even more configure flags than you used to. Shared/dl Expect libraries are built if you configured Tcl for shared/dl libraries. All support for earlier versions of Tcl and Tk have been removed from Expect. For example, if you're still using Tcl 7.4 (or earlier), don't bother to install this release. ====================================================================== ====================================================================== This section describes the changes from Expect 4 to Expect 5. The changes that people will find most interesting or annoying are as follows. Some of them may seem rather arbitrary but fix inconsistencies leading to much cleaner coding both internally and externally. -- Expect version 5.20 and above is designed to work with Tcl 7.5 and Tk 4.1. Expect 5.20 and above will not work with earlier versions. -- Glob-style patterns do longest-possible matches (from the earliest possible position) just like regexps. Previously, they matched the shortest-possible strings. However, the documentation didn't actually say this (it didn't say anything) -- Exact patterns are now supported from expect. Use the "-ex" flag. Exact patterns work just like those in interact. No special handling is made of *, ^, etc. -- The new command "expect_background" registers patterns that are to be tested against spawned process output whenever it appears (i.e., asynchronously). This only works in the Tk environment. The arguments are the same as the expect command. -- expect_before and expect_after now handle their arguments like expect_background. Previously, a command such as "expect_before" with no arguments deleted patterns for all spawn ids. Now, it only deletes patterns for the current spawn id. Similarly with the "-i" form. -- expect_background/before/after support an -info flag to query what the current patterns are. The results are returned in such a way that they can be re-used by a new expect command. The -info flag must be the first flag in the command. With no other arguments, it returns the setting for the current spawn id. With a -i descriptor, information is returned for that spawn id. The argument -noindirect may be used to suppress indirects which also match a direct spawn id. Only a single -i specification may be given with -info. With the argument "-all", all spawn id specifications are reported. -- There is now a sleep command. It understands decimal values such as sleep .5 Interrupts and other asynchronous events are processed while Expect sleeps. -- Traps now use Tcl's "Async" support. This has advantages and disadvantages. One advantage is that traps have no chance of screwing up the Tcl internals. One disadvantage is that trap handlers are delayed at certain specific times and places. For example, a handler cannot occur inside another handler. While one handler is running, all other handlers are blocked. This is probably the most noticable place where handlers are blocked. Others are generally small windows, so you shouldn't notice the delay in executing the handlers. Several traps are initially defined: trap exit {SIGINT SIGTERM} If you use the -D flag to start the debugger, the following trap is defined: trap {exp_debug 1} SIGINT You can, of course, override these. In particular, if you have your own "trap exit SIGINT", this will override the debugger trap. Because SIGINT is tied to exit (see above) by default anyway, you should remove your own "trap exit SIGINT" unless you specifically do not want to be able to get to the debugger by ^C. If you want to define your own trap on SIGINT but still trap to the debugger when it is running, use: if ![exp_debug] {trap mystuff SIGINT} Alternatively, you can trap to the debugger using some other signal. The ONEXIT trap is no longer available. Instead, say "exit -onexit ..." Traps are now deleted by using the empty ({}) handler. The current handler is returned if no action is supplied. With no arguments, trap returns the signal number of the trap command currently being executed. -- The wait command now returns a four element list if a valid child was waited on. Element 1: pid Element 2: spawn id Element 3: 0 (or -1 if there was an OS error) Element 4: status (or errno if element 3 == -1) -- expect and interact notes: The timeout and eof patterns were initially named "-timeout" and "-eof" but have been renamed "timeout" and "eof" to match those of expect. The ability to define default timeout/eof actions has been removed. (You can do this more easily by grouping spawn ids.) expect and interact now support a "null" keyword to match an ASCII 0. send supports -null and -break keywords. Since a large number of special keywords have been added to interact, a new keyword "-ex" for "exact" was added descriptive of its default treatment of patterns. This protects the next token from being misinterpreted as a keyword. The expect command provides "-gl" for "glob" for analogous reasons. Any string starting with "-" should be protected by the "-ex" or "-gl" flag, even those that are not keywords currently. (All strings starting with "-" are reserved for future options.) String start/end indices are no longer written to expect_out and interact_out unless the -indices flag is given. expect_out(spawn_id) is set to the spawn id associated with the spawn id that produced the last output in an expect command. For example, you can use this to delete files that have closed, by removing this element from an indirect spawn ids spec. The same effect is reproducable with interact (and interact_out(spawn_id)) but for efficiency reasons, it requires the -iwrite flag before each pattern. Expect's -i and interact's -i, -u, -input, and -output flags can now describe a list of spawn ids. So you can say things like: interact -input "$id1 $id2 $id3" .... -output "$id1 $id2" ... In this case, id1, 2, 3 would be sent to id1, and 