Changes for 3.30 since 3.24:

1.  An ANSI CUP (cursor position) with x or y values exceeding
screen size should set the cursor to the bottom or right of the
screen, not the top or left.  ECU did the latter; it is fixed.
The DSR (device status report) handler was broken also.  It is
fixed, and differentiation is now made between a type 5 and 6 DSR
request.  As a result of all this, the X11 resize program will work
with ECU.

2. The command line -l handler now treats an argument
with a leading equals sign as a Devices type rather than
an explicit line.  That is,

    ecu -l=VSAT

will search for an ``idle'' line matching any entry
in your HDB Devices file whose first field is "VSAT".
The term ``idle'' is strictly correct only on SCO machines
or where the utmp/ecuungetty code has been made to work.
The baud rate is automatically set to the highest speed
specified by the Devices entry, overriding any value set
by the -b command line switch.

Note that specifying a logical (directory entry name)
rather than an actual (numeric) phone number will cause the
directory entry choice of line to override the line specified on
by -l.  This has always been true of -lttyxx treatment as well.

Perhaps one day, I'll have time to make a directory entry
specifying "Any" *not* override -l and initiate it's own search for a
free line (based as it is on baud rate and type == "ACU" alone).
There is no time now.

3. The command line usage has been updated to show what
command line variables are overriden by a logical phone number.
Again, no time to put it in the manual.

4. The integer operators << and >> were added.

5. The procedure command eeod (erase to end of display) was added.

6.  Previously, you could bind a function key to a character
sequence to be transmited over the line.  You may now bind a
function key to a procedure invocation.  If the first token in
the third field of a key definition is "proc", the rest of the
third field is taken to be a procedure invocation.

Consider the following example entry in ~/.ecu/keys:

test
    F1:VT100 F1:esc O q
    F2:doit:proc doit arg1 arg2 argn
                 ^  64 chars max   ^
                 `-----------------'

The maximum length of the procedure name, arguments and
whitespace must be less than 64 characters; any excess is
silently discarded.

Procedure doit might, for example, send a prompt to a remote to
begin an XMODEM transaction, then start an XMODEM receive to
get the file.  (This helps make up for the unsuitability of XMODEM's
C or NAK header for automatic download startup.)

7. The fkey definition file (~/.ecu/keys in the documentation)
can be located in the library directory (usually /usr/local/lib/ecu).
This has always been in the code.  Now it is documented.

8. The code for saving a phone directory lacked good error detection.
This has been fixed.

9.  -D__STDC__=0 was added to gcc flags.

10.  The default for SCO UNIX configuration is now 'v' not 'u',
making 3.2v4/ODT 2.0 the default.

11. A discussion of how to "finish" a remote C-kermit server was
added to README.Q-and-A.

12. Under SunOS, DTR was not properly controlled.  This has been fixed.

13. Configuration for use with gcc now requires you have gcc 1.40
or later.  If you say you have gcc 2.3.3 or later, you will get -O6
compilations (full global optimaztion).  If you encounter problems
with the software compile with -O6, reconfigure answering no to
the 2.3.3 question and remake.  Please let me know of any problems.

14. -Wuninitialized was added to gcc compilations.
This is only used if you answered yes to the 2.3.3 question

15. ECU works again on XENIX 286 thanks to vancleef@netcom.com
(Henry van Cleef)

    a. CFLAGS from "-M2let128" to "-M2e -i"
    b. ZMODEM code unsigned Baudrate to unsigned long

(Not a bad port given I haven't seen a 286 in 5 years!)

16.  An experimental USR 2400 dialer (dceUSR24v.c/dialgUSR24v)
has been added to gendial for attended voice line autodialer
applications.  It is written for speed over reliability, so so
the careful initialization found in other gendial dialers is
absent.  This means dialgUSR24v works best with a modem that is
dedicated to the voice autodial function.

As delivered, dialgUSR24v will prefix a '1' to a number passed to it if
the number is ten digits long and has [01] as its second digit.
(I wanted to word click with xterm and my phone directory is an
ancient flat file with no '1' prefix to be found.)
You may need to adopt, hack and improve.

