C-Kermit Version 4C(058): Status, Bugs, and Problems As of: 19 March 1986 Note: Version numbering system changed from decimal notation (like 4.2) to number-letter notation (like 4C) to avoid confusion with Berkeley Unix version numbers. See ckuker.upd for list of bugs already fixed. -- SUPPORT FOR DIFFERENT SYSTEMS: -- 4.2BSD: "make bsd" works. Should work on VAX, SUN, Pyramid, & others. 4.1BSD: "make bsd" should work for this too (did not in C-Kermit v4.2). 4.3BSD: unknown, but "make bsd" should(!) work. 2.9BSD: Support in but untested: "make bsd29". Should be the same for 2.8. Version 7: works ok on the systems tested so far using "make v7", but some fiddling with the make file is necessary for proc table definitions; see the makefile and also ckuv7.hlp for details. PC/IX: should work ok with "make pcix". ATT 3Bx systems: should work ok with "make att3bx". One person reported that under Sys V R2, the macro 'unchar' declared in ckcker.h has a name conflict with the unsigned character typedef included from sys/types.h, preventing successful compilation, and changing the name to something like 'myunchar' fixes the problem (can do this in the makefile, as in -Dunchar=myunchar). Zilog Zeus: works with "make sys3" but all occurences of setjmp/longjmp in the source have to be replaced by setret/longret. HP Integral PC -- "make sys3", but either remove code that sets up lock files, or have the lock files put in a directory that is guaranteed to be there, like /tmp. Some problems reported when running under csh. Most other System V or System III based systems can build a working Kermit with "make sys3" PDP-11's running a System III or V based Unix and which have no I & D space should use "make sys3nid". DEC Pro-350 or -380 with Pro/Venix V2 (not V1) -- uses the regular "make sys3" or "make sys3nid", but the file ckcfio.c might have to be edited first to reduce the value of MAXWLD and/or SSPACE. See below under HINTS for details. Valid Scaldstar CAD system -- There's a "make valid" in the makefile, but reportedly one thing is still lacking: ckutio.c needs to #include. For other Unix systems, see the makefile. VAX/VMS: support added by Stew Rubenstein at Harvard and Martin Minow at DEC. Has to be built with CKV*.COM, needs development to handle all the VMS/RMS features and to improve performance. Has its own build procedure. See CKV*.*. VMS-specific bugs in CKVKER.BWR. Macintosh: Support added at Columbia, has own makefile, etc. See CKM*.*. -- HINTS -- - If the program dies with a message like "malloc fails in splitpath()" whenever it tries to parse a filename (as in the "send" command), then the amount of space allocated for filename expansion in the module ckufio.c must be reduced. This can be done by changing the #defines for MAXWLD (the maximum number of filenames) and SSPACE (the size of static string space) to make the numbers smaller. - When modifying or writing KERMIT code, do NOT pass to a function (e.g., "signal") the address of a static function. Doing so may break VENIX code mapping. If you must pass the address of the function, make it global and pick a "non-generic" name for it that will hopefully be unique and yet informative. -- BUG LIST -- First, a disclaimer must be made. C-Kermit attempts to support all post-V6 Unix variations on all machines. This is a tall order, and requires careful attention to certain details. As changes are made (and C-Kermit is still in stage of fairly rapid development), there is always the chance that a change -- made to introduce a new feature or fix a bug -- will not work as intended on some systems, even though it was tested successfully on others. The main area to watch out for is not system differences (which are handled fairly well in the system-dependent ck?[ft]io modules), but in compiler differences, especially int/char confusion. Characters should be stored in variables of type char, not int, and char/int conversion should be avoided because of problems introduced by sign extension. And i/o should not be used to read characters into int variables, because the way in which the system stores the character in an int varies from system to system (e.g. 68000s put them on the left, the VAX on the right). If you have received a C-Kermit release that does not work correctly (except for the bugs & restrictions noted below), it is not because the release was not thoroughly test -- it was -- but because it was not tested on your system since the last time changes were made, because of a lack of such a system to test it on. If this happens to you, please try to track down the problem and report as specifically as possible back to us. General problems: - The program is too big, with too many features; source is too large to fit on some disks. Needs to be reorganized so that a minimal Kermit can be built for any system, and then frills can be added on if desired -- interactive command parser, help strings, dial command, script command, etc. - There's not a full enough set of features available from command line invocation. Commands like "bye" are missing. This is mainly to keep the "kermit -h" help message small enough to fit on one screen. - Conditionalizations are not done clearly. In some cases it might be better to have compile-time flags for features, rather than systems, or generic system names, rather than specific vendor/machine