2. The spawn id may be given as a global variable name (called an "indirect spawn id specification"), in which case, the variable contains the list of spawn ids. Whenever the variable is changed, the new list of spawn ids is automatically used. This is particularly useful with any long running expect command such as expect_before, expect_after, expect_background, and interact. The -update flag was removed. Use indirect spawn ids (see previous paragraph). -- interact notes: Interact now support -input and -output flags that provide very flexible means of moving data from/to multiple spawn ids in complex ways (but very quickly). It is possible to write most expect loops using a simple interact statement. For instance, the three way interaction inside kibitz (between two users and a process) is written this way: interact { -output $shell -input $userin eof { . . . } -output $shell -input $shell -output "$user_spawn_id $userout" } -- send command notes: It is possible to send a break by using the "-break" flag. Any string starting with "-" should be protected by preceding it with the "--" flag, even those that are not keywords currently. (All strings starting "-" are reserved for future options.) -- The spawn command now takes an "-open" flag which in turns takes a Tcl file as an argument. This lets you treat raw devices, files, and pipelines as spawned processes without using a pty. This was actually in Expect 4, but I forgot to document it. Oops! -- The old "debug" command (which describes what Expect is doing internally) was renamed "exp_internal". "debug" (and "exp_debug") now invoke the interactive debugger. -- The new command "stty" now takes over the job of "system stty". It works much better, allowing POSIX-style redirection to affect other ttys. It otherwise takes arguments as "system stty" did. -- The "-tcl" option to "return" has gone away. (This was dangerous to anyone that actually happened to return the value "-tcl".) Instead, use inter_return. -- Added exp_timestamp command to produce very fast timestamps. -- Added exp_pid command to return pid of given spawn id. -- The close command now takes an argument of "-onexec" with a following 0 or non-zero value. For example, the follow command stops the current spawn id from being closed when another process is exec'd or spawn'd. close -onexec 0 While "-onexec 1" returns it to the default condition where it will be closed upon exec or spawn. -- log_user now returns previous value. It is acceptable to call now, without arguments just to get the value. -- The following forms are deprecated. They will be allowed indefinitely but not advertised or supported if they break. -eof, -timeout in interact (reason: didn't match expect. Instead, use eof or timeout.) -- in expect or interact (reason: no easier to read. Instead, use -gl in expect or -ex in interact.) continue -expect (reason: when mixing in extensions, you have to use exp_continue, so -expect becomes irrelevant. Instead, use exp_continue.) getpid (reason: Tcl now supplies same functionality as "pid". Instead, use pid.) expect_version and expect_library (reason: the name implies they have something to do with the expect command, which they doesn't. Instead, use exp_version and exp_library.) -timestamp for obtaining tm and ctime in expect and interact (reason: separate command now exists for this purpose. Instead, use exp_timestamp.) system stty (reason: bad interaction with redirection. Instead, use stty.) -- New examples have been added: "dislocate" lets you disconnect and reconnect to processes. "tkpasswd" illustrates passwd embedded in a GUI. They may not be overwhelmingly useful, but run them once to see what they do. If you ever need to do anything similar, you can look back at them. "tknewsbiff" pops up a window or plays a audio clip when you have unread news. -- Changes to the Expect libraries: The expect-tcl library (libexpectcl.a) has been integrated with the expect library (libexpect.a). So references to libexpectcl.a should be removed. The Expect C library now supports buffering, multiplexing, null matching, full buffer matching. Basically, all of the features in Expect are now in the library. Buffering and multiplexing has caused the biggest change to the library. Previously, exp_match contained the entire buffer that matched. Now exp_match just points to where in the buffer the match started. exp_buffer points to the beginning of the buffer. Previously, the previous buffer was thrown away at the beginning of each expect function call. Now, previously unmatched characters are eligible for matching. To match on different file descriptors, exp_match, exp_match_end, exp_buffer_end must be restored to their previous values. Initially, they should be zero. The meaning of timeout == 0 in the Expect library has been changed. See the man page for more info. ====================================================================== ====================================================================== This file describes the changes from Expect 3 to Expect 4. The improvements that people will find most interesting are: 1) Expect version 4 is designed to work with Tcl 6.7 and Tk 3.2. (Earlier versions of Expect will not work with Tcl 6.7) Expect can now be layered in with any Tcl program. Note that in Tk, Expect's send command is called "exp_send". (It used to be called "send_spawn" but this bit the dust.) 2) A debugger is provided. 3) The interact command has been totally rewritten and supports regular expressions, timeout/eof patterns, and a number of other new things. 