Queries

Write information about teletype to the standard output stream.

Special Control Characters

Must be followed by either the empty string, meaning 0x00, a single byte, used verbatim, "^-" or "undef", disabling the character, "^?" for DEL (0x7F, bottom 7 bits set), "^b" where b is any byte, of which the bottom 5 bits are used (this most likely corresponds to holding the Control key on your keyboard, then pressing the desired character, and yields distinct favourable results for @, the alphabet, brackets, \, ^, and _), or an integer.

These are also part of the Special Control Characters, but aren't actually characters, and control non-canonical input; as such, they take an integer only:

Input Attributes

Except where specified otherwise, these all correspond to flags of the same name, but in upper-case. A BREAK means being on the receiving end of a tcsendbreak(3).

If -ignpar parmrk, framing/parity errors yield bytes 0xFF, 0, then the erroneous one; to disambiguate, normal 0xFF bytes are doubled unless istrip in this case. If -ignpar -parmrk — a single 0 byte.

If -ignbrk -parmrk, a break condition yields a single 0 byte, if -ignbrk parmrk — 0xFF, 0, 0.

Output Attributes

These allow correct output to quirky or moving teletypes.

Control Attributes

If cmspar -parodd: space (always-0) parity; if cmspar parodd: mark (always-1) parity.

Local Attributes

Blocking SIGTTOU has the same effect as setting -tostop for the process would.

On systems where xcase does anything, a backslash followed by a lower-case byte is converted to an uppercase byte, and a backslash followed by any of '!^()\ becomes `|~{}\, respectively. Output is written with the inverse mapping (~ becomes \`, &c.). This is useful exclusively if you're calling from an extraordinarily cut-rate upper-case-only or punctuation-lacking teletype.

When erasing (icanon) (via erase, werase, kill with echok echoke, &c.), if echo:

On systems where flusho does anything, entering discard enables it until any subsequent input.

Line Speed

Where speed is any of:

all of which correspond directly to the gross-bits-per-second speed of the link, or one of the following:

Under Linux, 38400 potentially means 57600, 115200, 230400, 460800, or some division of the clock, depending on additional link-specific configuration (ioctl(TIOCSSERIAL), setserial(8) spd_hi/spd_vhi/spd_shi/spd_warp/spd_cust divisor),

Windowing

These take a single integer; the "window" is the systems's concept of the window dimensions — these won't resize an emulator, but can help get full-size windows and user-space wrapping on a remote system.

Line Discipline

Takes a single integer as "line discipline", which determines how bytes are packaged for delivery to/from the peer; in many ways this could be considered the "protocol" — "raw serial", X.25, GSM, PPP, &c. There should be no reason to set this manually.

Flushing

Combos

Aliases

Gang Aliases

Saved Attributes

Input Attributes, Output Attributes, Control Attributes, Local Attributes, then all Special Control Characters, hexadecimal, colon-separated.

Naturally, this doesn't preserve/restore Windowing information or the Line discipline.

Standard Attributes

Queries

size, as described; exactly one space.

Special Control Characters

intr, quit, erase, kill, eof, eol, start, stop, susp, with arguments as-described (except for empty-string-as-0, an extension), but ^b is defined explicitly for the alphabet, brackets, \, ^, and _.

min, time, with just an integer, as-described.

VMIN is allowed to be VEOF, and VTIME VEOL, which is why raw and cooked set them, respectively.

Input Attributes

All as-described.

Output Attributes

opost and -opost are part of the base spec, and as-described. All others are shaded XSI:

Control Attributes

All as-described.

Local Attributes

All a superset of described.

Line Speed

Either "number" for "terminal baud rate" or "ispeed number"/ "ospeed number" for "terminal input baud rate"/"terminal output baud rate", with 0 semantics as described. The speeds supported are at least:

It's unclear what the minimum required mapping to the numbers for these is.

Windowing

rows and cols, as described.

See also the standard LINES and COLUMNS environment variables in environ(7).

Line Discipline

Not part of the standard.

Flushing

The standard doesn't require stty to use tcsetattr(3), and doesn't specify any flushing characteristics.

Combos

Aliases

All as-described.

Gang Aliases

raw (and -raw, cooked) is shaded XSI, and supposed to enable (or disable for the inverse) "raw input and output". This raw mode is supposed to be equivalent to "cs8 erase ^- kill ^- intr ^- quit ^- eof ^- eol ^- -post -inpck"; what cooked is supposed to mean is unclear. This implementation's aliases match the GNU system's.

The additional -onlcr for nl and onlcr -ocrnl -onlret for -nl match the GNU system, and, since the additional attributes are shaded XSI while [-]nl isn't, may be considered a safe extension.

Obsolescent Aliases

Version 3 of the Single UNIX Specification (“SUSv3”) removes the following, previously marked LEGACY and shaded EX (equivalent to present-day XSI).