DialgUSR24v, like all gendial dialers, are supplied "as-is" in
hope the code may be useful. (Okay, and also to promote the
excellent SCO solution to modem madness.)

17. Previously, a Devices-specified  SCO dialer program had to
have a full pathname specified for ECU to use it (otherwise,
ECU would treat the token as the name of an entry in Dialers).
Now, when ECU sees a simple name in a Devices entry, it checks
for an executable in the UUCP library directory before searching
Dialers.

18. PgUp and PgDn have been added to the directory menu repertorie.

PgUp   ^U    page up
PgDn   ^D    page down

19.  The new conxout interactive command was missing from
interactive help.

20.  I had a hack cpio entry in the main Makefile for my use
in making gifts for ftp.uu.net:/tmp.  Folks have started
using it, so I beefed it up.  Now it makes a proper cpio
file even if your distribution directory is not named 'ecu'.

For instance, make cpio in directory my-ecu-3 will produce
a cpio file containing files named my-ecu-3/....

21.  The Configure/config.c SCO LNG353-related logic is obsolete or 
soon so.  It was removed.

22.  Added to README.Q-and-A a sad discussion of the state of
ecuungetty on non-SCO systems.

23. I got a guest account on a Consensys SVR4 system and
put the code through it's native compiler.  Y'all have been
either too nice about reporting warnings or else this compiler
is more righteous than others.  I fixed a few warnings, but
couldn't get far in the short time I had.

24.  The receiver process was not being killed when the ecu
internal dialer was used.  Did the thing actually work for
anyone?  I haven't used it for years (since HDB arrived on
XENIX in days of olde).

25.  Using cursor left or right in a ttygets bombed due
to a bad call to tputs() in each handler.  Now Home xx<cleft>
won't explode on you.

26. While ECU does not support the VT-100 terminal type, it
is close for most uses. ESC > (select numeric keyad)
and ESC = (select application keyad) are now ignored.
They used to hang ECU until the escape sequence collector
timed out.

27.  Previously, you had no way of knowing if an lgets procedure
command execution with a specified delimiter actually detected
the delimiter or not.  If you wait for a single carrige return
with anything but a ridiculous timeout interval, how are you to
know whether the read timed out or the return was received.

Well, now in addition to setting $i0:

    $i1 is zeroed upon execution of the lgets command
    $i1 then receives 1 if you specified a delimiter and
        it was read

28. A bug prevented /usr/local/lib/ecu/colors from being read
if a ~/.ecu/colors file was found missing.

29.  I admit it.  I've been using Windows 3.1.  Starting with
3.30, if you are shown a field on a curses screen for editing, it
is displayed in "standout" mode (usually reverse video).

If the first key you press is an editing key (e.g., backspace,
cursor left), the standout display is replaced by a normal one
and editing proceeds.

If the first key you press is not an editing key (e.g., 'a'), the
standout display is replaced by a normal one, the field is
erased, and your first character becomes the first character in
the field.

If you do not wish this functionality, you may define -DOLD_WINGETS.

Pressing ESCape while editing always the original field
unchanged.

29.  For special applications, -DTURNKEY.  With this switch in
effect, executing ecu with no arguments opens the default
line and places you in the dialing directory.

30. Make depend no longer includes system header files
(those included with <> instead of "").

31.  The ioctl to F_GETFL in lopen() has been buggy since ECU
0.1.  It depended upon random stack contents being right.  It had
`worked' for seven years until I called lopen a bit differently
in 3.30.  This kind of bug makes me a bit ill thinking about what
else I've left floating out there.

32.  Thanks to Carl Wuebker <clw@f.rose.hp.com>, we have an HP-UX
port.  Robert E.  Laughlin <bel@nosc.mil> helped with the
funckeymap entry for hpterm and with testing the port.

33.  Thanks to Robert_Broughton@mindlink.bc.ca, a Linux port
lives.  See README.Linux.

34.  Daniel Harris <daniel@reubio.apana.org.au> in working on a
BSD port found a long-present bug: in termecu, shm_done() was
being called before restore_initial_colors() was being called;
the catch is that the latter refers to the shm segment.