names, to avoid excessive nesting or repitition of compile-time variables. Constructions like "#ifdef FOO | BAR" are avoided because many compilers don't understand them; the alternative is to repeat code under different conditionals (to accomplish an OR) or to include it within nested conditionals (AND), sometimes applying De Morgan's law to achieve the desired result... - Program's return code might be wrong in some cases (in 4.0, it was always zero; in 4C some attempt is made to return correct codes for failure and success). - On some systems (e.g. TRS-80 Model 16 with Xenix V7, or HP-9000 HP-UX) C-Kermit reportedly runs VERY SLOWLY. The program could certainly do with some tuning -- but not at the expense of modularity and transportability! -- but in the meantime, it can probably be sped up a lot by removing the -DDEBUG from the makefile, eliminating hundreds of function calls, many of them in critical code (one user reports a 1250% improvement doing this on the TRS-80 Model 16!). Another speedup could come from allowing more systems to take advantage of "myread()" -- the nonblocking version of read(); see below. - Even on systems that use myread(), performance is much worse when parity is "on" because single character reads are still used so as to allow detection of the end of packet. This will be improved in a future release. - Micros like the DEC Pro-350, when in "IBM mode" (set flow none, set handshake xon), might not be able to keep up with incoming packets, even at relatively low baud rates. The Pro-350 OS relies heavily on XON/XOFF to avoid losing characters at the port, but XON/XOFF must be disabled for communicating with IBM mainframes. The only solution is to lower the packet size to 20 or 30 to reduce the probability that data is lost in any particular packet, and maybe also reduce the baud rate. - Need 'set' commands for retry-threshhold and server timeout. - The program could be a little bit less cavalier in its treatment of files. For instance, when receiving a file (with "warning" turned off) it will overwrite any existing file of the same name. That's ok, but what if the user types ^F or ^B to interrupt the transfer? This does not save the existing file -- it's already been destroyed by the open() of the new copy, which in turn is discarded as a result of the interruption. Maybe Kermit should always make a temporary, unique name for incoming files, and then rename them to their real names only after the transfer is complete. But that's no good on systems (like the Macintosh) that don't have disk space to burn. - Local versus remote mode is not, and probably can not, be determined automatically upon startup. For instance, if you build Kermit with "make sys3" on a 3B20 and a 3B2, the program has no way of knowing whether it's running on a timesharing system (the 3B20, where it should be remote by default) or a workstation (the 3B2, where it should be local by default). If you find that Kermit comes up on your system in the wrong mode, you can put the appropriate "set line" command in your .kermrc file -- "set line /dev/tty" will always put you in remote mode. - Local mode file transfer display could be improved. On tty-style displays, the "percent done" could be shown by doing something like "0...1...2...3...4...5...6...7...8...9...". - If the program crashes or is halted (killed) from outside while it has the terminal in a not-normal mode during command parsing or file transfer, the terminal mode is not restored, and lock files are not cleaned up. There can be no fix for this within C-Kermit itself. Subsequent Kermit runs may still fail because the device is already opened by "another process" (you have to log out & log in again to clear this one up). - The shell's interrupt, delete, and kill characters may interfere or conflict with those used by the Kermit command parser. In any case, there is no way to change Kermit's editing characters to conform to user's taste. - Status info should be updated only for real file transfers, not transactions like "finish". - "!" command requires a space after, contrary to the Unix user's normal expectation. - Many people have asked for a system-wide startup file similar to the user's .kermrc file, perhaps with a conditional way to escape from it if the user has her own .kermrc file. This notion might be extended to include the entire hierarchy system -- home -- current directory. - A deeper problem with the initialization files is that the file is only taken when the program enters interactive command dialog. To be consistent, it should also be taken when the program is run via command line arguments. - Some users report that it would be more desirable to have an error during execution of a take file return directly to command level, rather than pop to the invoking take file, in case the error in question is of such severity that it would cause all subsequent commands in the stack of TAKE files to fail. Best to have a SET command to control this behavior. - Some users report that the program should make no internal distinction between running in foreground or background, so that program exit codes are consistent, making it easier to debug shell scripts that invoke Kermit. - Since Kermit opens and closes the communication line with each command line invocation, it is not convenient to use it in scripts in which it