(Nominally, IEEE Std 1003.1-202x (“POSIX.1”), Draft 2.1 doesn't contain -xcase, but this appears to be a mistake in the CHANGE HISTORY.)

UNIX Programmer's Manual

gtty(II) and stty(II) syscalls appear in the first edition of the UNIX Programmer's Manual, operating on the DC11 asynchronous communications interfaces attached (UNIX Programmer's Manual, /dev/tty0 ... tty5 (IV)) , but not the console (ibid., /dev/tty (IV), p. 161) Teletype Model 33 ASR typewriter which is an integral part of the PDP-11.

stty(II) (ibid., p. 135) takes a file descriptor and a pointer to three 16-bit integers, the first of which (dcrsr) "is placed into the typewriter's reader control and status register", and the second (dcpsr) "is placed in the printer control and status register" — it then defers to the DC11 manual for their format and "until the typewriter is quiescent" (drain, TCSADRAIN); the BUGS section notes:

Indeed, the DC11's two registers are the Receiver Status Register (RCSR), laid out as such (in short; read-only bits and bit 8 omitted), compared to the present-day equivalents:

The SPEED SELECT fields select between the four speeds available on the specific DC11 model — according to /etc/init (VII) (ibid., p. 202) and logging in and logging out(VII) (ibid., p. 208) the default teletype is a 150-baud Model 37 Teletype with a BREAK at the "login:" prompt selecting the 300-baud GE TermiNet 300; type(I) (ibid., p. 101) supports printing to 134.5-baud IBM 2741s — the remaining speed required to disambiguate the model used is 1200 baud, given in /dev/tty0 ... tty5 (IV), points squarely at the DC11-AG.

These are described in grueling detail in DC11 asynchronous line interface manual, DEC-11-HDCD-D, 5th Printing, http://bitsavers.org/pdf/dec/unibus/DEC-11-HDCD-D_DC11_Asynchronous_Line_Interface_Sep73.pdf, Chapter 3 (Operational Programming), Digital Equipment Corporation, Maynard, Massachusetts, September 1973.

The third integer is the mode, whose contents "determine the system's treatment of the typewriter" (octal):

   200  even (M37 tty) parity allowed
   100  odd (non-M37 tty) allowed
   040  raw mode: wake up on all characters
   020  map CR into LF; echo LF or CR as CR-LF
   010  don't echo (half duplex)
   004  map upper case to lower case on input (M33 TTY)

Raw mode is the moral equivalent of present-day, well, raw, but spelled as the inverse of the default icanon (with min 1 time 0) implied: "every character is passed back immediately to the program. No erase or kill processing is done; the end-of-file character (EOT), the interrupt character (DELETE) and the quit character (FS) are not treated specially."; today, erase, kill, and eof are predicated on icanon directly, and intr and quit on isig.

020 is equivalent to icrnl onlcr, "used for GE TermiNet 300's".
010 can be recognised as -echo and 004 — iuclc.

The delays are hard-coded for the Model 37. It's unclear if they're the same at 300 baud or if they're altered for the TermiNet, except that it (supposedly) works.

The Special Control Characters are unchangeable:

intr and quit represent a primitive form of signals (even before signals got signals, job control is king. it is known): "for practical purposes at random" a process whose control typewriter — the first /dev/tty* file opened by the process, inherited on fork(II) (cf. today's similar but much more complex credentials(7)); this means all processes in a login session share the same one — matches the one on which the intr (quit) was received and which hasn't ignored the "signal" by passing 0 to intr(II) (UNIX Programmer's Manual, p. 122) (quit(II) (ibid., p. 127))), is "forced to exit" (and dump core). (If a funciton is passed to intr(II) (quit(II)) it is executed, but this causes a cascade of errors that make it not worthwhile, and, as the BUGS proclaim: "I don't know how to make it work.".)

Hanging up the call makes all future reads return empty (the Manual calls this simulating end-of-file); nowadays, a SIGHUP is also delivered to the foreground process group.

gtty(II) (ibid., p. 120), expectedly, copies those three integers into the provided buffer.

/dev/tty (IV) (the "hard-wired" "console typewriter") is nominally out-of-scope since [gs]tty(II) don't apply to it, but it explicitly cannot be placed in raw mode (it's unclear how that would happen), quit is generated with the "alt mode" key, it's always in iuclc (004) mode, and "a carriage return is echoed when a line-feed is typed" — this doesn't appear to be a documented stty(II)able mode, and doesn't appear to have survived as a modern one either.

type(I), or

Of interest is also the hup(I) (ibid., p. 63) program, which "hangs up the phone on the typewriter which uses it", much like stty 0 would, but, per BUGS, "should not be used; sometimes causes the typewriter channel to be lost".

UNIX Programmer's Manual, 2nd Edition

The argument names are aligned with the DC11 manual in K. Thompson and D. M. Ritchie, UNIX Programmer's Manual, Second Edition, https://www.tuhs.org/Archive/Distributions/Research/Dennis_v2/v2man.pdf, pp. 166-167, Bell Telephone Laboratories, Inc., June 12, 1972, stty(II). dcrsr becomes "the typewriter's receiver control and status register" and dcpsr — "transmitter control and status register".