35. More Linux patches from Bob Broughton seem to flesh out the
Linux port.

36.  The 8 retries at locking a port in the setup screen has
been extended to 20.  Bob Lewis found that was better for slow
turnaround on some getty/initcond situations.

37. Bob Laughlin <bel@nosc.mil> sent an kbdtest3.out file for
hpterm.  He says it is preliminary and that he'll look into why
the cursor keys are not being read properly.

38. hdb_choose_Any could not handle utmp status DIALOUT.
Go figure.

39. Tim Sailer found a long standing, rarely tripped bug.
If the local console had no "mb"/"me" blink attribute
in it's termcap (and only SCO systems do), then the ECU
receiver would coredump if the remote system sent a blink
escape sequence.  He found it because of modem garbage on
a bad connect.  It is hard to believe no one has had a
blink sequence thrown at their ECU session before now.

40. A configuration item for the internal dialer timeout
was added (INTDIAL_TO).

41. USE_FIONREAD replaces NO_RDCHK. USE_SELECT replaces
WORKING_SELECT.  USE_DECIMAL_PIDS replaces/inverts HONEYDANBER.
USE_GETIMEOFDAY enables compilation of an ftime() substitute
using gettimeofday().

42. A Motorola Delta 88k port is compilable.  Who knows if it works?

43. Daniel Harris ported the program to NetBSD.  More work to follow.

44. No more references to obsolete protos.h and USE_PROTOS.

45. BETA 3.28.00 released to a few.

46. Fixed Linux /bin/bash wildcard expansion.

47. Robert Lipe fixed several gcc warnings which make the way
for 64-bit machines (overflow in integer constants, primarily).

48. When the highest level procedure exits, ecu is supposed
to close all files opened with `fopen'.  Now it does.

49. WHT ported it to Motorola SVR41.

50. Added models/ecu-ansi.tinfo and models/ecu-ansi.tcap.

51. cfsetispeed()/cfsetospeed() is not used in lieu of
cfsetspeed() in all termios use.  cfsetspeed() appears to be
a rare beast while the i/o pair seems ubiquitous.

52. gcc 2.5.8 disallows `const' if -traditional, so -Dconst=
for all gcc compilations.  I'll never code one of the pinky-extended
striped-pants things myself.

53. Finish port to ODT3 DS.

54. config.c scans /usr/include/sys/{types,select}.h for fd_set.
Code uses it for select() in lieu of old BSD &int.

55. config.c scans /usr/include/sys/{time,select}.h for struct timeval.
Some systems define it in select.h for hysterical reasons.

Changes for 3.30 patch 1:

56.  Fixed bug in winget_single whereby End, etc.  could not be
entered in a curses single character input field.  Robert
Broughton <Robert_Broughton@mindlink.bc.ca> reported this three
time during alpha and beta testing, but I never got it into the
code.

57.  A long standing bug with funckeymap support has been fixed.
ASCII special characters with two character representations (such
as `gs') would be rejected by ECU.  This was pointed
out by David A.  Avery <daa@CERF.NET>

Changes for 3.30 patch 3:

58. The termio (vs. termios) version of disp_termio() was
displayed incorrectly.

59. Once SCO CRTSFL flow control was turned on with a value of 4
or 7, it could not be turned off with an argument < 4.
In addition, using an argument greater than 4 caused *input* flags
to be trashed instead of control flags being manipulated.

60. Steve Feuerstein <sgf@acsu.buffalo.edu> reported that the
procedure command `fopen' returns an error on a successful open.
Indeed, I found it would return normal success on a failed open.
This has been fixed.

61. A problem with use of select() under Linux was fixed.
Thanks for that goes to Bob Broughton.

Changes for 3.30 patch 3:

62. yzrnur!rene (Rene Lampe) reported a DTR flash 
fix for ESIX SVR4. Now, `hangup' works.

63. Screen size defines SCREEN_LINES_MAX and SCREEN_COLS_MAX
have been renamed  CFG_ScreenLinesMax and CFG_ScreenColsMax.
Configure now asks for max screen lines and columns. Experimental
but confidence is high.

