This is the HISTORY file for Expect. Modifications made by Cygnus support are in ChangeLog. - Don Date Version Description ------- ------- ------------------------------------------------------ 4/19/93 4.5.2 Achim Flammenkamp noted that I documented full_buffer as buffer_full. Ted Stockwell noted that wait arg was missing an & in configure test. Scott Hess noted that systems can have wait4 without waitpid. Jonathan Kamens noted/fixed some things that weren't autoconf'd correctly: pid_t, RETSIGTYPE, malloc. Gary Shea noted that a recent change to expectk made it not default to interactive. 4/12/93 4.5.1 At request of Rusty Wilson , added "-console" to spawn. Pang Wai Man Raymond reported that passmass didn't grok DEC's passwd prompts for root. 4/7/93 4.5.0 Fixed bug in interact regexp preventing match of multichar literals. 4/6/93 4.4.3 Bennett Todd noted missing example scripts timed-read and time-run. 3/29/93 4.4.2 Bill Houle on HP, make kibitz use domainname as fallback and used whoami instead of env(USER). Fixed bug in the generic pty code that could report out of ptys because an earlier slave slowly deleted the lock file. 3/25/93 4.4.1 Stephen House reported exp_tk.c wouldn't compile on HP. Fixed. 3/24/93 4.4.0 Added back SVR4-style pty allocation which got omitted in the autoconfig process. Fixed bug in interact's -update handling. Fixed bug in weather script that cut off long reports. 3/15/93 4.3.0 Cleaned up /tmp files used during pty locking. Added command "parity" to enable parity stripping. Fixed match_max to do -i correctly. 3/15/93 4.2.4 Fixed to work on new SGI which returns slave-close via excep (select) or POLLERR (poll) rather than thru read(). Why do you people do things like this? 3/12/93 4.2.3 Fixed to work on AIX (using /dev/ptc) and UTS (using getpty). 3/11/93 4.2.1-2 Fixed numerous bugs relating to HP ptys. It's amazing that for their bewildering complexity, they couldn't support generation of EOF to the master (or at least enable trapping of just close), rather than forcing the code to know about opens, too. 3/8/93 4.2.0 Integrated Rob Savoye's autoconfig code. Interact mishandled new -eof flag. Added -update. Gary Shea noted that tkwait hung if expect had been called. Rewrote most of tk_event.c and fixed some other problems related to efficiency & multiple timeouts. E Beck suggested mods to more easily support Extended Tcl. Bill Mitchell reported problems on 4.3+BSD. Added support for TIOCSCTTY. Dana Burd noted that "exit" caused by ^C during expect didn't work - just returning to expect. Fixed, and then removed "feature" of ^C to abort a timeout. This feature proved a lot less useful than I thought it would. 2/21/93 4.1.0 Bill Tierney noted that double close dumped core. Rewrote fd_to_f and close/adjust functions. Interactive interpreter() didn't properly wait in get_next_event, so Tk stopped responding to events. Wrote version of interpreter that shares expect's input buffers but can't think of a use. Left as an ifdef SHARE_CMD_BUFFER. 1/26/93 4.0.1 Added eof check to xpstat. Removed incorrect and unnec. #includes from exp_main_exp.c Chip Rosenthal found my refs to tclRegexpError need externs on systems that don't use Tcl's string.h. string.h should probably be changed not to refer to tclInt.h. Added FAQ about Expect's copyright status. Mark Christopher pointed out some really stupid errors in the HP support for select. 12/16/92 4.0.0 Rewrote interact. Made re-entrant thru event-handler for Tk. (Same for Expect.) Abstracted out common code so that remainder is specific to select vs poll vs tk (although "simple" was impossible to handle). Added timeouts, regexps (at request of numerous people), ability to set up arbitrary graphs of process flows, and some miscellaneous but useful functionality. New flags are: -input, -output, -re, -echo, -flush, -eof. Added "-noecho" to spawn command. Added getpid command. Something with this functionality should be added to the Tcl core. When it is, this function will go away. Removed assumption of global "interp" handle. Rewrote init and other routines for use as libraries. Added appropriate glue to Makefile. At request of Rob Savoye added "send_log" and disabled buffering on all output. The only affect unbuffered output will cause users is if they pass large strings in multiple args to send. Ray Davis reported Convex could not do job control from spawned procs. I added a symbol DO_SETSID to force this. Martin Leisner modified rftp to understand iftp. I added it to the publicly donated scripts directory. 11/17/92 3.24.1 Martin Leisner suggested Makefile use $(MAKE) and support Tcl as a Sun shared library. Seth Perlman suggested interact support timeout. I've added this as "-timeout" in inter_select but left undocumented while we experiment with interface. Joe VanAndel pointed out that su2 script still used old syntax. Fixed. Konrad Haedener fixed a bug in POSIX tty handling on AIX. Surprisingly, we discovered AIX worked just fine when pty_bsd is used and without -DPOSIX! Doug (George Jetson) pointed out that a spawn_id for /dev/tty would be really handy. I added tty_spawn_id for this purpose. 11/4/92 3.24.0 After problem reported by James Ward added to man page describing delays required by hardware such as for UART switching. Recoded all \C sequences as \### in examples and man page in anticipation of them going away in next version of Tcl. Switched to printing errorInfo during errors instead of the command and only the top-level error message. Since this includes entire stack, this should be very helpful. J. Cazander reported that purify found a write beyond the end of an input buffer. Lucked, it was just before a double-word boundary, so it probably isn't a problem. I fixed it anyway. 10/9/92 3.23.1 Tor Lillqvist supplied support for HP 8.0.7 in POSIX-mode, and a bugfix for POSIX tty mode switching 10/8/92 3.23.0 Larry Rogers reported that "weather" blew up in spawn. I'll add a catch-all to the script to report similar problems (out of ptys, processes, etc.) Ting Leung notes that log() in human_write can receive a 0 (domain error). Fixed unit_random to avoid that. Tony Primavera notes that the sample archie script needs to understand mcgill's limit of 10 users. Tor Lillqvist noted that a lesser-used pattern ("unknown...") is incorrect. Grant Taylor found a problem when forking (using Expect's fork) multiple processes, each of which spawned something. In the BSD pty support, I had used the pid to build a temporary file for testing the pty before actual use. When multiple processes tried to use the same tempfile, it blew up. 8/12/92 3.22.13 Corey Satten pointed out that -u on cat caused kibitz to slow down on Ultrix. I see the same behavior on SunOS. I added an option to fix it for systems that need it. Corey also noted arg miscounting in kibitz, and pointed out that world-readable fifos could be a security problem. He gave a fix for this and also a fix to force ptys to be put into raw mode. Terrence Brannon reported rftp referenced the undefined variable 'transfer'. Turned out to be a bug in the code to handle symbolic links. 