ckc192.upd

linux终端仿真程序

Fixed a bug that caused TELNET negotiations to be ignored if C-Kermit was
used to make a TELNET connection, then a serial connection, and then a
another TELNET connection, in that order.  ckuus7.c, 25 Oct 93.

Merged in support for a whole new and different proprietary operating system,
Stratus VOS, from David Lane at Bellsouth <[e-mail]>.  This
is a full new C-Kermit implementation, including TCP/IP and X.25.  Changes to
all modules are within #ifdef STRATUS..#endif.  For X.25, the symbol is
STRATUSX25.  Elements common to SunLink and Stratus X.25 are now combined
under the new ANYX25 symbol.  All ckc* and cku* modules affected (except
cku[ft]io.c).  All modules must now #include "ckcsym.h" first thing to pick up
other symbol definitions -- the VOS C compiler doesn't allow -D/-d for this.
David also VOS-ified the ckbmkb and ckbunb programs, wart, and the malloc
debugger.  25 Oct 93.  Supporting system-dependent modules -- ckl*.[ch] --
still to come, along with VOS-specific documentation, build procedure, etc.

Last month, Olaf Seibert <[e-mail]> pointed out that C-Kermit,
particularly on the Mac, would seem to operate in stop-and-wait mode,
even though a nonzero window size was successfully negotiated.  The behavior
would sometimes change during the transfer a particular file -- windowing
one minute, stop-and-wait the next.  Olaf figured out that, in the "PARSENSE"
case (in which C-Kermit's packet reader is length-driven rather than waiting
for the packet terminator), the packet terminator is left over in the buffer.
Meanwhile, back in sdata() -- the Data-Packet sender in ckcfns.c -- the
following code:

	spack('D',pktnum,len,data);	/* Send the data packet. */
	x = ttchk();			/* Peek at input buffer. */
	debug(F101,"sdata ttchk","",x);	/* ACKs waiting, maybe?  */
	if (x) return(1);		/* Yes, stop sending data packets */

checks whether *any* data has arrived, in which case it breaks out of sending
the current windowful and goes to read a packet.  And that's just what happens
when the packet terminator is still waiting to be read.  The fix turned out to
be simple:

	if (x) return(1);
becomes:
	if (x > 4) return(1);

The minimum number of characters in a Kermit packet is 5, so we don't break
out of the send loop unless there are at least that many characters waiting.
This tiny change should speed things up quite a bit in situations where
windowing is needed.  ckcfns.c 5A(092) 25 Oct 93.

Fixed an incorrect test (nettype == NP_TELNET) in ckvtio.c.  5A(030) 27 Oct 93.
Noticed by David Lane.

Fixed incorrect setting of network type via command-line arguments.
Noticed by David Lane.  ckuusy.c.

Changed the UNIX CONNECT command to buffer keystrokes, rather than reading
them one at a time.  This turns out to be important when C-Kermit is "in the
middle", in CONNECT mode, between two other computers that are transferring
data.  Fulvio Marino of Olivetti in Italy recoded C-Kermit's communications
and console i/o for this very reason, but his changes are massive and would
require months or years of testing, debugging, adaptation to every UNIX
variant in the world, etc, which might well be worth it, since his generalized
i/o buffering scheme makes most kinds of i/o in Kermit MUCH faster and easier
on the system in terms of cycles, system calls, etc...  In other words, the
present change is just a quick fix to get a modest improvement in this
particular case.  Benchmarks using a 58K file on a network connection:

					   No Buffering   With Buffering
  SEND through C-Kermit in CONNECT mode:      102 sec         47 sec
  TRANSMIT thru C-Kermit in CONNECT mode:     237 sec         94 sec

So, about a 100%+ speedup.

The buffer size is only 256 bytes, since that's the biggest amount that UNIX
itself seems to buffer from the console (i.e. tests on SunOS showed that
conchk() never returned a value greater than 256).  No ill or side effects
were noted -- CONNECT works in the normal way during interactive use.
ckucon.c 5A(055) 27 Oct 93.

This code could not be adapted to VMS C-Kermit because of the insane way that
it reads characters from the keyboard.  It was not adapted to OS/2 because
OS/2 C-Kermit is never "in the middle".

Changed the SET NETWORK keyword table to allow any combination of networks
without getting compile-time syntax errors caused by too many or too few
commas.  Problem noticed by David Lane when he built VOS C-Kermit with X.25
and no TCP/IP.  ckuusr.c 5A(107), 28 Oct 93.