64. Andrey Chernov provided patches for eight-bit clean input and
for porting to FreeBSD.

65. I reworked a great deal of conditional code to get rid
of some "porting rot" -- hope I didn't break anything.

66. The file transfer raw tx and rx data counts and other stats 
were not updated on XMODEM or unspecified length ZMODEM xfers
(undefined file length). Fixed by ache@astral.msk.su.

67. Again thanks to Andrey, upon file receive timeouts, the stats
mentioned in 66 above are updated. Particularly, RTS/CTS and
driver queue information is useful to see in these circumstances.
(Not all ports of ECU report driver level statistics and EIA status;
neither do many tty drivers have means of reporting this information.)

Changes for 3.30 patch 4:

68. On SunOS and possibly other termios users the send-break function
hung the transmitter process. The break is signalled, but control is not
returned. A fix for SunOS and Solaris was applied.

69. Commands 'erto' and 'erverbose' have been added to the
interactive and procedure command repertoire. Below is the interactive
help information.  The procedure commands function in the same
manner, but require arguments.


    ERTo
    Usage: erto [<msec>]

    This command specifies the default timeout period (in milliseconds)
    of the next "expect" in an ECU-executed expect-respond script.
    Remember that ECU expect-respond scripts can change timeouts on their
    own, so this number may not "stick." Entering the command with no
    arguments displays the current value (which, again, may not be the
    value ECU started with or the value you last selected).

    ERVerbose
    Usage: erverbose [<true-false>]

    This command specifies the verbosity level of an ECU-executed
    expect-respond script.  Entering the command with no argument
    displays the current status of verbosity. To change the value, enter
    and argument beginning with an lower-case 'y' or 'n'.

    Examples:
        erv
        erv y

There is some clash between the erverbose commands and the -v
flag of the procedure expresp command. Complaints will get
an immediate fix, but otherwise, the fix is in the stack.

70. Fixed many typos in the manuals (thanks to Bob Lewis
at SCO for the time he spent editing!).

71. Cleaned up support for variable max lines and columns.

72. GREATLY simplified timed line reads (lgets_timeout and lgetc_timeout)
may be used by defining SIMPLIFY in eculine.c. This reduces
timed reads to ASR-33 speeds (10 char/sec), but if you have trouble
during a port, here is a way out of swimming in molasses.

73. Made many Chernov FreeBSD patches standard for all versions
(eight-bit cleanness).

74.  On systems supporting ecuungetty, a choice is now provided
by Configure as to how to handle lines which do not appear in
your UUCP Devices file.  Lines in Devices (generally) are
involved in shared dial-in/dial-out or UUCP services; lines which
are are not in Devices are not involved in such use.  

If you answer yes to the new Configure question 'Do you want
ecuungetty to manage only lines in Devices?' then ECU will manage
the ownership and permissions only of lines in Devices.  This
means ECU will not `monkey' with the ownership or permissions of
serial lines NOT in Devices (which of course implies permissions
must be already be set correctly for you to be able to access the
chosen line).

75.  The undocumented interactive command hack 'sgr' had a minor
bug preventing the "Stimulus:" display on the console.
Ecu.scr.log was always updated properly.  I needed to beef it up
anyway and added `sgrto1' and `sgrto2' while I was at it.

Changes for 3.37

76. Added `con80x25' and friends to funckeymap for later Linux.

77. Non-vt console attribute mapping under Linux improved.

78. Later gcc does not permit "volatile" keyword with -traditional.
I quit using "volatile".

79. My apologies to Andrey Chernov for mispelling his name Andrew
for five years. Rather than generate the 12kb patches to fix the
numerous occurrences, I'll try to remember and make the fix in
the next full release.

80.  BSD 4.4 defines st_size as "long long" instead of long, so printf %ld
fails. Andrey supplied a patch.

81. lstat() is no longer provided in lstat.c.  It was only there
for folks with 3.2v2 DS that were too cheap to buy the 3.2v4 DS when 
they moved to 3.2v4 OS.  It causes problems with all post-3.2v4
and post-ODT 2 SCO systems.

82. If ecu is chown uucp/chmod u+s, then uucp owned lines can be
directly opened on any ecu platform (without ecuungetty on systems
which normally require it).