The [-]parodd and iuclc equivalents are clarified:

   200  even parity allowed on input (e. g. for m37s)
   100  odd parity allowed on input
   004  map upper case to lower on input (e. g. M33)

   010  echo (full duplex)

   020  map CR into LF; echo LF or CR as LF-CR

   002  echo and print tabs as spaces
   001  inhibit all function delays (e. g. CRTs)

The reasoning for 020 (icrnl onlcr) is expanded past the TermiNets, to "other teminals without the new-line function" — i.e. ones for which a newline doesn't return the carriage.

The top byte of the mode is "used to indicate that the terminal is an IBM 2741 and to specify 2741 modes.":

   400  terminal is an IBM 2741
  1000  the 2741 has the transmit interrupt feature
        (currently ignored)
  2000  use correspondence code conversion on output
  4000  use correspondence code conversion on input
        (currently ignored)

The counterpart to the "correspondence code" is the default "EBCDIC (e. g. 963 ball and corresponding keyboard)" — since the 2741s are IBM Selectric typewriters, these refer to the two common type balls: the "correspondence" ball (and keyboard) are those of the standard retail Selectric for office correspondence. It's unclear what EBCDIC actually means here, since the counterpart to correspondence is PTT/[E]BCD, neither of which even remotely correspond to a character set encoding¹.

With this, type(I) (ibid., pp. 211-214) becomes

tty0(IV) (ibid., pp. 211-214), now with eight devices, now explicitly notes the input buffer of 150 characters and echoes a # for each one lost. The newline, tab, and form-feed delays are appropriate for the Model 37, but the carriage return delay is appropriate for the GE TermiNet 300. The speeds now listed are also 110 and 600, which implies heterogeneity in the devices used since any given DC11 is capable of being on-line programmable for one set of four speeds, but this is excluded by the SPEED SELECT fields being hard-coded for autodetection.

ASCII EOT, which causes the Model 37 to hang up, is stripped from the output (previously it was sent with wrong parity). No mention is made of formerly-similar special processing for NAK, which would lock the keyboard.

Arrangements for the 2741 (and other supported teletypes) are made automatically with the new getty(VII) (ibid., pp. 251-252) as cs7 -cstopb -echo 134.5. echo is off because 2741 is half-duplex (transmission-wise; all characters typed are always hardware-echoed, much like on a typewriter, which it is), and incoming data excludes the keyboard being unlocked. Indeed, the keyboard is only unlocked when a read is being performed, intr is done by catching a BREAK from the ATTN key, quit and eof are impossible, the encoding conversion renders swathes of important punctuation missing and even more of the commonly-used ASCII plane with entirely unrelated characters. It's astonishing this doesn't simply render the 2741 unfit for purpose, but letter-quality prints may well have made it just about worth it — Joe Morris and Dennis Ritchie, IBM Selectric parallel/serial interface, alt.folklore.computers, https://groups.google.com/g/alt.folklore.computers/c/M7H9LZrK894/m/-ncoZ9gVzPAJ, https://groups.google.com/g/alt.folklore.computers/c/M7H9LZrK894/m/zqHmiGrbqh4J, 1998-09-03/04.

Login processing and rudimentary 150-baud/300-baud detection in /etc/init (VII) is delegated to the new getty(VII), executed after the call is established: the speed selection is generalised slightly, in that the "login:" message at 150 baud is prefixed with ‘\033:\a’ to put the Model 37 in full-duplex mode; at 300 baud — with ‘\033;’ to do the same for the GE TermiNet 300; this "is acceptable for other 300 baud terminals also."; at 134.5 — with nothing, see above.

The username is read in raw mode (with standard erase and kill processing emulated) with either parity allowed, which allows the next speed to be picked and the message re-printed when a zero byte is read (as part of a BREAK). At 150 baud, the 2741's (134.5-baud) ATTN and enter "looks like the ascii »~« (1748)" (actually 1768; 1748 is also recognised but it's "|") and, if any of these is read, the 134.5-baud speed will be picked immediately (it is noted, however (login, logout(VII) (ibid., pp. 257-258)), that: "If the greeting doesn't appear after a few seconds, hang up and try again; something has gone wrong."). The "speed" here is loose, and includes model-specific flags (parity, 2741ness).

If the username is terminated with a carriage return instead of a newline, the post-getty mode will also have 020 (icrnl onlcr) set. If it didn't have any lower-case characters, it's mapped to lower-case and 004 (iuclc) is set as well.

Following this, it executes the also-new login(I) (ibid., p. 86), or

hup disappears, but stty(I) (ibid., p. 117) appears, and bears recalling in its entirety:

These map trivially to the stty(II) mode argument. Of these, [-]echo, [-]raw, [-]lcase, and [-]nl survive semantically equivalent, [-]canon as [-]icanon, [-]tab similarly to [-]tabs, and [-]even [-]odd similarly to [-]evenp [-]oddp.