7/20/92 3.22.12 Added O_NOCTTY (if defined) in pty_bsd.c to avoid gaining control terminal while testing pty when running as daemon. At request of Michael D. Riley added explanation to man page - how expect_after/before deal with spawn_id. Charles Hannum discovered the problem with AIX (see earlier) was a missing "extern" in the errno declaration. Also, the compiler was sensitive to a lack of access to the defn of struct expect_special. Dave Coombs gave me yet another fix for the weather server to accomodate its ever continual change. 7/2/92 3.22.11 Yet more work. Discovered that SunOS and Ultrix really like setpgrp(0,0) much better than setpgrp(0,getpid()) but the manual doesn't describe well why this seems to work better. (The old call worked inconsistently.) 6/30/92 3.22.10 Did more work on modifications to dissolve connection between stdio and devtty. Eventually, I'd like to add a separate spawn_id for devtty (expect_devtty?). 6/5/92 3.22.9 Hansel Wan noted that $errorInfo was clobbered by prompt1. To prevent this, I added a default definition for prompt1 (and prompt2 while I was at it). Unnati Amin noted that the example scripts checked for $ in prompts which didn't work. This bug was created when $ was turned into a "match end-of-input" char in the transition from v2 to v3. Solution: backslash the $. A few parts of code assumed spawn_id was always stdin, which caused "send" to send to stdout, which meant succeeding expect's hung, waiting forever. Fixed is_user macro. This was a problem with scripts that redirected stdin or somehow reused fd 0. Surprising that no one ever did that before - also surprising that it didn't bother cron jobs. 6/2/92 3.22.8 Man pages fixes from Matt Crawford crawdad@fncent.fnal.gov. 5/12/92 3.22.7 Missing ; in Makefile, screwed up chmod. Fixed bug that caused interact to think the modes had changed when they hadn't. 5/11/92 3.22.6 Added regression paper to ftp archive - published in the 1992 USENIX San Antonio Proceedings. Swapped setpgrp and fork in disconnect command for sysV88. According to Dave Schmitt , original code (right out of Stevens) starts the child with closed stdio fds. Fixed bug in interact that changed /dev/tty modes even if interact was used to connect two completely different ttys. Had never been a problem before, but today I wrote some code that actually calls interact from cron! Also, copied the experimental fix from 3.22.5 to inter_poll. Jeremy Nussbaum says cat needs "-u" in kibitz for his HP 8.0 system to work. I wonder why this has never been a problem on earlier HP and other systems? Forced Makefile to mark scripts executable. 4/12/92 3.22.5 Fixed bugs reported by Matt Ranney including a syntax error (!) in expect.c on ecases_inuse. I didn't even compile this before pushing out? He also noted some # were not in column 1. I put in an experimental fix to interact (only in select version currently) to fix when pattern matching from master and user needs to continue typing in order to complete match. 4/3/92 3.22.4 Charles Hannum (mycroft@gnu.ai.mit.edu) pointed out that I screwed up a comment in the brand new pty_aix3.c. He also gave me a fix for an arg-less expect, which did a malloc(0). And he said that AIX ptys return EOF in yet a new way - read() returns -1 with errno == 0. Yuck. 3/29/92 3.22.3 Jay Schmidgall gave me yet another pty_aix3.c. He also gave ifdefs for POSIX terminal support. 3/18/92 3.22.2 Jay Schmidgall modified pty_sgi3.c to make a pty interface for recent versions of AIX. Steve Summit noted that "trap 0" could actually call signal(0...) Martin Leisner noted that rftp was broken. It seems I never handled symlinks. They are interesting. You can't tell from the listing whether they are files or directories, so you just have to blindly go ahead and assume it's one or the other and see what happens! 3/11/92 3.22.1 In talking to Dave Schmitt ,daves@techmpc.csg.gss.mot.com>, realized the documentation for wait had never been updated from the way it used to work in v2 (returning any pid). 3/11/92 3.22.0 Another question from Ron, prompted me to find another bug. interact -o wrongly manipulated the user buffer at one point. 3/10/92 3.21.0 Ron Young found that spawn failed on a DECstation 3100 running Ultrix 4.2. I had forgotten to test that cmdfile was valid before comparing against stdin in fix related to fflush in 3.20.0. While I was on a DECstation, I noticed that it does not accept setpgrp(...,0). Changed 2nd arg to getpid(). 3/6/92 3.20.2 Stefan Farestam provided a new version of pty_sgi.c which uses _getpty. I renamed the old one pty_sgi3.c 3/3/92 3.20.1 Brian Woodson requested I update the dates and version numbers. 3/1/92 3.20.0 Prompted by a question from Ken Mandelberg, added -raw to noidle and kibitz script. Fixed fflush(cmdfile) again, having been authoritatively told by net wisdom that there is no way to portably fflush a shared read-stream. (I take back my claim about a bug in HP's fclose!) John Sellens gave me some more fixes for non-DEC MIPS OS. 2/22/92 3.19.1 John Sellens gave me a bug fix for NOWAITPID. 2/21/92 3.19.0 Found a bug in HPUX fclose!! It moves the I/O pointer in the shared file table entry! This explains the symptoms I reported earlier. Fortunately, it's easy to code around (by me - it is no longer necessary to fudge the scripts). Added some stuff to the man page to explain why expect behaves the way it does in an emacs shell window and how to live with it. 2/21/92 3.18.0 Worked on the HP port some more. The HP causes a real problem by insisting SIGCLD be delivered in order for wait to return a status. This royally complicated the code, partly because of the special casing all over the place in the trap command, the asynchronous delivery of SIGCLD and also because Tcl itself is not prepared to have system calls be interrupted. Cleverly, the HP also defines both CLD and CHLD which threw my macros off at first. Thanks, but I don't this kind of help! Anyway, the end result is that on the HP, SIGCLD is ignored. The manual claims wait status will not be delivered but it seems to be anyway. Good grief! (Even if it were ignored, it would not be such a calamity, since wait is used mainly to discard zombies on other systems.) A remaining problem is that there appears to be some odd interaction, perhaps with fork, such that the script is rolled back at eof if a spawned process happens to exit at the same time. The solution for now is to exit all scripts via exit rather than letting exit be called implicitly. There must be some real bug, but I'm unable to find anything after lots of testing, line and Saber. At the moment, I'm highly suspicious of the HP itself rather than expect. Bob Proulx and Jeff Okamoto supplied me with patches for inter_select.c. HP transmits some pty interactions via the exception field in select. Michael Grant gave me a mod to grok ~ in the logfile and debug commands. 2/17/92 3.17.1 Brian Keves pointed out that the man page still referred to "expect_match" instead of "expect_out". 2/12/92 3.17.0 Eric Arnold ran into a problem when running in the background. interact did ioctl(0...)s to change the terminal mode, ignoring the -u flag. Fixed a bug in kibitz which blew up when asking for a password due to a spelling error. The drawbacks of interpreters... 