83. The maximum screen geometry is now shown by the stat
interactive command.

84. An early ALPHA version bug with configuring lines and columns
has been fixed.

85. For SCO, -nointl has been removed.

86. Support for SCO 3.2v5/Open Server 5 has been added.

87. A hack for SCO UNIX cross-compile of XENIX binaries was removed.

88. Support for automatically installing ECU setuid UUCP was added.
Such is automatically done for OS5.

89. There are KNOWN PROBLEMS WITH USING ECU WITH SCOTERM
which there was no time to diagnose:

> From: emory!raider.raider.net!root (Mark J. Bailey [HOME])
> X-Mailer: SCO System V Mail (version 3.2)
> To: n4hgf!wht@uunet.uu.net
> Subject: ecu and scoterm
> Date: Wed, 21 Jun 95 10:36:48 CDT
> Status: O
> 
> hi warren
> 
> still have ecu freeze up TIGHT any scoterm I try to run it in the second it
> (ecu) closes down.  can't even click the scoterm away.  i iconized it and it
> did die off after 30 minutes or so.  weird.  ideas?  i am sco open server 
> enterprise 3.0 - compiled ecu 3.342 with cc
> 
> mark

Changes FOR 3.37

89a. I cannot find this problem with OS5.  I had no way to
test under earlier versions because I have not had a monitor
or keyboard on my ODT3 box for over a year :).

90. Configure now take a cue from the user's local/bin directory
for it's default for local/lib/ecu directory.

91. The rtscts command value may now be set before a local line is attached.

92. The dialing directory supports setting the rtscts value.

 | description       don't call this_________________________ |
 | debug level       0  (dialer -x value 0-9)                 |
 | DCD watch         n                                        |
 | RTS/CTS flow ctl  n  (0=off,7=best,n=no change)            |
 |                                                            |

93. Wonder of wonders, ecu.man has been updated for changes 91-92.

94. make cpiogz does the same job as make cpio, but creates
a gzip file instead of a compressed file.

95. SCO 32v5 converts ALL characters of a tty name to lower case
for a lock file name.  This is now supported by ECU for 32v4 and beyond.

96. Examination shows the Linux version does not support
DCD watcher set to '1' (catch connection terminations, but
continue ECU with communications I/O intact).

97. ANSI SU (scroll up) did not work except for native
SCO multiscreens.  

98. An updated ECU/ANSI terminfo file ecu-ansi.tinfo is
supplied in the models/ subdirectory.

99. Drop pretense of XENIX 286 support.

100. Drop SEAlink support.

101. Added 57600 and 115200 baud support for Linux.

102.  Finally an unlazy moment after 6 years of wishing.  All(?)
references to "baud rate" have been replaced with "baud" or "bit
rate." "Baud rate" is a redundant and meaningless term.  The
usage has become acceptable like saying "It's me" over a
telephone.  I want to turn the tide (heh heh, likely story).

103. select() now used for all timing functions. nap() decommitted.

104. Serial-line oriented commands moved from eculine.c to ldserial.c.

105. Now funckeymap entry names (like vt100, sco, etc.) may have comments
after them.

106. Minimum geometry is 80x24 now. No one has complained that
there are still terminals out there with only 24 lines??

107.  FOR MADMEN ONLY:
Added a very preliminary but stable telnet option.  From command
line only, an internet connection may be established by
specifying a logical "telephone number" with a period in it.  For
instance,

    ecu blah.org
    ecu 127.0.0.1
    ecu kepler.

The last example shows how to call a host with a simple name like
"kepler": suffix the simple name with a period.

To enable this experimental feature to systems
other than Linux or SunOS, you must add (or add to)
a file in the ecu distribution directory named 'config.local':

The file must contain the line:
LOCAL_CFLAGS = -DCFG_TelnetOption

On SCO or similar, you must add the line
LOCAL_LIBS = -lsocket

The -z switch enables display of telnet option negotiation.  The
interactive and procedure 'telopt' commands also control the
display, but the show is mostly over by the time you get a login
prompt, so consider using-z or putting 'telopt yes' in an _rc.ep
file.