UNIX Programmer's Manual, 2nd Edition, add'l bibliog.

¹:
Three keyboards/line codes/type balls for the 2741 are the "Standard Selectric® Typewriter" (correspondence), "PTTC/BCD", and "PTTC/EBCD" — IBM 2741 Communications Terminal, TP-09 A24-3415-2, Third Edition, http://bitsavers.org/pdf/ibm/2741/A24-3415-2_2741_Communication_Terminal.pdf, figs. 5-10, pp. 15-20, International Business Machines Corporation, Data Processign Division, 112 East Post Road, White Plains, N.Y. 10601.
The line is stateful for the ball hemisphere selected (lower case/upper case headings in even-numbered fig. ibid.) et al., with the sent/received codes corresponding to the rotation and tilt of the ball: BA1 and C42 are: g/G and 6/¢ for the correspondence models, a/A and 6/{' for the standard charset, : for the A and H ones, and ¢ for the typewriter(?) one} for the PTTC/BCD ones, and a/A and 6/' for the PTTC/EBCD ones. Trivially, since EBCDIC (or anything contemporary) isn't stateful, none of these are any given encoding.

UNIX Programmer's Manual, 3rd Edition

The three substantive changes herein are the correction of the 200 mode description to "M37s" and removal of the "It is all too easy to turn off your typewriter." warning from BUGS in K. Thompson and D. M. Ritchie, UNIX Programmer's Manual, Third Edition, https://www.tuhs.org/Archive/Distributions/Research/Dennis_v3/v3man.tar.gz, ./man2/stty.2, Bell Telephone Laboratories, Inc., February, 1973, stty(II).

As well as a drastic culling of modes allowed in stty(I) (ibid., ./man1/stty.1), now:

   even    allow even parity.
   -even   disallow even parity.
   odd     allow odd parity
   -odd    disallow odd parity
   raw     raw mode input
           (no erase/kill/interrupt/quit/EOT)
   -raw    negate raw mode
   -nl     allow cr for lf (and echo lf cr)
   nl      allow nl only
   echo    echo back every character typed.
   -echo   do not echo characters.
   lcase   map upper case to lower case
   -lcase  do not map case
   -tabs   replace tabs by spaces
   tabs    preserve tabs
   delay   calculate cr and tab delays.
   -delay  no cr/tab delays
   ebcdic  ebcdic ball conversion (2741 only)
   corres  correspondence ball conversion (2741 only)

UNIX Programmer's Manual, 4th Edition

Sees a rewrite of the kernel in C, of the manual in troff(1), and is infested by gnomes — K. Thompson and D. M. Ritchie, UNIX Programmer's Manual, Fourth Edition, https://www.tuhs.org/Archive/Distributions/Research/Dennis_v4/v4man.tar.gz, ./man0/basinf, Bell Telephone Laboratories, Inc., November, 1973, How to Get Started, Logging in..

getty(VII) (ibid., ./man7/getty.7) cycles between just 150- and 300-baud and kernel support for the 2731 is removed entirely, and type(I) (ibid., ./man1/type.1) once again becomes

The input buffer is bumped to 256 bytes (dc(IV) (ibid., ./man4/dc.4)), and overrun characters are "thrown out without notice". ASCII EOT is only stripped from output if not raw.

The "upper-case mode" (004) is additionally described as present-day xcase (with the mapped characters in the same order, even! This (and the already-extant lower-case) directly corresponds to the characters untypeable on a Teletype Model 33 (and 35), present on all DEC PDP-11s as standard²), but not predicated on icanon.

Recognisable (but unreliable, so not quite present-day) signals appear as signal(II) (ibid., ./man2/signal.2) and hanging up sends signal 1 (what we'd call SIGHUP), default action terminate without core, to all processes whose control typewriter matches the hanged-up one, then "returns with an end-of-file indication" as previous. This is as present-day.

This courtesy is extended to DEL's "interrupt signal" (2, SIGINT) and FS' "quit signal" (3, SIGQUIT). It's implied that but unclear how the shell protects itself from either signal.

stty(II) (ibid., ./man2/stty.2) sees a major rework with a SYNOPSIS of

(stty = 31.)
(file descriptor in r0)
sys stty; arg
...
arg:  speed; 0; mode

stty(fildes, arg)
int arg[3];

The speeds listed are the ones "supported by the DH-11 interface", lifted directly from DH11 asynchronous 16-line multiplexer user's manual, EK-ODH11-OP-002, 1st Edition, http://bitsavers.org/pdf/dec/unibus/EK-ODH11-OP-002_DH11_Asynchronous_16-line_Multiplexer_Users_Manual_Sep76.pdf, Chapter 3 (Programming), 3.2.3 (Line Parameter Register), p. 3-5, Digital Equipment Corporation, Maynard, Massachusetts, September 1976. Which are:

With two important differences: speed 0 is defined as "(turn off device)", and speeds 14 and 15 are defined as "External A" and "External B". This has made a lot of people very angry and been widely regarded as a bad move.