4) The default behavior of ^C (SIGINT) is exit whether or not you are in a read. 5) Expect uses "sane" terminal parameters by default, allowing scripts to work the same whether inside emacs shell mode or not. (See man page on "spawn" for more info.) 6) All the hard parts of the installation process are automated. This was done primarily by Rob Savoye at Cygnus. Thank you, Rob! 7) It is now possible to buy a support contract for Expect from Cygnus. The changes that people will find most annoying are: 1) send now only sends a single string. (It used to send any number of strings with spaces jammed in between.) 2) getpid was renamed pid. 3) interact's -flush was renamed -nobuffer (much more descriptive). 4) interact now runs all actions in raw mode unless the flag -reset is used. -f and -F are ignored. send automatically understands how to do the right thing. The most likely thing to watch out for are actions like "exec kill -STOP 0" which almost certainly need the -reset flag. 5) argv0 is initialized to script name. argv no longer contains it. argc is initialized [llength $argv]. This follows new Tcl style. All differences are described in the man page. Some of the less significant differences are described in the HISTORY file. The debugger is described in a separate document (see the README). This version also introduces one incompatibility that may require changes to scripts. While this may initially annoy you, the final result will simplify the process of writing scripts. Namely: In version 3, the expect command accepted lists of glob-style patterns such as: expect "a\ b c" action where "a b" or "c" would cause action to be executed. The problem with this is that the pattern list is hard to write and hard to read. Patterns with control-characters, backslashes and dollar signs were very difficult to deal with. Regular-expression patterns provide a much simpler solution. Via the alternation feature (using a "|") the above pattern can be written as: expect -re "a b|c" action I was concerned about people having a significant investment in code that depended on the old syntax but responders to a comp.lang.tcl poll about such a change in pattern handling were 100% in favor of it. (I even proposed hooks for backward compatibility, but there was no interest in it.) Fortunately, most simple things will work as before including: expect foobar expect {foobar} expect "foobar" expect "foo\ bar" However, some things won't work as before. For example, the following will behave differently - now the braces will be considered as part of the pattern. expect "{foo bar}" Here are examples of patterns in my own code that I had to change: was changed to Version 3 pattern list Version 4 pattern {Whois:\ } "Whois: " {No\ match} "No match" {250*ftp>* 200*ftp>*} -re "2(5|0)0.*ftp>.*" {{Press Return to continue*}} "Press Return to continue*" {*\r\n*\\\\\r\n} "\r\n*\\\r\n" Future Change Alert John Ousterhout has pre-announced a future change in Tcl that may affect you. In particular, backslash sequences other than those explicitly listed in the Tcl documentation will be handled as if the backslash was not present. The likely place this arises is when quoting characters that are special to the pattern matcher but not to Tcl. For example in Tcl 6.7, the following command matches a period. expect -re "\." In Tcl 7.0, it will match any character, because Tcl will throw away the backslash. If you want to match a period, you will have to say: expect -re "\\." or expect -re {\.} The following command will find occurrences of this. (It may find other things, but it will at least find the problem cases.) egrep '(\\$)|(\\[^][bfnrtv\0-9{}$ ;"])' *.exp ====================================================================== ====================================================================== This section describes the changes from Expect 2 to Expect 3. If you used to use Expect version 2 (any version written before September '91) you will find that the current version of Expect (3) introduced minor but significant incompatibilities. The HISTORY file describes these briefly. They are described at length in the man page. I'm sorry if you feel annoyed at the incompatibilities, but Expect has been out for a year and a half, Tcl even longer. Both Tcl and Expect are using this as a last chance to make significant changes, so that we will not disturb even more users in the future. There is no automated conversion procedure (although see note below) for Expect or even raw Tcl. For now, I suggest that you not bother fixing things that already work - just keep the old Expect around. The binary isn't very big after all. If you do write a translation script, let me know. Thanks. Of course, I felt obligated to convert the examples distributed with expect. I did this by hand while writing the new version itself, partly as an aid but mostly to test lots of variations. In 90% of the scripts, all I had to do was change: (changes due to Tcl) 'index' to 'lindex' 'range' to 'lrange' 'length' to 'llength' 'print' to 'send_user' or 'puts' depending on how you use it 'function .... c' with '[join [function [split string ""]] ""]' (changes due to Expect) 'expect_match' to 'expect_out(buffer)' 'set match_max' to 'match_max' (perhaps with -d flag) '*' to '-re .+' If anyone wants to write a script to do this, note the pitfalls: 1) functions and variables do not share the same namespace, so it is a inappropriate to just globally rename things. A number of optimizations can be made: 1) If you are doing multiple split/joins, you should probably cache the split string. 2) Virtually all uses of scan are unnecessary now, due to exec's automatic stripping of terminating newlines, and expect's support of regexps. 3) String manipulation can be reduced or avoided entirely if you use expect -re. 4) exec is no longer necessary to retrieve environment variables, since they can now be retrieved from $env. 5) If you have been really careful about testing for timeout and eof, you can dramatically reduce the size of your scripts by using expect_before and expect_after. This is more efficient, as well, since those actions are only parsed once.