2/4/92 3.16.3 Dongchul Lim noted that scripts can hang in the background. I had assumed isatty(0) was enough to contrast bg/fg but it returns 1 if the script was started with a & from the terminal. I added code to watch if any ioctl(0)s were done. If so, than it is safe to do more, in particular in the exit handler to reset the terminal modes. 1/28/92 3.16.2 Fixed a bug on SV systems causing errors when trying to do further reads after a SIGCLD had already arrived on a spawn_id. Peter Funk gave mods for SCO XENIX 386. 1/24/92 3.16.1 Oops. Forgot to add pty_svr4.c to shar. 1/13/92 3.16.0 Karl Lehenbauer a tiny change for getting a clean compile on SCO 3.2.2. My getimeofday-avoidance code wasn't right, sigh. Kibitz noticed. Fixed two other bugs in kibitz - password request was for wrong user and it timed out but shouldn't have. Note: seems to work fine with new version of Tcl: 6.2 1/13/92 3.15.1 Added a bit of code to avoid gettimeofday system calls when timeout == -1. Fixed minor bugs in kibitz relating to cleaning up and returning error messages. Redid support for stdlib.h including making it default to fix problem in Ultrix 4.2 reported by Oliver Kretzschmar . Ian Johnstone said his system (DYNIX 3.2) needed an additional include in inter_select. Dave Coombs added logic to test/weather to accomodate a new feature in the weather server. Hal Peterson fixed some SV code that I just added for handling SIGCLD properly. He made the Cray-extra-child timeout in half the normal timeout to allow distinguishing between eof and real timeout. Finished rest of Jeff Okamoto's fixes for HPUX. Wally Strzelec provided mods for Amdahl which has its own pty-handling functions. Ifdef'd into pty_usg. 12/30/91 3.15.0 Fixed a bug that struck when eof occurred when reading from multiple processes simultaneously and no user-supplied eof handler. 12/26/91 3.14.1 Ted Gibson gave me some mods for a 3B2 having to do with termio vs termios, etc. 12/24/91 3.14.0 Deprecated expect 2. Expect 3 is now the official version. Parag Patel gave me some #includes necessary for A/UX 2. Brian Woodson noticed "send a b" generates incorrect debug output. Working with Jeff Okamoto to run expect on HP/UX 8.0, we fixed SIGCLD catching (he says HPUX doesn't ignore them by default?), obviating longjmp from stomping locals, fixed a bug in cmdWait that would prevent the wait status from being collected in rare situations. 12/17/91 3.13.1 James Davis suggested fixing Makefile to handle case where no example scripts should be installed. I added similar logic for script man pages. Pete Siemsen fixed a bunch of things in the Makefile including where to get expect when invoking fixline1. He suggested defs for supporting install and multiple MAN targets. 12/12/91 3.13.0 Matthew Freedman noted mismatch between lib man page (said "stty_init") and lib code (said "exp_stty"). He also found a screwup in the library such that the pty slave wasn't being set up correctly. Added note to kibitz man page on how to kibitz with 3 or more. 12/12/91 3.12.0 "expect *" worked incorrectly if it was first expect after spawn, due to buffer not being initialized. Added a good example for "expect -continue" to man page. Added an FAQ about a gcc problem that seems to be common. 12/11/91 3.11.2 James Davis noted I forgot to put kibitz.man in distribution. I changed kibitz to read domain from resolv.conf instead of calling domainname(1) for systems upon which NIS domainname differs from Internet. Pete Siemsen noted slight error in libexpect man page. 12/10/91 3.11.1 A couple tiny mods to the Makefile courtesy of James B. Davis and Michael Grant (guest worker from Sun, temporarily at . Both of them also noted a problem caused by incorrect installation of gcc that caused expect to say "ioctl(set): Invalid something or other" upon exit. Fixed complaint about exit() while compiling without STDC. 12/9/91 3.11.0 beta! Hal Peterson provided fixes for UNICOS 6.1 and 7.0 on both CRAY-2 and CRAY Y-MP. He also fixed a problem in interact where malloc(0) could've occurred. Added support for allowing user to set interpreter prompt. Added forgotten -d flag to match_max in rftp script. Made kibitz understand user@host. Expect's internal buffer-full-handling incorrectly copied the latter buffer half beginning from the end of the buffer. 12/5/91 3.10.1 Massaged Makefile to allow for more flexibility in installation, especially with regards to scripts. #! is now reset. Added "kibitz", a really cute script to let two people control one program. Example users are for one person to help another remotely, logging a conversation (run emacs or whatever inside kibitz and your conversation can be logged, scrolled backwards, etc., or of course, playing games together. 12/4/91 3.10.0 Tightened up arg checking for "wait" - it core dumped when it should've said "syntax error". Rick Cady noted minor inconsistency in man page describing strace. I fixed a bug that prevented "system stty -echo raw" from working. The raw data was clobbering the -echo data. 12/3/91 3.9.0 Brian Woodson noted that "close -i ..." evoked a syntax error. I had parsed the arguments incorrectly. After the nth request, I finally set up pub/expect/scripts as a directory for scripts. 12/2/91 3.8.0 Phil Sheperd fixed a major bug in exp_spawnv() preventing one side of the pty from being set up correctly. Thus nothing worked! He also reported that his system didn't have strdup, so I added an explicit defn of it. James B. Davis fixed a couple nroff-bugs on the man page, and said someone already gave him a dump script (see below). Richard (R.C.) Vieregge found a $ was missing from test/ftp.exp. 11/22/91 3.7.2 James B. Davis straightened out a couple things in the Makefile and asked if anyone had written a script for dump. Jeff Okamoto had a couple changes for HPUX 7 and 8 compat, involving termio stuff. Prompted by Andy Norman, added note to man page describing how to disable all argv processing while using #!. Converted passmass and rftp over to new version. 11/15/91 3.7.1 Brian Woodson asked me about the Tcl_WaitPids "got unknown process" panic. This is a Tcl bug that John has promised to fix. I'll document how to avoid it in the man page. Incidentally, I'm going under the knife tomorrow for three torn cartilage in my wrist. The doctors say it may be a couple days to couple months. Until I get back, hang in there. 11/13/91 3.7.0 Yet another bug discovered (and fixed). "expect eof" was failing to remember the buffer, and expect_out(buffer) was empty upon return. Brian Woodson noted I forgot to document the -i flag of close. 11/12/91 3.6.0 Sean Cunningham reported that he couldn't open /dev/tty from 'at'. 'spawn' was incorrectly not executing code to claim it was a controlling tty. BSD only. 11/11/91 3.5.1 Brian Woodson notes that version 2 and 3 treat the following differently. proc p {} {spawn s}; expect In v3, spawn_id is locallized by the proc, and thrown away when p returns. Unfortunately, in v2 due to some sloppy coding on my part, spawn always affected the global value of spawn_id. This differed from the handling of other variables, and in v3, this unusual behavior had to go, because the multiprocess handling and the large number of variables implicitly set (especially by the expect command) demanded that I be more systematic about how this was done. Since I never depended on this behavior, I never documented it as being something you should rely upon. Alas. To fix it, add the line global spawn_id to the beginning of any proc that calls spawn and needs the value of spawn_id implicitly defined outside of the proc. 11/6/91 3.5.0 Drew Whitehouse hit a bad pointer. I forgot an initialization in expect.c which caused problems when an EOF occurred which had no eof pattern. 