108. %telnet is a procedure integer function which returns zero
if the current connection is a serial connection, 1 if an
internet connection.

109. mk.cpio now uses /bin/pwd rather than a simple pwd. Linux
does not behave as we would like.

110. The rcvr ANSI filter has grown up.

     a. NO_ANSI_EMULATION hack for BSD now fully implemented
        as CFG_NoAnsiEmulation.  Turn this on in config.local
        if you want to break a nice toy.  No screen image
        support.

     b. If included, it defaults to on.

     c. controllable by interactive and procedure `ansif' commands;
        `ansif' may be used in _rc.ep

     d. startup switch -n disables the ANSI filter.

111. All serial line-oriented configuration commands may now
be used in _rc.ep.  Previously, setting rtscts or the like
would fail because no line was attached.

112. F12 is now the Shift-BkTab (redraw screen from ANSI filter's POV)
key for Linux's main console. The console keyboard driver sends
0x09 for both Tab and Shift-Tab.

113. make targz and mk.targz added to make standard Linux tar.gz files.

114. README.Linux updated and ecu-3.37.tar.lsm created for tsx-11 release.

Changes FOR 3.37.90

115. ZMODEM file transmit working! Rx is next!

116. ZMODEM file receive works. What was needed was an sz that
would properly escape IAC characters.  See README.telnet.

********************************************************************
* tsx-11.mit.edu:/pub/linux/sources/Communications/ecu-3.37.tar.gz *
* was released at this point                                       *
********************************************************************

Changes for ECU 3.38.02

117. Setuid() macro expands to seteuid on systems which have it,
setuid on others. This enables setuid uucp to work on SunOS and the like.

118. Shell wild card characters may now be used in specifying
the interactive log command file name character.

119. The procedure function %uname now returns the node name
(utsname.nodename) rather than the system name (utsname.sysname).
A function %sysname was added to return utsname.sysname.

Changes for ECU 3.38.03

120.  Fix hanging built-in dialer if tty not stdin, as when using
ecu from cron job.  Thanks to Jim Melton <jmelton@i-link.net> for
reporting the bug.

Changes for ECU 3.38.04

121. Bob Lewis <robertle@sco.com> suggested tty1A is a better default
than tty1a for SCO. config.c was changed to comply.

Changes for 3.40

122. Andrey Chernov supplied more configurations fixes for FreeBSD.

Changes for 3.41

123. Keep uninitialized ecutcap code from trying to send
console control sequences before being initialized.
If your session hung after saying,
Terminal type 'xxx' does not have the beef.
Try again with a screen-oriented terminal.

124. Make sure insert/delete line/character is present.
Warn if not.

125. Add al,ac,dl,dc to models/ecu-ansi.{tinfo,termcap}

Changes for 3.42

126. Fixed erroneous reference to `ac' instead of `ic' for TERMCAP
insert character identifier. 

127. models/ecu-ansi.tcap had `ac' instead of `ic'.

128. Better diagnostics for termcap/terminfo shortcomings.
    Warning: Terminal type '%s' has
      No standout/standend (`so')
      No insert line (`al' or `il1')
      No insert character (`ic' or `ich1')
      No delete line (`dl' or `dl1')
      No delete character (`dc' or `dch1')
    The first `code' is the TERMCAP identifier for the feature
    while the second `code' is the TERMINFO identifier.
    We will continue, your screen may be inaccurately painted.
    Press RETURN to continue or ESCape to exit:

129. At certain times, ECU is supposed to be killable with a
keyboard interrupt.  Now it is, though the feature was broken.

Changes for 3.43

130. Better diagnostics in error report for bad screen geometry.

Changes for 3.44

131. Rearrange curses use to minimize version differences (now
watch it break everywhere I cannot run it! :>).
3.40 on would not compile on Solaris.  I had problems
with whline and wvline, so I took the opportunity to
revise all curses use, collecting various conditionals
into universal (HA!) code.

132. Apply Andrey Chernov <ache@nagual.ru> patches for FreeBSD:
a. improve console color handling
b. use getwd() to get current directory.