is repeatedly invoked (e.g. a print spooler). - Variable names are sometimes confusing, especially the send/receive parameter pairs (spsiz/rpsize, mystch/stchr, npad/mypadn, rtimo/timint, etc). This is mostly history... they tend to agree with the names used in other Kermit programs. Still, they should probably be changed. Ditto for some of the procedure names. - Some C compilers do not support variable names longer than 6, nor function names longer than 5, and some are not case sensitive (the DEC-20 C compiler has all these restrictions, and the V6 Unix C compiler has some of them). To ensure the widest possible portability, all identifiers should comply with these restrictions -- currently many do not. - When the C-Kermit server is given a host command (or even some generic commands like 'space'), it might have to think for a long time before returning output. In this case, it should set a timer while waiting for input from the fork and whenever the timer goes off, it should send a null data packet to prevent the other Kermit from timing out. - When connecting back to C-Kermit after a transaction, or after finishing the server, it may be necessary to type a ^Q to clear up an XOFF deadlock. There's not much the Kermit program can do about this... - When interrupting a send with ^F or ^B, an appropriate message does not seem to make it into the transaction log. - The transaction log should record file type (text or binary). - The debug log may contain bare CR's, which can be disconcerting. The debug() function should remove them. - On some systems, "unchar", which Kermit uses to perform the inverse operation from "tochar", is a reserved word (e.g. in ctypes.h). Get around this by doing something like "-Dunchar=xnchar" in the makefile. ckucmd.c: - Some editing commands don't work exactly right, like ^W after "show param". - Interactive Kermit commands that come in from a pipe (as in "cat foo | kermit") are echoed. Some people think they should not be. The fix can be made (at some expense) in getwd() by adding a line to the first if () {} block, "if (!isatty(0)) echof = 1;". ckcfn*.c: - Some people have reported that CRLF between packets rather than just CR make rpack spuriously detect an invalid packet (not verified). ckuus*.c: - 'show parameters' gets some of the send and receive parameters mixed up? ckufio.c: - ckufio currently goes to a lot of trouble to traverse the directory in order to expand "*" and "?" in wildcards. Maybe it should just fork the user's customary shell and have it do the expansion. This would allow fancier filespecs to be given, like "~/ck*.[cwh]". But it would slow down features like filename completion and menus in the interactive command parser. (Find out how Unix FTP does it...) ckutio.c: - Myread() should be expanded to include systems that can't do nonblocking reads, but can find out how many characters are in the input buffer -- this would eliminate calling read() in a loop for each character during packet transmission (e.g. Pro/Venix V1 could use "ioctl(x,TIOCQCNT,&y)", V7 could use its buffer-queue-peeking trick, etc). - There should be a timed option for ttoc() for use during connect so you can escape from XOFF'd or other stuck conditions. - There's no good way to select baud rates higher than 9600. Most Unix systems don't supply symbols for them (unless you use EXTA, EXTB), and even when they do, the program has no way of knowing whether a specific port serial controller supports those rates. - Use of RAW|TANDEM in 4.1 BSD gives only unidirectional, not bidirectional, flow control -- xoff's from the terminal are treated as data rather than control signals. Symptom: possible loss of characters during CONNECT. - On some systems, there is an undesired interaction between the various open() modes, fnctl(), and ioctl() invocations when modem-control lines are involved. Some of this due to bugs in some Unix implementations or to inconsistencies between them (e.g. as to the behavior of TIOCEXCL, etc). ckudia.c: - Some systems do not allow users to manipulate dialers directly. - Should support a "phone book" (this would actually go in ckuus*.c). - Pro/Venix V2 (and some other Sys V-based systems) are having DTR-dropping problems when dialing. With Pro/Venix V2, a workaround is to get the system to ignore the modem control signals and treat the line as a direct line by issuing a "setline -d xxx" command, where "xxx" is the device node (e.g. com1), rather than "setline -m xxx". - Hayes modem dial string should be ATD, not ATDT, so that dialing will occur in the mode that the modem is set up for (does everyone agree?). ckuus*.c: - When an alternate filename is given for an incoming file name, and the alternate name is uppercase, the name is lowercased. Alternate names should be taken literally. - The send command should have the same syntax as the get command to allow multiple filenames to be given on a single line: send foo bar baz <-- send all these files or: send Local file(s) to send: foo bar baz Name(s) to send it/them under: x y z The latter form could be risky, of course, when mixed with wildcards, and in any case would require major rewriting of many parts of the program (which probably will never be done). - Baud rate selection currently requires user to type a number, which is then verified against a system-dependent table. Better to have