11/2/91 3.4.0 Added FAQ from various questions people have sent me and my replies. Made CONVERTING file on converting from 2 to 3. Nelson H. F. Beebe found a missing declaration for exp_tty_original in bye() of main.c. How come the Sun C compiler doesn't complain about this!?!!? Nelson also reported that SunOS 4.0.3 had a problem including varargs. It turned out that old varargs had check for reinclusion, and tclInt.h also includes it. So I added an #ifdef va_dcl and put my inclusion after tclInt.h. 10/31/91 3.3.0 Converted most of the examples. Three more to go. Worked on man page some more. Modified expect so that if timeout > 0, and nothing in the buffer matched, it will force a read, no matter how long the preceeding code took. This may be hard to understand, but is the intuitive behavior that I always desired. 10/30/91 3.2.0 Fixed bug in eof handling. Converted some more of the examples, and added to Makefile. 10/29/91 3.1.0 Fixed slight bugs in tty mode switching, pty initialization (via stty). Fixed expect library. Fixed compatibility code for non-BSD systems. As usual, I could only test it so far, not having all these systems at my disposal. I don't expect major problems though, since the basic functions I depend on haven't changed. Completely rewrote handling of continue, return, etc in expect, interact, interpreter. It's actually systematic now. Checked with John O. about some code to bounce wild return codes, which he said was a mistake and would remove, so now I can pass my own return codes different from Tcl's. To get | to return -> TCL_RETURN TCL_OK (no return) V expect return default continue -expect interact return -tcl return default interpreter return -tcl return default What this table says is, to get "interpreter" (for example) to return TCL_RETURN to its caller, you must say "return -tcl", because "return" makes it return TCL_OK. The "argumented" versions are considered to be the uncommon form. In particular, I'd be surprised if anyone ever uses the -tcl argument, but it's there for completeness and consistency now. Put together a FAQ. Needs more work, but hopefully worthwhile as is. Computing Systems with Expect article appeared a couple days ago. How ironic that it describes the old version of Expect. Nonetheless, it looks ok. 10/25/91 3.0.0 alpha! First release of Tcl-6.0-ready code. It might fly for a couple seconds. Here is a quick list of changes. Besides Tcl incompatibilities, Expect incompatibilities are flagged below as: ** major - scripts definitely won't run if they depend on this * minor - scripts probably will run but there is some subtle change that should be examined). ** Select renamed 'ready' and undocumented. Seems pointless now. Added support to expect command for waiting on patterns from different processes. The old version implemented this via 'select' but but it is much simpler via expect. Added -i to a number of commands to signify a spawn_id which overrides the variable. Added any_spawn_id to match any spawn_id. An explicit null pattern, forces a spawn_id to be considered when all it can possibly match are any_spawn_id patterns. * output is no longer flushed to expect_match upon timeout. May be multiple buffers now, so it doesn't make sense to flush just one.) -n was added to disable transfers from input buffer to expect_match var. I suspect it will only be used for experimentation. Added expect -re for regular expressions. Added expect_out array to retain indices and strings of partial matches for ** for both glob and re. expect_match has been renamed expect_out(buffer). A la Tcl, added -nocase for both types of patterns. (Oddly, Tcl's case only does it for regexps.) By popular demand, unanchored glob patterns. Old patterns will continue to work, since earlier interpretation was much stricter. Unfortunately, unanchored matches make certain user errors easier. For instance, people will send answers before seeing all of the question. Typically, output can 'look' ugly, as answers land in the middle of other things. To anchor patterns, use ^ in beginning and/or $ at end. Added expect_out(spawn_id) to report which spawn_id was read. Made expect and variants understand all args as one arg. Added 'default' pattern. Added continue_expect command. Added expect_before, expect_after commands which take same args as expect, but continue to stay in effect for all expects. ** Added match_max command, deleted it as a variable. The old way was too coarse for use over multiple spawn_ids. With no arg, returns current max. Takes -i flag and -d for default. Added globbing to spawn command. Added optional -i spawn_id to wait. Added optional -i to send (and all it's variants). Renamed trace to 'strace' since it conflicts with Tcl's new trace command. Since 'trace' traces variables, I figured 'strace' wasn't too bad (for "statement trace"). I felt obliged to make it short and not as obliged to make it as meaningful since it will probably invariably be typed by hand. Made timeout == -1 mean infinity. Made interact do pattern matching in both directions via use of -o flag. Added -F flag for convenience. If -f or -F used, interact can no longer be overrun. In particular, if more characters arrive then match a pattern, remaining characters will be buffered rather than thrown away (old behavior). Patterns may now be substrings of one another. Made interact optionally take all args as one. Default action is now 'interpreter' (see below). interpreter now forces cooked mode, and echos results so you don't have to constantly say "send_user [...]\n" * Interact reads characters that have been buffered but not matched by expect. And vice versa. Does anyone care? (My rogue script did.) From discussion with John Conti, I decided to make 'interpreter' a separate command to start up interactive command processor. Changed default action in interact to this. Added eval depth and event id to prompt to interpreter. Added expect_library which contains path for commonly sourced expect scripts. Automatically source expect.rc out of expect_library unless -N given. Automatically source ~/.expect.rc unless -n given. Added expect_version command to print and/or verify script is compatible with running expect. Tcl version is also tested. Felt it was worth making this a command because it's such a pain to tear apart version strings. Tcl's close and exit are both subsumed by expect's commands of the same name. Rewrote mode switching code so that "system stty" is handled specially. This allows interact and interpret to get the modes they want, without burning the user. It is now much easier to leave expect in raw mode all the time, but the choice is up to the user. Added vgrindefs, courtesy of Brian Fitzgerald. 