The attentive reader may notice that the speeds are listed for an entirely different device – a DH11 incessantly called a "DH-11" – than supposedly has a manual page and supposedly contains the /dev/tty? device family – dc(IV): DC11 (likewise, "DC-11"). Indeed, the reference implementation of the UNIX system has just "DC-11 interfaces actually present", but the PREFACE states (Fourth Edition, ./man0/intro):

The manual lists asterisks after the speeds supported by its DC11s (unwaveringly 134.5, 150, 300, 1200), warning that "if a non-starred speed is selected, it will be ignored and the present speed left unchanged.". Even further:

One could say that this is a primordial version of inter-UNIX portability, like the outsourcing of mounting of /usr (and /sys), changing the mode of the DECtape nodes and spawning dpd(VII) (ibid., ./man7/dpd.7) from the init(VII) process itself to the new-in-Fourth Edition /etc/rc. Or, given that getty(VII) still uses a fixed table of remote teletype types and init(VII) — of local teletype lines to wait on, one could be reaching.

Beyond removal of the 2741 400, 1000, 2000 modes, 020 is re-described to echo "as CR-LF" despite a subsequent paragraph describing it and dc(IV) remaining unchanged as "LF-CR", and a new mode:

The console teletype kl(IV) (ibid., ./man4/kl.4)'s "line disciplines" "are identical to those for tty?", and it accepts gtty(II)/stty(II), with the only "salient differences" being the different key for SIGQUIT and that "the speed-select word is ignored". One could say that the latter isn't a difference, since that's the behaviour for most speeds on dc(IV), but.

stty(I) (ibid., ./man1/stty.1)'s raw parenthetical gains "parity bit passed back"(?), -nl also turns around as "CR-LF" to match the syscall, [-]delay also mentions form-feed (no functional change, presumably? given that form feed is mentioned in the Second Edition as well), and ebcdic corres are replaced with

The BUGS section becomes

UNIX Programmer's Manual, 4th Edition, add'l bibliog.

²:
KL11 teletype control manual, DEC-11-HR4C-D, 3rd Printing, http://bitsavers.org/pdf/dec/unibus/KL11_TeletypeControlManual.pdf, Chapter 2 (General Description), 2.1 Introduction, para. 2, Digital Equipment Corporation, Maynard, Massachusetts, September 1970.

H720 power supply and mounting box manual, DEC-11-HR5B-D, 5th Printing, http://bitsavers.org/pdf/dec/pdp11/pwrSupply/H720_PowerSupplyMan.pdf, Chapter 4 (Installation), LT33 Teletype, LT35 Teletype, pp. 4-7, 4-8, Digital Equipment Corporation, Maynard, Massachusetts, September 1970. As well as ibid., Table 4-1 (PDP-11 System Configurations), p. 4-1.

Technical Manual, Model 32 and 33, Bulletin 273B, http://bitsavers.org/communications/teletype/33/273B_Mod32_33_TechMan_Jan63.pdf, Section 7 (Principles of Operation), 7-2. Code, Fig. 7-1.a. (Character Arrangement), Teletype Corporation, 555 Touhy Avenue, Skokie, Illinois, 1962, 1963.

The Model 35 has a Shift key, so it could be used to type lower-case letters, but it inverts bit 5, so it can't be used to type special characters from xcase: 35 Keyboard and Base for Keyboard Send-Receive (KSR) and Receive-Only (RO) Teletypewriter Sets, Description and Operation, Bulletin 221B, Section 574-221-100, Volume 1, http://bitsavers.org/communications/teletype/35/281B_Mod35_TechVol1_Apr73.pdf, Section 2 (Principles of Operation), 2.20 (Shift Key Mechanism), p. 12, Teletype Corporation, 555 Touhy Avenue, Skokie, Illinois, 1962, 1964, 1965.
It can't print lower-case characters: 35 Typing Unit, General Description and Principles of Operation, Bulletin 221B, Section 574-220-100, Volume 1, http://bitsavers.org/communications/teletype/35/281B_Mod35_TechVol1_Apr73.pdf, Section 2 (Principles of Operation), 2.04 (Signaling Code), Fig. 5, p. 9, Teletype Corporation, 555 Touhy Avenue, Skokie, Illinois, 1963.