a baud rate keyword (cmkey) table defined in the system-dependent module, so user can abbreviate (e.g. '9' for '9600'). - No way to put trailing comments on commands. - ckuus2.c makes the C optimizer run out of space under PC/IX and some other systems. Protocol (ckcpro.w, ckcfn*.c): - No way to interrupt a protocol transaction until after it starts successfully. For instance, no way to interrupt the timeouts and retransmissions of the initial packet when the other side is not responding, except for killing the whole program. Cure: check for keyboard "interrupts" during the send-init process, set c[xz]seen. But doing this will make the state table a lot more complicated... Maybe change ^C trap to get back to command mode rather than exit the program. - When parity is in use and the other Kermit cannot do 8th bit prefixing, the user is not warned that binary files will not be transferred correctly. This can be easily remedied by calling screen(SCR_WM,...) after the send-init exchange has occurred, but this would make a dialog box pop up during protocol on the Macintosh... ckucon.c: - Doesn't do any particular kind of terminal emulation. It wasn't meant to. Filters can be used for this. Or a replacement module can be written (as has been done for the Macintosh). - When sending BREAK, should clear output buffer first (this is done in BSD, should be added for other systems to ttsndb() in ckutio.c). - Performance is poor on systems that can't check the input buffer or do nonblocking read() calls. See the horrendous code that was added for V7 to get around this (peeking into tty buffers in kernel memory). - As structured, connect mode can't include commands to toggle logging on and off or to change settings, because the fork that reads keyboard input doesn't share variables with the fork that does port i/o. - The program may become stuck if the two sides get into an XOFF deadlock. There should probably be a timer around (or in) the ttoc() call. - Reportedly, a control-S typed at the keyboard doesn't always seem to "take" when doing terminal emulation under Pro/Venix V2 (DEC micros, terminals, devices, systems are notorious for their insistence on doing XON/XOFF and attendant problems. Remember the VT180?) ------------------------------ Undigested: Date: Sun, 24 Nov 85 04:16:02 CST From: Stan Barber Subject: more notes on C-kermit 4C(057) Organization: Neurophysiology, Baylor College of Medicine, Houston, Tx One more suggestion: I would provide some mechanism to allow SYSIII compatible sites to configure what signal that might like to use to allow the child and parent to notify each other of problems. At my site, SIGUSR1 can not be used by kermit, so I modify ckucon.c by replacing SIGUSR1 with SIGUSR2. That fixes everything just fine. At least a warning should be noted somewhere (in the .bwr file, I guess) so that people will know to change it. Alternatively, I would suggest a unique name (SIGKERMIT) that the installer can define in the makefile (e.g. -DSIGKERMIT=SIGUSR2) to keep people from mucking in the source file. ------------------------------ Date: Tue Nov 26 11:45:41 EST 1985 From: dolqci!irsdcp!scsnet!sunder@seismo.CSS.GOV Subject: Zilog Zeus Kermit, Os9 Kermit Warning I am the contributer of the Zilog Zeus support for C-Kermit. Kermit WILL NOT WORK with the newest version of the Zeus operating system, i.e. 3.21. We have gotten C-Kermit to run under this new release but we had to rewrite ckutio.c. Do you want us to submit this new mod? [Ed. - Sure, especially if the new ckutio.c also works with the old release of Zeus. Or do we have to have two separate compile-time options for Zeus systems?] ------------------------------ Date: Fri 28 Feb 86 11:15:39-EST From: Chris Lent Subject: Kermit new versions, fixes, problems and comments Just put C-kermit on a Pixel 1000 which is a 68000 based Berkley Unix box. Make bsd almost works but it seems the getppid() function which is used to determine the parent process id of the shell running C-kermit is missing from the run-time library. So I made a change to ckufio.c adding an #ifdef PIXEL and substituted a kill(0,9) for kill(getppid(),1). The unix diffs of the current version of ckufio.c and ckuker.mak follow: $ diff ckufio.c ckufiopix.c 198a199,201 > #ifdef PIXEL > return(kill(0,9)); > #else 199a203 > #endif $ diff ckuker.mak ckukerpix.mak 174a175,179 > #Pixel 1000 (Almost) Berkeley Unix 4.1 or 4.2 (and presumably also 4.3) > #Pixel 1000 V2.1 (Has no getppid() function) > pixel: > make wermit "CFLAGS= -DBSD4 -DDEBUG -DTLOG -DPIXEL" > ------------------------------ Date: Sat 15 Mar 86 17:13:07-EST From: GH0N@CMCCTC Subject: Re:Kermit(057) on the HP Integral This note is in reply to my previous post about an incomptibility between C-Kermit and csh on the HP Integral that has the new (SVR2) ROMs. Installing the latest version of C-Kermit(057) did not have any effect on the supposed incompatibilty. Running with csh alone or sh on top of csh resulted in the loss of communication from the modem to kermit. When escape to local control (^-\C) was attempted the csh process stopped. It is not known whether csh is recieving EOF or some signal such as SIGINT or SIGKILL. Sometime in the near future (hopefully) I will get the oportunity to run C-Kermit from debug to find out what is happening. Gordon Haverland ------------------------------