9/23/91 Tcl 6.0 released. This new Tcl has some incompatibilities with the old Tcl, so as long as everyone is changing their scripts already, I'm taking the opportunity to make some incompatible changes to Expect that I've wanted to do for a long time. 9/11/91 2.67 Ed Klein added support for SVR4 in the form of pty_svr4.c and mods to command.c. Added explanation to man page of how to create unreadable but executable scripts. (No, chmod 111 doesn't work.) Mark Diekhans pointed out to me that there is a potential problem with the trap command: "There is no control over when the signal will cause Tcl_Eval to be executed. There is a chance that code in the Tcl library will be executing when the signal comes in and the interpreter data structure will be in an inconsistent state. This could cause all sorts of nasty things to happen. In our Extended Tcl (4.0) we added signal handling. but the way we implemented it was to have the signal handler set a global flag. We modified Tcl_Eval to check the flag after it finishs executing each command. If the signal came in, Tcl_Eval then returns an error such as: "SIGINT signal received". Signals may then be caught with the catch command and processed." 9/10/91 2.66 Don Jackson found a syntax error in the usage error message of the example ftp-rfc script. Marty Olevitch provided mods to support MORE/bsd. Namely, added #include types.h to expect.c and extern int errno to a number of files. Scott Hess noted a potential problem in interact. Since interact only checks patterns at beginning of reads, user can conceivably type fast enough so that patterns are typed in the middle of a read. In reality this doesn't happen, but Scott was driving one expect with another expect and in this way provoked the behavior. The solution is to read chars one at a time, either by read(,,1) or buffering in a stdio-like way, but I'm not going to do that because the code should really be rewritten entirely and it just isn't worth it, since it is so easy to get around at the user level. Steve Legowik found that spawn-disconnect sequences fail. The pty testing I added in version 2.55 causes expect to regain the slave as a controlling tty, which generated SIGHUPs. If anyone knows a clean way to avoid regain controlling ttys, let me know. For now, I just set SIGHUP to SIG_IGN in the disconnect command. 8/14/91 2.65 Old passmass script changed root password. I renamed it to passmass.old, and made a new one which works for any account. It also supports yppasswd, telnet/rlogin, different names for accounts on different machines. Handles VMS machines, too. Added Computing Systems paper to expect distribution and moved all expect-related things to separate expect directory in our ftp directory. 8/5/91 2.64 Achille Petrilli found that on an SGI, the expect command ocassionally returned "no more processes". He traced the problem back to O_NDELAY in the open, which was taken as-is from the man page, by someone else who's code I didn't look at too closely at the time. The result works now. Oddly I thought I fixed this error myself when the SGI support was first installed, but I cannot find it. I evidentally screwed up. 7/31/91 2.63 Steve Legowik wanted to implement callback by having a modem dial out and NOT go away, but interact in the reverse direction. I added "interact -u" to support the idea of changing the user from the default stdio to a second spawned process. The result is that we can now write a modem callback program that doesn't depend on the cute trick of having getty recognize DTR which only worked when the modem was directly connected to the computer. In Steve's case, there were several network switches in the way. Added "overlay" function which is similar to plain "exec" in shell. (Too bad Tcl took the name already.) Added robohunt scripts to the test directory. I wrote these back in January, '91 and forgot about them til now. But I suppose they are illustrative (at the very least of how to generate truly random numbers). Ha. 7/20/91 2.62 Carl Witty pointed out my fdset implementation (for systems that don't have it) wasted some space. I had commented it correctly, however, making the incorrect code obvious (except to me). Robert Howland pointed out that expect complained about not running from a real terminal under cron. Oops! So I added a test to skip saving/restoring terminal modes when fd 0 is not a tty, since this is obviously pointless. 7/19/91 2.61 Oops. Forgot to include getline and getline.exp examples even though they have been documented! 7/17/91 2.60 UMich changed interface to weather system necessitating change to weather script. 7/9/91 2.59 Didn't correctly comment things right in Makefile. Fixed. Changed 'close' in gethostbyaddr example to 'catch close'. 6/22/91 2.58 Made new file (pty_sgi.c) for supporting Silicon Graphics ptys. Silicon Graphics select fails to see eof immediately but poll works ok. Unfortunately, there was an error in inter_poll (bad_io was uninitialized). Silicon Graphics works now. Andy Norman notes that linking expect with the BSD compatibility library under HP-UX, libc.a must be loaded before libBSD.a. Modified Makefile to reflect this. He notes that there is a problem with expect not reading an EOF from the current process. This should go away with HP-UX 8.0 when select has been enhanced to flag exceptions in the readfds argument. Probably inter_poll would work. Edward Haines notes that close returns EPERM ("Not owner") on his Sun 4.0.3. This is rather startling! (That's what I get for checking the return value of close!) He said it is possible that they have modified things (viz. DDN X.25 is loaded), but it still sounds incredible. For now, I told him to either "catch" all closes or to remove the check in the source code. Added example scripts: ftp-rfc retrieves an RFC from uunet via ftp. archie mails back a listing from the archie server. Add the rest of Hal Peterson's changes for Cray support, 1) fixing a problem where spawned processes flushed unread I/O upon process exit, and 2) creating processes with the correct uid. See his comments in command.c for more info. 6/6/91 2.57 (On Cray) made signal handler declarations right. Added missing #endif. Added includes to pty_unicos.c. Fixed bug in two bugs in CmdSend, one involving send_stderr, the other send_user. All of these are from Hal Peterson. Added gethostbyaddr as example script. Given an internet address, it returns the domain name. By querying neighboring hosts if the name server fails, a much higher probability of returning the name is obtained. 5/30/91 2.56 Mispelled "match_max" as "max_match" in rftp script. This caused files after the 2000 byte mark (per directory) to be skipped. 5/21/91 2.55 Revisited BSD pty code to reject ptys that have either slave or master side already open. This fixes problems rare problems such as expect not being able to see EOFs from the child proc. (because another process still has the pty slave side open). USG and Cray pty code could probably use this code, too. Fixed bug in expect library (lib_exp.c) which caused output to be copied to stderr instead of logfile when logfile_all was set. Per Sreedhar Muppala . 