UNIX Programmer's Manual, 5th Edition

300-baud teletypes are now the default — K. Thompson and D. M. Ritchie, UNIX Programmer's Manual, Fifth Edition, https://www.tuhs.org/Archive/Distributions/Research/Dennis_v5/v5man.pdf, p. iv, Bell Telephone Laboratories, Incorporated, June, 1974, How to Get Started, Logging in.. So the getty (VIII) (ibid., p. 267) speed ordering is reversed to try it first, then fall back to 150-baud on BREAK. The 300-baud "login:" prefix additionally returns "the GSI terminal to non-graphic mode".

init (VIII) (ibid., p. 268) now reads /etc/ttys (ttys (V) (ibid., 211)) as a list of lines to listen on and programs to spawn when called (but "the only program currently specifiable is /etc/getty (q.v.).", since the program is an offset into an init-internal table) and accepts "the hangup signal (signal #1)" (SIGHUP) to reload and reprocess it "as in a reboot" — remnants of this remain in all present-day inits.

tty (IV) (ibid., p. 194) is re-instated as a description of all "low-speed asynchronous communications ports", with the hardware-specific peculiarities relegated to their respective dc (IV) (ibid., 180), dh (IV) (ibid., 181), and kl (IV) (ibid., 184) pages, all of which unchanged and equally uninteresting, except to see that the DC11 /dev/tty[01234567abcd] is described as "not currently connected", and appears replaced on the reference implementation by the DH11 at /dev/tty[f-u]. The only significant change is that the default speed is also 300 baud and individual settable speeds are not listed.

The "parity bit passed back" bit is clarified in the raw paragraph as

"If you find it hard to type" FS, "try control-\ or control-shift-L."; the other half of the new BUGS is that "Half-duplex terminals are not supported." — this is supported by the nonexistence of type.

stty (II) (ibid., p. 131) loses asterisks after "supported speeds" and generalises the unknown-speed ⇒ ignored behaviour to all the aforementioned devices for the aforementioned reasons. Besides that, it appears unchanged — the dc (IV) speeds are stil listed as if globally true and mode 10000 (tab0) is crossed out by hand entirely, with the description of mode 001 (nl0 cr0 tab0 ff0) crossed out and "hangup on last close" penned in — this is [-]hupcl, with that exact phrasing of the flag surviving to IEEE Std 1003.1-202x (“POSIX.1”), Draft 2.1 .

It's unclear what this means, especially given that stty (I) (ibid., p. 79) is unchanged entirely.

UNIX Programmer's Manual, 6th Edition

Sees the introduction of /dev/tty, corresponding to the process's control typewriter (the console is exclusively /dev/tty8, a synonym in the Fifth edition), as present-day, "a variety of delays after function characters", rather than model-class enumeration for each "function" (NL/CR/tab/FF), speed changes explicitly predicated on "if the hardware permits" — K. Thompson and D. M. Ritchie, UNIX Programmer's Manual, Sixth Edition, https://www.tuhs.org/Archive/Distributions/Research/Dennis_v6/v6doc.tar.gz, ./man/man4/tty.4, May, 1975, tty (IV). ‘#’ and ‘@’ are "By default", and EOT is finally parenthesised as "control-D".

init (VIII) (ibid., ./man/man8/init.8) always runs getty, and the third ttys (V) (ibid., ./man/man5/ttys.5) column is instead passed as an argument thereto:

getty (VIII) (ibid., ./man/man8/getty.8) mentions no such modes, simply predicating the description of the former 300-/150-baud operation on being passed ‘0’. However, the getty of the reference system is available at https://www.tuhs.org/Archive/Distributions/Research/Dennis_v6/v6src.tar.gz, ./s1/getty.c. And reveals: a ‘-’ mode for the "Console TTY" whose 110 icrnl onlcr tab3 iuclc olcuc xcase cr1 point to it continuing to be the same Model 33, with an entry in the default table after the 150-baud mode.

There's also a ‘1’ table, which contains only the 150-baud entry, and a ‘2’ one, described only as "9600" — 9600 icrnl onlcr tab3 ff1.

stty (II) (Sixth Edition, ./man/man2/stty.2) finally sees a new SYNOPSIS:

(stty = 31.)
(file descriptor in r0)
sys stty; arg
...
arg:  .byte ispeed, ospeed; .byte erase, kill; mode

stty(fildes, arg)
struct {
        char    ispeed, ospeed;
        char    erase, kill;
        int     mode;
} *arg;

The erase, kill members are self-explanatory. The mode table is vastly expanded:

The delays are very similar to present-day, but described in terms of narrow-bit integers, rather than bit-fields, and form-feed is overloaded with vertical tab, and a zeroed delay value corresponds to no delay for that function byte; respectively, they correspond to: BSDLY, FFDLY=VTDLY, CRDLY, TABDLY, NLDLY, and are expounded as follows.