133. Apply Don Yuniskis <dgy@rtd.com> patches for FreeBSD:
a. use /etc/uucp for UUCP library directory
b. fix serious problems with more esoteric features
   of configuring/handling internal dialer

Changes for 3.45

134. Don Yuniskis <dgy@rtd.com> work turned out to be
on an older version of FreeBSD.  He gets no less thanks
for having taken the time to submit.  FreeBSD uses Taylor
UUCP nowadays.  More porting work is needed before ECU
can use its `Devices'-like logic under FreeBSD.

Changes for 3.46

135. Andrey Chernov <ache@nagual.ru> sent me a set of
patches he describes as:
1) 8bit/ctype fixes.
2) locale tuning.
 
Few words about 1). According to ANSI and POSIX standard valid
ctype macros range is EOF,0..UCHAR_MAX. In most implementations
it means -1,0..255. So, you can use is.*(char) only if you are
absolutely shure that "char" here is ASCII char. Small example:
char c = `\200'; for signed char compilers expands to -128.
It means that isalhpa(c) becomes isalpha(-128) which is illegal
per ANSI/POSIX range definition. Most common way to fight
it is (unsigned char) cast for all char* strings, since char
can't become EOF. I.e. isalpha((unsigned char)-128) is legal,
because it is equal to isalpha(128) and 0 < 128 < 255. Don't use
this cast for "int" arguments, f.e. getchar() result,
because you truncate EOF (-1) to 255 (valid char) in this way.

136. I got rid of attempts to use AT characters in window
box drawing.  There are too many ports around now and
curses grovelling started taking more time to maintain
than everything else.

137. Screen dumps now dump the virtual screen map
in all ports. SCO console capable of reporting their
hard display contents used to be the source of
screen dumps.

138. Command line switch -C and interactive command 'rev'
shows compile-time configuration.

139. MOTSVR4 excited some kind of Motorola bug. After
exiting a child csh process, ECU (and /bin/vi!) get a SIGTTOU which
I have hacked into submission by using signal(SIGTTOU,SIG_DFL).
Like vi, ecu prints "Stopped (tty output)" after HOME !csh.
Just type `fg'.

Changes for 3.47

140. Now we use cua0, not ttyS0 for linux for the default port.
ttyS0 is for incoming connections.

141. Clean up MANY unfortunate gcc "comparison between signed and unsigned"
warnings.

Changes for 3.48

142. Consolidation/cleanup of chtype/unsigned char and other curses
incompatibilities.  Using CFG_UseSetupterm, CFG_UseACS _locally_
for now, globally (config.c, Makefile) in the future.

143. MAKING AND INSTALLING portion of README moved to INSTALL.

Changes for 3.49

144. Uh, I forgot one relatively important note for 3.48: normalized
telnet behavior with async so that one can arbitrarily switch between
them during one ECU session.

145. Fixed bad 'else' procedure command that improperly execute
a command appearing on the same line as the else. i.e,

   ifnz 1 echo 'good'
   else echo 'bad'

used to print good and bad, whereas

   ifnz 1 echo 'good'
   else
        echo 'bad'

would only print good.  Now both codings work as expected.

146. Added unsupported stuff.

Changes for 3.50

147. Tildes are now converted to hyphens when saving phone directories.
Why one might have wished to used a tilde in a directory entry
is past my wisdom: my apolgies.  Now, colons in fields
(like "localhost:25") are protected from the passwd-style
database by converting them to colons.  When the entries
are reread from the directory files, tildes are restored to
colons.

A nasty side-effect: those who edit phone directories directly
have to remember to type localhost~25, not localhost:25.

148. Created ecuicmd.1 interactive command man page.

149. Auto-configuring for telnet option for
SCO (and potentially for other SVR3+-lsocket platforms).

150. A carriage return was missing from the redial "no line attached"
message.

151. Merged ecuicmd.1 plus two hours of drudgery into ecu.1.

152. Lengthened TTY field in setup and phone directory.
Motorola SVR4 port names can get loong. E.g., /dev/port/m332_c0d0.

153. ECU RCVR now handles a dying socket: it signals XMTR to do 
funeral and probate.

154. break/lbreak in telnet sends IP.