5/16/91 2.54 Fixed weather script to accomodate occasional Weather Watch that would cause an unexpected initial question to pop up. 5/15/91 2.53 Added comment to BUG section of man page describing pty misbehavior with non-interactive programs (search for "553061" below), as per Hal Peterson . Removed note from README about asking Ousterhout for SV TCL at his request. 5/11/91 2.52 Fixed a syntax error that Bruce Larson found in inter_poll.c 4/23/91 Computing Systems accepted paper on Expect for issue 4.2. 4/18/91 2.51 Added some example scripts: weather - retrieves weather forecasts from National Weather Service via University of Michigan server. rftp - ftp a directory hierarchy (i.e., recursively). 4/18/91 2.50 Changed timeout to apply to total time in expect rather than per read(). Original behavior hung forever when my modem test script started listening to a modem than spit out 1 spurious character every 10 seconds (very consistently). Hal Peterson noted that exp_spawnv's args didn't match documentation. Fixed in favor of documentation. Several other funcs don't match header file (but typechecking is avoided during compilation), because it was too hard for me to make the header file ANSI compliant and support varargs (which is undeniably more portable than stdargs at this point). Fixed prototype declarations (again) in expect.h for C++ and Standard C. Verified with GNU, G++ and Sun C (proto-less). Added exp_disconnect to library. Moved alarm calls closer to read() to tighten windows. 4/11/91 2.49 Changed passmass script to use timeout of 1000000 instead of 10000000000 after discovering that Ultrix sleep(3) doesn't sleep at all for large values! Added support for systems without dup2 (SVR2) per . Added test/Makefile to shar as per Chris Pribe . 4/4/91 2.48 Fixed possible problem with poll in inter_poll.c for systems that check for a valid address even though no members are used. 3/27/91 2.47 Added support for Cray Unicos 6.0, which of course is different from Unicos 5.1 (which was different from everything else)! This and other minor bugs fixed courtesy of Pete Termaat. 3/19/91 2.46 Removed a "feature" which caused patterns with no whitespace not to be run through SplitList. While not documented not to do so, this was mystifying even to me when I saw it. For William Waite. The result actually simplified the internal handling of multiple patterns, removing some excessively complex logic that I thought would be helpful for speed, but that in retrospect, was not that important. 3/16/91 2.45 Added my own definition of FD_SET, fd_set, etc, test for SIGABRT, and support different types of signal arg func definitions to support SunOS 3.5 as requested by William Waite . 3/14/91 2.44 Removed redundant def'n of pty_stty in pty_usg.c, redef of sprintf and added signal.h to command.c to make compiles cleaner on SV3 and HPUX machines. All compliments of Mike Gourlay. 3/10/91 2.43 Added -s (for slow) and -h (for human) flags to send. This had been requested by several people including Frank Terhaar- Yonkers (who actually wrote and tested a "send_slow" command), and Steve Simmons who suggested the "human" option (over a year ago), and Brian Woodson (brianw@swqa-sun.ESD.3com.com), who requested both! Thanks to NIST statistician, Keith Eberhardt, who taught me about the Weibull Distribution. According to Jim Thomas , 3b2 requires defines for R_OK and W_OK. Added to pty_usg.c. Added support for "-" as file name on command line to mean stdin as requested by Steve Clark . Wrote passmass (change root password on a set of machines) as requested by Ken Manheimer . Added to test directory. 2/21/91 2.42 Removed reinstallation of signal 0 in signal handler. Added hook for setting initial pty parameters when started in the background. Should've done this a long time ago, but I was never really happy with my solution and had hoped I would think of a nicer method. I only hope this is clean enough. 2/10/91 2.41 Added buffer_full keyword to solve Brian Fitzgerald's problem. It disables "forgetfullness" so that when expect's internal buffer hits match_max, whatever it has returns at that point. Didn't add this to the library version, because I want to think for awhile about the cleanest way to do it. 2/4/91 2.40 Per Brian Fitzgerald (fitz@mml0.meche.rpi.edu), fixed error in interact example on man page which incorrectly implied that "kill" was built-in. Added fork/disconnect functions. This solved the problem of Jerry Friesen (jafries@snll-arpagw.llnl.gov) who wanted to run an expect script that asks for a password and then goes to sleep for awhile before waking up to run in the background (to run a program using Kerberos). 1/30/91 2.39 Per Jim Johnson (jaj@mlb.semi.harris.com), added declaration and documentation for exp_pid in libexpect. 1/10/91 2.38 More mods from Frank Terhaar-Yonkers. Also, some requests from Pete TerMaat (pete@willow.cray.com) for features: 1) a single-step facility. Yeah, that would be nice. No ideas on how to do this easily. 2) Generate scripts automatically after watching a session. This is hard. Read more about this in the FAQ. 1/10/91 2.37 Added support for Cray Unicos 5.1, all courtesy of Frank Terhaar-Yonkers (fty@sunvis.rtpnc.epa.gov). Most of it had to do with pty support. 1/8/91 2.36 Modified expect.h to support C++ and ANSI prototypes. Added appropriate example in test directory based on chesslib.c. 1/7/91 2.35 At the request of Jan Norden (jano@imdpy1.im.se) added NO_MEMCPY and NO_STRING_H defines for Pyramid. 1/3/91 2.34 Added a check to protect against a longjmp occurring between i_read and alarm(0). Didn't think this would be a problem but evidentally a function return modifies the stack, so it cannot be returned to again. Drat! This appeared in the robohunt script I wrote which plays hunt automatically and uses 1 second timeouts. 12/19/90 2.33 Add signal to sighandler, to reinstall signal for those systems that need it. 12/12/90 2.32 Removed test for args to expect. I only recently realized that no args still allows a valid way to check for timeout and eof! 12/6/90 2.30-1 Mike Gourlay (mike@penguin.gatech.edu) found and fixed quite a few SV-related problems that I had introduced since Clem's fixes. We eventually got it to run on his HPUX machine, a mixed breed of BSD/USG stuff. But spawning a shell worked but always produced a complaint about "no access to tty" which we were never able to get rid of, and he had a problem with exp_fexpect (but not exp_expect), although it still isn't clear if that was expect's fault. He said he would speak to some HP engineers about what he found. 12/5/90 2.29 Fixed a malloc off-by-one bug in new C library. After contemplation, revised interfaces. Decided that rather than following the original 'expect' style, it should be more like what a C programmer is used to, so I made the file descriptors be parameters to exp_expect rather than globals, added an exp_popen which is a popen equivalent, and added exp_fexpect versions which are stream equivalents. Am not happy with exp_fexpect. It is much less efficient than exp_expect, because there is no way to (portably) get fread() to return the way read() does, with less then the number of characters you supplied a buffer for. Instead, I have to call fgetc for every char. Ugh. Add a couple new examples, including lpunlock, time.exp, chesslib.c (using file descriptors) and chesslib2.c (which uses stream pointers). 