Added vanity herald for NetBSD based on preprocessor symbol __NetBSD__.
ckuver.h, 28 Oct 93.

Disabled -j command-line option if TCPSOCKET not defined.
Disabled -X and -Y command-line options if ANYX25 not defined.
Suggested by David Lane.  ckuusy.c, 29 Oct 93.

Added a new variable, \v(parity), to allow script programs to get the current
parity, in case they need to save, change, and restore it, e.g. around a
dialog with a modem that doesn't work right if it sees a parity bit.
ckuusr.h, ckuus4.c, 29 Oct 93.

Added a new variable, \v(dialstatus), a return code from the last DIAL
command, either success or else any of twenty-some specific error codes (busy,
no answer, no dialtone, etc).  This allows users to write script programs that
can decide whether to redial based on the failure reason.  For now, the value
of \v(dialstatus) is a number.  ckcker.h, ckudia.c, ckuus4.c.  The failure
reason codes are:

#define DIA_UNK  -1			/* No DIAL command given yet */
#define DIA_OK    0			/* DIAL succeeded */
#define DIA_NOMO  1			/* Modem type not specified */
#define DIA_NOLI  2			/* Communication line not spec'd */
#define DIA_OPEN  3			/* Line can't be opened */
#define DIA_NOSP  4			/* Speed not specified */
#define DIA_HANG  5			/* Hangup failure */
#define DIA_IE    6			/* Internal error (malloc, etc) */
#define DIA_IO    7			/* I/O error */
#define DIA_TIMO  8			/* Dial timeout expired */
#define DIA_INTR  9			/* Dialing interrupted by user */
#define DIA_NRDY  10			/* Modem not ready */
#define DIA_ERR   20			/* Modem command error */
#define DIA_NOIN  21			/* Failure to initialize modem */
#define DIA_BUSY  22			/* Phone busy */
#define DIA_NOCA  23			/* No carrier */
#define DIA_NODT  24			/* No dialtone */
#define DIA_RING  25			/* Ring, incoming call */
#define DIA_NOAN  26			/* No answer */
#define DIA_DISC  27			/* Disconnected */
#define DIA_VOIC  28			/* Answered by voice */
#define DIA_NOAC  29			/* Access denied, forbidden call */
#define DIA_UERR  98			/* Unknown error */
#define DIA_UNSP  99		/* Unspecified failure detected by modem */

From Jamie Watson <[e-mail]>:

- On AIX 3.x, the xon/xoff and rts/cts values in the tty line discipline
  stack are saved on startup and restored on exit.  This fixes a very
  irritating problem in which C-Kermit could leave a tty in a state that
  was unusable by uucp or cu.

- The REDIRECT command is enabled and working on AIX 3.x and on DEC Ultrix.
  The AIX version has been tested on AIX 3.1.5 and AIX 3.2.x (heaven only
  knows what the "x" might be; I defy anyone to assign a specific value
  to any currently running AIX 3.2 system).
  The Ultrix version has been tested on a DECsystem 5000/25, Ultrix 4.3,
  and on a MicroVAX 2000, Ultrix 4.2.

ckutio.c 5A(116) 1 Nov 93.

Jamie also noticed that the new keyboard-input buffering invariably caused
CONNECT/TELNET to fail under AIX, as it would have under most any System V
implementation.  The culprit turns out to be that the read() in kbget() was
getting interrupted by the lower CONNECT fork as a result of TELNET echo
negotiations, thus read() would fail with EINTR.  Solution: ignore EINTR
errors when read()ing from keyboard during CONNECT.  ckucon.c 5A(056) 1 Nov 93.

Fixed prototype-mismatch complaints in new buffering code in SCRIPT and INPUT
commands.  ckuscr.c 5A(019), ckuus4.c, 1 Nov 93.

Added a compile-time symbol for ckufio.c: ISREGBUG.  If this symbol is
defined, it tells ckufio.c to define S_ISREG and S_ISDIR itself, rather than
using definitions from header files, when the latter don't work.  There is
already code in ckufio.c to do this for certain systems, but the new symbol
allows it to be done from the CC command line (or with KFLAGS on the "make"
command line) without modifying the source code.  This is necessary for at
least one of the NCR UNIX versions.

Added numeric C-Kermit program version number to OS/2 C-Kermit ENQ/Answerback
string so now it says "OS/2 C-Kermit 501190".  ckocon.c 5A(104), 2 Nov 93.