155. Added qmake to make my life easier.

Changes for 3.54

156. Added whilez/whilenz procedures commands.

157. Added auto-search for sys/filio.h.

158. Added SIGINT interrupt and 20 second alarm() timeout
for connect().

159. Added fork and kill procedure commands.

Changes for 3.55

160. Support && and || integer operators in procs.

Changes for 3.56

161. Add %dates, %timesm and %timesmz.

162. Bob Lewis <robertle@sco.com> verified ECU
runs on SCO UnixWare 2.1.

4.00 RELEASED

Changes for 4.01

163. Add fflush proc command.

164. Fix fputc/fputs procedure command documentation.

165. Implement ++,-- operators.

Changes for 4.02

166. Fix bug in ecuicmd.c where non-telnet build gets a
syntax error.  Reported by Robert M. Bailin <72027.3605@CompuServe.COM>.

Changes for 4.03

167.  Manuals (and other, obscure, files) were omitted from the
4.00 release.  Oops.

168. Catch qmake at 'make clean' time.

169. Add modem dialer procedures.
    "Modem Dialer Procedures (dialproc)"

    You may use a procedure to do the modem dialing function.
    Just create a subdirectory called "dialproc" in your personal
    .ecu directory or in /usr/local/lib/ecu.  In that directory,
    place a procedure whose name matches the port's base name.
    That is for "/dev/ttys0", create dialproc/ttys0.ep".  For
    "/dev/port/m332_c0d3", create "dialproc/m332_c0d3.ep".

    The procedure is passed one argument, the telephone number to
    be dialed.  It is the responsibility of the dialproc merely
    to establish a connection.  Procedure status is returned in
    $i0 and $s0.

     $i0 values
     0    connected
     1    did not connect
     2    interrupt attempts (unrecoverable)
     3    modem error (unrecoverable)
     4    try ecu dialer
     
     $s0 is returned set to the modem result string, such as "CONNECT 9600"
     or "BUSY".


Changes for 4.04

170. Fixed procedure commands sx and sy due to ECU not passing ecusz
a baudrate.  PLEASE, will the fellow who reported this and whose mail
I lost please step forward to take credit for this fix??

Changes for 4.05

171. Inconsequential bug in telnet_open never caused any harm;
getservbyname call had wrong protocol. getservbyname("telnet","telnet")
now corrected to getservbyname("telnet","tcp");

172. Andrey Chernov <ache@nagual.ru> or <ache@null.net> wrote that
> BTW, here small patch to fix 'date' POSIX-ly. Previous one work only
> for subset of systems. Latest POSIX test suite requires direct set to 'C'
So the ecu makedate is now made with 
echo 'char *makedate="@(#)'ecu `LC_TIME=C date` $(SYSTEM)'";'  >makedate.c

173.  A BIGGIE: Bob Lewis <rll@felton.felton.ca.us> fixed the
code for SCO 3.2v2, ODT, OpenServer 5.0.x and UnixWare so that
actual box ruling characters are displayed on the computer
console across the whole SCO product line.  I had let this lapse.
Thanks, Bob, for restoring/extending/hardening the feature!

Changes for 4.06

174. A bug in winget_single() made single character form input
misbehave if a function key delimiter (ie, <END>) pressed.
The bug has been in there since 3.46 (August 1996!) so GO FIGURE!!
Thanks to Jeff Liebermann <jeffl@comix.santa-cruz.ca.us>
and Bob Lewis <robertle@sco.com> for the report and diagnostic help.

175. The setup screen showed 'E' instead of 'H' for half duplex.
(Part of jeffl and robertle work).

Changes for 4.07

176. Updated models/iarc.ep, my killer-app procedure; it is
probably operationally useless to you unless you use a Motorola
UTA220 ISDN terminal adapter to access iarc.net's ISP.
However, the style and usage of the procedure language in iarc.ep
is the best effort yet I have put into .ep hackery yet.

177. A bug in expresp.c qallocated one too few bytes for a string.
Some mallocs and string lengths worked, some not, because of slop that
may have been part of the allocated memory. Mea culpa.