12/3/90 2.28 Created C library version of expect. 11/29/90 2.27 Fixed bug in interact - when no string actions were defined, the mapping table length wasn't set at all. Made interact call printify when debugging so that crlf and other nonprintables are visible. Fixed bug in printify which interpreted some characters wrong due to parity. Added some more examples to the distribution (lpunlock, dvorak, timed_read) and put in another tip in the TCL HINTS section of the man page. 11/18/90 2.26 Fixed mismatched comment per Craig Warren (ccw@deakin.oz.au). Also improved man page entry for "interact". 11/17/90 2.25 Added -f (fast) on interact options, and made default case a little more efficient. Added explicit support for SIG_IGN and SIG_DFL in trap command. Added ability to specify signals symbolically for portability. 11/15/90 2.24 Craig Warren (ccw@deakin.oz.au) wanted to exit expect while in interact with a single character. Dan Bernstein (brnstnd@kramden.acf.nyu.edu) wanted to suspend with a single character. So I generalized interact's escape character to string-action pairs. 11/7/90 2.23 Tired of getting reports that various (Ultrix 3.1, BSD4.3) C compilers can't handle ternary conditionals returning ptr to func returning void. Made all (2) such statements into if-then-elses. Per Steve Simmons (scs@iti.org). 10/8/90 2.22 Allow "log_file" even when no log is open. This makes user programming a little simpler - they don't have to remember whether they opened the log or not. 9/27/90 2.21 Fixed bug, v2.19 introduced. debuglog(unknown string) requires a "%s" as formatting for protection against %'s in the unknown string. 9/17/90 2.20 4 syntax errors in interact_poll.c, vik@sequent.com. Added quotes to all the sends (now that this is more efficient) in the examples and man pages. Also removed a misstatement in the man page about the behavior of double quotes. 9/15/90 2.19 Removed buffering from send command. Originally, I buffered the args, so I could do it all in one write. But to send variables bigger than the buffer didn't work. I didn't think about this before. But Joe Gorman (Joe.Gorman@elab-runit.sintef.no) asked me if you could "send" a file in one command, and of course you can using [exec cat] as the argument to send, but the damn buffering prevented big files from being sent. Anyway, now it works. 9/14/90 Fixed the declarations of nflog and nferrlog. Added a #define so lack of pid_t could be controlled from the Makefile. Per Andy Holyer (and@ux.rfhsm.lon.ac.uk) 9/4/90 2.18 Added trap command to catch signals. This is nice as (among other things) it allows you to turn off the conversion of ^C to timeout which was requested by John Conti . 8/21/90 2.17 Fixed bug in printify. Forgot to reset ptr to beginning of print buffer. Made debugging info wrong. Possibly screwing up other things on overflow. Paper accepted into USENIX LISA! 8/15/90 libes Cleaned up man page. Made tabs line things up correctly, finally. Found another problem with ptys (at least under SunOS 4.1 and earlier). When last pty-slave fd closed, any unread output is lost after a short window of time (around 10 seconds on a Sun 3/60). Sent example ptybug.c to Sun demonstrating this and EIO problem found earlier. (Service Order #553061) 8/6/90 2.16 Added -f to debug command, -a to log_file command. This required significant changes, including revisiting all the logging routines, plus miscellaneous output done in special places. Noted that it cannot be done with getopt, since it could be called during main's getopt, and getopt is not reentrant! (Guess how I discovered this!!) I'm not particularly happy with the design, but maybe others won't be. In any case, I like the benefit of it and am now glad that -a was asked for. Per Harry Bochner and Ira Fuchs (fuchs@pucc.bitnet). Changed behavior of argv, so that 0 == [length $argv] when no script/args supplied. 8/4/90 2.15 Added debug command, so -d-ness could be changed while expect is running. 7/20/90 2.14 Fixed small bug in -d output from expect, which printed ^Z as ^: 7/18/90 2.13 Added wait command. A waitpid/waitspawnid would be nice and cleaner, too, but since csh doesn't need it, it is probably not worth much. Consequently, removed SIGCHLD handling from command.c. It worked under SV but not BSD. By forcing users to explicitly code waits, resulting scripts are more portable. Rewrote rogue example. rogue sometimes misses EOF (generated by close on our side) and continues reading. 7/16/90 2.12 Removed buffering from variadic log routines. This was faulting when the buffers overflowed. Cleaned up the -d output from expect, so it is much more readable. For example, control characters are now visible. 7/14/90 2.11 Added declaration for errno, to support 4.3BSD. Per Alan Crosswell. Added -i flag and related behavior. 7/12/90 2.10 Fixed bug where timeout = 0 waited forever rather than not waiting at all. 7/11/90 Fixed man page example which didn't include the blank on the end of an ftp prompt. 7/9/90 2.9 Fixed bug in send when spawn_id = $user_spawn_id. 6/27/90 2.8 Integrated some mods from clem cole (clemc@ccc.com) to support System V.3 (386/ix Version 2.02). Unfortunately, he didn't do "select". 6/25/90 2.7 Test that cmdfile and logfile are open before fclosing in child while spawning. Per Corey Satten 6/24/90 2.6 Pty master returns EIO instead of EOF when pty slave closes. Bug in pty driver? Until I figure this out, I have put in code to interpret EIO to EOF. 6/21/90 Added new section to expect man page - Tcl hints. 6/14/90 Spoke at USENIX. Went well. Added USENIX paper as separate ftp archive. 6/4/90 2.5 Fixed bug in ^C catching during expect. Changed man page to accurately describe what ^C does. Fixed bug that caused "send" to screw up when handed 0 arguments. All per Harry Bochner. 6/1/90 2.4 Made trailing empty action in expect optional, primarily to make straightline code easier to read. 5/15/90 2.3 Changed expect to strip nulls from program output since there is no way for Tcl to handle them, per Harry Bochner. 5/5/90 Added "send_error" command. 4/26/90 Got USENIX paper back from Kolstad to proof. Am depressed at how awfully they formatted it. 4/25/90 2.2 Eric Newton found that expect's special variables weren't being found inside of user subroutines. Had to do with new Tcl, which now differentiates between variables that are undefined vs. empty. 4/24/90 Upgraded Tcl from 2.1 to 3.3. 4/22/90 Added special behaviors of ^C in expect, and when profiling. Profiled rogue (at urging of Ousterhout). 4/10/90 2.1 Added select command. Added support for user_spawn_id so that you could treat user just like another process (i.e. with send and expect). Decided to leave send_user/expect, since scripts are more readable with them. 4/2/90 2.0 Changed syntax of expect to provide alternatives (a la Tcl case), per suggestion of John Ousterhout. Note that this breaks pre-2.0 scripts. 