From CKUKER.BWR:
"If a file arrives that has the same name as a directory, and C-Kermit's FILE
COLLISION setting is BACKUP, C-Kermit will either: (a) rename the directory,
or (b) not be able to rename the directory and so will refuse the incoming
file with the message "Can't transform filename".  Similarly, if C-Kermit's
FILE COLLISION setting is OVERWRITE, C-Kermit will, hopefully (!), not be able
to delete the directory, and so will refuse the incoming file."

And from CKOKER.BWR:
"You can't give a command like "RECEIVE A:".  C-Kermit will fail to open
the output file.  You must also include a filename, e.g. "RECEIVE A:X.X"."

These are now fixed.  C-Kermit -- at least the UNIX and OS/2 versions -- now
has the ability to recognize disk and directory names when they are given as
output filespecs.  When this happens, primarily when "as-names" are given in
RECEIVE or GET commands, or in a '-a' command-line argument, C-Kermit now
changes its directory (or disk) to the one given for the duration of that
command, and then restores the previous current disk/directory when the
command is done.  So now in UNIX, you can say "receive foo", where foo is a
directory name (relative or absolute, need not end in slash), and C-Kermit
will CD to that directory and store all the incoming files there under their
own names until protocol mode exits, at which time C-Kermit goes back to the
directory that was current when the RECEIVE command was given.  If "foo" is
not a directory, then C-Kermit stores the (first) incoming file in the current
directory, under the name of "foo", as before.  This feature depends on the
symbol CK_TMPDIR, which is automatically defined in ckcker.h for UNIX and
OS/2, and a new supporting CK?FIO.C function, isdir(), which has been filled
in for UNIX and OS/2.  Also, files that arrive under names which correspond to
directory names are no longer able to cause renaming or other modification to
directory files, and useful error messages are now returned in this case.
Also, 'kermit -r' is bit more careful about checking whether the output file
can be created before going into protocol mode.  ckcker.h, ckcfns.c 5A(093),
ckuus[r56y].c 5A(108), ckufio.c 5A(075), 2 Nov 93.

(Ideally, the receiving system should be able to send back a fully qualified
file specification in the ACK to F packet, but most systems where C-Kermit
runs are not able to do this.  The temporary directory name might be absolute
or relative, might have or lack a terminating separator, etc, so we'd need a
ton of system-dependent code to do this right.)

Added the ability to repaint the fullscreen file transfer display screen, for
example if the screen gets messed up by talk or broadcast messages during file
transfer.  User types Ctrl-L (or L) to repaint.  Internally, two methods are
used.  If the symbol CK_WREFRESH is defined, which can be added to ckcdeb.h
for a particular version (as I have done for SunOS, NeXTSTEP, and AIX), or
added to a makefile entry, or given as "KFLAGS=-DCK_WREFRESH", or otherwise
defined at compile time, then the real-curses-library functions clearok() and
wrefresh() are called, which totally repaint the screen -- just what we want.
Otherwise, Kermit uses a do-it-yourself method which clears the screen and
repaints most of the fields, but not all of them, because certain information
might be lost by the time repainting occurs, such as the filename and the
direction of transfer.  The latter method works for all implementations of
fullscreen file transfer display, but making this method work for all fields
would be a major recoding task, and would add a lot of complexity and storage
space, basically duplicating the functionality of curses itself (keeping each
field, or an entire screen image, for repainting purposes).  ckcdeb.h,
ckuusx.c, ckuus5.c, 3 Nov 93.

Added CK_WREFRESH for AT&T System V R3 and later, since it's in the SVID.
Also for BSD 4.4 and derivatives.  However, this type of change is always
risky, so I also added a new symbol, NOWREFRESH, to allow this decision to be
overriden from the CC or 'make' command line.  In other words, if your linker
complains that _clearok or _wrefresh are undefined, then "touch ckuusx.c"
and rebuild with -DNOWREFRESH.  ckcdeb.h, 3 Nov 93.

Added VMS support for CK_WREFRESH, contributed by William Bader
<[e-mail]>.  ckcdeb.h, ckuusx.c, 4 Nov 93.

Added a makefile entry for DECstation 5000 models based on R4000 "MIPS-3"
chip, Ultrix 4.3A, from Jamie Watson <[e-mail]>.  4 Nov 93.

A few more minor Stratus VOS changes, having to do with X.25, from David Lane.