Mode 000001 corresponds directly to [-]hupcl, and "is useful when a port is to be used for some special purpose; for example, if it is associated with ACU used to place outgoing calls.", with "ACU" presumably standing in for an Automatic Calling Unit.

stty (II) is "also used with certain special files other than typewriters" — it's the only non-I/O entry in the device driver interfaces, cf. Dennis M. Ritchie, The Unix I/O System, https://www.tuhs.org/Archive/Distributions/Research/Dennis_v6/v6doc.tar.gz, Character device drivers, Bell Telephone Laboratories, ./iosys/iosys. And compare v6src.tar.gz ./usr/sys/conf.h bdevsw/cdevsw — "but since none of them are part of the standard system the specifications will not be given."; one could consider this a primordial ioctl(2).

stty (I) (Sixth Edition, ./man/man1/stty.1) allows no arguments, in which case it "reports the current settings of the options", much like present-day, The "[-]delay" and "[-]tdelay" options are removed in favour of:

Of these, cooked, [-]hup, and the speeds are retained verbatim (but see STANDARDS, Line Speed), ek using a system default (CERASE, CKILL), and erase kill are equivalent for the cc-is-single-byte case.

v6src.tar.gz ./s2/stty.c reveals that there are [-]LCASE, aliases for [-]lcase, as present-day, a 134 alias for 134.5, 33 37 05 tn ti aliases for tty33 tty37 vt05 tn300 ti700, respectively, which replace delay flags with cr1, ff1 cr2 tab1 nl1, nl2, cr1, cr2, ff1, respectively. The bs[01] delays, conspicuously missing from the manual, are there, but just conspicuously missing from the manual.

The no-argument output is ideologically much like present-day sane, deviations except much less formalised (and written to the standard error stream): the first lines are either

input speed  ispeed baud
output speed ospeed baud

speed speed baud

erase = 'cc'; kill = 'cc'

UNIX Programmer's Manual, 7th Edition

Many more and different models are listed, as expected from four years of increasingly affordable cash cows being bred; the TTY 37 terminal section is generalised as 1200- and 150-baud terminals to cover causing BREAKs until the correct speed is achieved, and a new Hard-wired terminals section appears, saying that they "usually begin at the right speed, up to 9600 baud", and defers to the normal instructions otherwise — UNIX™ Time-Sharing System: UNIX Programmer's Manual, Seventh Edition, Volume 1, https://www.tuhs.org/Archive/Distributions/Research/Henry_Spencer_v7/v7.tar.gz, usr/man/man0/intro, Bell Telephone Laboratories, Incorporated, January, 1979, Introduction to Volume 1, How to Get Started, Logging in..

The "'break' or 'interrupt' key" — the one that generates the BREAK used to cycle between speeds — sends SIGINT, with the usual DEL method seemingly being second-class (ibid., How to communicate through your terminal., para. 4).

As for Stty and gtty, "These system calls have been extensively altered, see ioctl(2) and tty(4)." (ibid., Converting From the 6th Edition).

Indeed, they're now described in the new ioctl(2) (ibid., usr/man/man2/ioctl.2) as shims for ioctl(TIO[GS]ETP), and are provided as such by Standard C Library (libc, -lc); the formerly-stty (II) information is merged into tty(4) (ibid., usr/man/man4/tty.4), though it's now about "terminals", rather than "typewriters". The individual-device pages are shed, as are references thereto.

"The set of processes that thus share a control terminal is called a process group" — this has no bearing on the actual semantics of the system, and process groups are acquired (and shed) differently in present-day, but this is the first notion of the process group — kill(2) (ibid., usr/man/man2/kill.2), when signalling pid 0, signals the members of the caller's process group: this is as present-day.

The argument format is the same, but now defined in <sgtty.h>, designated struct sgttyb, with members prefixed with sg_, and mode — sg_flags. The speeds themselves are defined with their present-day Bintspeed, EXTA, EXTB macros.

The TIOCSETN ioctl() is like TIOCSETP, but doesn't wait for output empty then dump input buffer. Mode 000001 (HUPCL) is moved to ioctl(TIOCHPCL) (description retained), and becomes "TANDEM".

The delay mask macros appear (humorously, with "VTDELAY" governing FF[01]), and "XTABS" becomes the third tab delay mode (present-day TAB3). Backspace delays are downgraded to "might".

Besides those, the low-order byte bit layout is the same:

Raw mode is extended with

The new "CBREAK" mode is described as

The TIOC[GS]ETC (misdocumented as TIOCSETC and TIOCSETP again) ioctl()s, allows changing the "special characters", with defaults outlined below, via the following <sgtty.>> type:

struct tchars {
	char	t_intrc;		/* interrupt */
	char	t_quitc;		/* quit */
	char	t_startc;	/* start output */
	char	t_stopc;	/* stop output */
	char	t_eofc;		/* end-of-file */
	char	t_brkc;		/* input delimiter (like nl) */
};

Indeed — this is the first system with truly settable Special Control Characters and it includes a disabling functionality, though spelled as -1 (rather than today's _POSIX_VDISABLE).

start and stop are fully formed (in ixany mode).

eol is fully formed:

The new "TANDEM" mode sends t_stopc/t_startc in response to the input buffer water level: "It is useful for flow control when the 'terminal' is actually another machine that obeys the conventions.". The name hints that this was intended for talking with Tandem Computers, Inc. systems, presumably; it's not used in the userland of the distribution tape at all.