3/31/90 Got great comments from Ousterhout. (This time he said that he really liked the idea. Maybe he realizes how much it will promote Tcl!) 3/30/90 Got comments from dpk. Made me think more about Perl. 3/28/90 Evi said I should turn the paper in unformatted and they will format it. (She's kidding, I hope.) 3/27/90 1.8 Rewrote interface so that raw arguments can be passed in like a shell. I'd been thinking about this for some time, but Eric Newton finally prodded me into action. 3/25/90 Got first corrections for paper - from Sue Mulroney! 3/24/90 Observed that it is possible to use the #! syntax with expect. I asked John O. about this (his choice of # as a comment character), and he said it was pure coincidence. Deprecated request to end scripts in ".exp". Ted Hopp volunteered to be my Center WERB reader. 3/23/90 Finished 1st draft of USENIX paper and sent copies to John Ousterhout and panel chair, dpk@morgan.com. 3/20/90 1.7 Deprecated "stty", and added more general "system" command. Sent Evi some complaints about the business of not allowing camera-ready at USENIX. 3/17/90 Sent copies of man page to Doug Gwyn and Larry Wall for comments. Note that gwyn downloaded it. 3/16/90 Am really irritated by USENIX. My paper has been put in a session against another session, the BSD people. Furthermore, they called my paper an application, when it is no more so than any other shell or language. Better I should be in "lessons learned". Mashey said take a hike, i.e., it was too late to change the schedule. On top of that, our session has four people in it, so I'll have very little time to speak. Grrrr. 3/13/90 1.6 Added "stty", because without it you can't do things like turning off echo to accept a password. 3/8/90 1.5 Abstract was accepted into USENIX!!!! Time to start writing it! Sent man page to Ousterhout. He didn't seem too impressed. Added "send_user/expect_user" after listening to Ken complain about how shell could not do timed reads. Actually it can, but expect does it much more naturally. Deprecated echo. Now, I realize that expect can be viewed as a shell! Changed logfile/loguser to log_file/log_user to match all the variables with underscores in them. Barry Warsaw asked if there was any way one could execute any command from interact (apparently without any reason in mind). Nonetheless, it is a wonderful idea, and I changed the "abort character" in interact to an "escape mechanism". After escaping, you may execute any command. return duplicates the old action of the abort character. Now you can do interactive job control, recursive interacts, etc. You can bet I didn't get this right the first time! At Scott's request, fixed bug related to pty initializing. Scott was putting expect in the background which disassociated it from a tty, and I was blindly copying the tty parameters without checking to see if they were meaningful or not. Tightened up exit code. Fixed bug in spawn so it would print error messages when it failed. Spawn sends back the error message in the pty, if the fork succeeds but exec fails. Cute! Added "close" command. Makes scripts much shorter and cleaner. Return string matched by expect directly, rather than setting a special variable. Added "match_max" feature. Probably no one will ever use it. Added trace command. 3/6/90 1.4 Rob Densock was the second user, and suggested (demanded?) the idea of the loguser command. I added it and changed "log" to "logfile" making the first incompatibility with existing scripts (sorry, Steve). 3/1/90 1.3 Trying to make pty code more robust. Many questions unanswered by manuals. Did a lot of guessing. While debugging, looked through pty code in gnuemacs to see if I might increase portability somehow. I almost barfed when I saw all the funky ifdefs on weird ioctls. Found lots of comments like "this might work". 2/28/90 Sent a short Tcl bug list to John Ousterhout. He thanked me! 2/22-3/90 1.2 Hooked my first user, Steve Ray. Surprisingly, he only found one bug in the code (exit didn't handle args correctly), but it was obvious that I need to put more explanation in the man page. Many of the examples in the man page are based upon his probl.. questions. Thanks, Steve! 2/20/90 Posted news about expect to "general" newsgroup locally. 2/15-20/90 Talked to local POSIX reps and then to Steve Albert (AT&T) about portability of select, wait and other system calls. I'm not impressed by 1003.1. 2/9/90 John Ousterhout answered some questions I had about Tcl syntax. I like this language! 2/8/90 1.1 Sandy Ressler suggested the idea of being able to spawn multiple programs at the same time although he didn't say how. It took about a day to design and code the spawn_id hook. Extremely difficult to support this with uucp-style kludge. Switched to using select. So much for portability. I investigated how to do this portably, and spent some time talking to NIST & AT&T POSIX representatives. Unfortunately, portability (especially when it comes to select()) remains a dream. Provided multiple versions of "interact" depending upon what OS you are running. 2/7/90 1.0 Completed first cut of "sex" (for "Smart EXec" or "Send/EXpect"). Supports send, expect, echo, log, spawn, interact. Spent a lot of time making "log" write to log in just the right order (across fork and while debug flag enabled). Ended up writing a bunch of variadic log routines. Fooled around with uucp-style multiple processes versus one process doing select() to read asynchronously. Using uucp-style for now, since it is more portable. Gave up on pipes, and switched to ptys. Pipes seem to be messed up by ftp, perhaps because it goes into raw mode? Ptys are more efficient and cleaner to program albeit less well-documented and portable from system to system. Ken Manheimer helped me explain what the program does. I kept saying it does "send/expect" processing, and he kept insisting that was meaningless to everyone. (In fact, it comes from uucp, and I guess uucp hackers are indeed a dying breed.) Ken gave me an elegant enough sentence that I expanded it into an abstract and sent it in to the USENIX conference the following day (two days after the deadline). I noted that the uucp documentation I referenced in the submission is dated October 31, 1978! 1/30/90 0.0 Got a copy of Tcl and went to work. Tcl was exactly what I need. Plus, it is easy to use, AND it is documented. 1/25/90 Attended Winter 1990 USENIX in DC, with the goal of banging heads with some other gurus in hopes of finding a good send/expect language for a generalized stelnet. Had looked at uucp and kermit but found nothing general enough. Listened to John Ousterhout's presentation on Tcl. By the middle of the talk, I had found religion. At the end when he said it was public-domain, I was ready to orgasm. */*/88-89 Spent a lot of time telling Scott how useful his program could be if he made it more general. I thought it wouldn't be that difficult to make more generic. Scott was interested but not enough to do it. 9/25/87 Helped Scott Paisley write a program called stelnet, that forked a telnet and did very simple send/expect processing. It used pipes, not ptys. It had no pattern matching, and only straight-line control without error handling. Nonetheless, this got me to thinking about making stelnet more generic.