The BUGS lose the "parity should be transmitted" sentence and gain an even more indicting paragraph:

init(8) (ibid., usr/man/man8/init.8), rather than checking the console switches, just starts /bin/sh </dev/console (the latter replaced tty8), and boots multi-user when the shell exits; as expected, SIGHUP restarts the init sequence at single-user mode.

getty(8) (ibid., usr/man/man8/getty.8) is generalised to say that the "login:" message "can contain character sequences to put various kinds of terminals in useful states." and enumerates the following ttys(5) second-column values:

Special Control Characters are reset to their defaults together with setting the initial username-asking mode.

stty(1) (ibid., usr/man/man1/stty.1) lists bs[01] in the manual, replacing the ominous "(see below)"s on all delay modes and "various delay algorithms" note with a "(see ioctl(2))" on the first (cr), adds

[-]hup is replaced with just hup, calling ioctl(TIOCHPCL). It's unclear how you'd unset this. Removal of -hup is undocumented.

An undocumented gspeed mode, equivalent to ispeed 300 ospeed 9600 (but keep in mind neither primitive exists yet) appears, and tab3 is, naturally, removed (but not from the manual),

The no-argument output formats nonprintable (control) erase and kill in the expected way; the flag line is unchanged as much as possible, with the explicit singling-out of XTABS as -tabs (and no tab[012]).

Third Berkeley Software Distribution

The section covering @# notes that "Most users prefer to use a backspace rather than '#', and many prefer control X instead of '@'", pointing to tset(1) (cf. Lies by Omission) and stty(1) to alter this — UNIX Programmer's Manual, Seventh Edition, Virtual VAX-11 Version, https://archive.org/details/The_CSRG_Archives_CD-ROM_1_August_1998_Marshall_Kirk_McKusick, Marshall Kirk McKusick, The CSRG Archives, CD-ROM 1, Berkeley Systems 1978-1986, 3bsd, usr/man/man0/intro, Computer Science Division, Department of Electrical Engineering and Computer Science, University of California, Berkeley, California 94720, December, 1979, Introduction to Volume 1, How to Get Started, How to communicate through your terminal., para. 3.

getty,vm(8) (ibid., usr/man/man8/getty.vm.8) loses the table enumeration in the manual (and moves some tables around, but there are 13 with 16 total modes, one of which being an "LSI Chess Terminal"(?) whose login message is "*"; these are out of scope) and special escapes before the login messages (except for table ‘1’), sends a line consisting of "Virtual myname", where the latter is a primordial form of hostname defined in <ident.h>, before the "login:" one, explicitly ignores (i.e. re-does the identification and prompt at the same speed) login names without letters, and sleeps for a second after prompting and before trying to read the name, then stty()s again to dump the buffer:

                /*
                 * Wait a while, then flush input to get rid
                 * of noise from open lines
                 */

There is a stty(2) (ibid., usr/man/man2/stty.2) but it's an odd mishmash of Sixth Edition values (esp. HUPCL) with Seventh Edition tty(4) format. tty(4) (Seventh Edition, Virtual VAX-11 Version, December, 1979, usr/man/man4/tty.4) is expectedly as in Seventh Edition.

stty(1) (Seventh Edition, Virtual VAX-11 Version, December, 1979, usr/man/man1/stty.1) removes the '^X' description, lists default values for erase kill and finally allows setting the Special Control Characters:

By peeping the source (ibid., usr/src/cmd/stty.c), we find that despite the '^X' removal, it's both still valid, and ^? is accepted as DEL (0177, 0x7F), as well as that Special Control Characters not followed by a character is ignored instead of segfaulting.

The no-argument output is suitably updated as well: after the speed lines, the Special Control Characters are listed as such:

erase = cc; kill = cc; intr = cc; quit = cc
start = cc; stop = cc; eof = cc; brk = cc

The settings for non-delay modes (incl. XTABS ([-]tabs)) are always written, format otherwise unchanged.

Fourth Berkeley Software Distribution

Refuses input and rings the bell on input overflow (≅imaxbel) "if your prompt was a %" — that is, with csh(1), that is, with the "new" line discipline — and declares that "many prefer control-U instead of '@'" now; additionally, "Many users change this interrupt character to be ^C (control-C) using stty(1)."; finally,

Indeed, as the PREFACE (ibid., usr/man/man0/intro) states: "This release finds the system in transition: a number of facilities have been added in experimental versions (job control, resource limits) and" "Applications which use facilities that are in transition should be aware that some of the system calls and library routines will change in the near future." — the extent to which the latter is not the case is remarkable: these facilities remain enshrined, essentially unchanged, to this day.

tty(4) (ibid., usr/man/man4/tty.4) more than doubles in size; this is reasonable, given that it describes double the amount of "terminal drivers used for conversational computing" (nominally three, but the third is the "net" NETLDISC one, used for Berknet, described in full in bk(4) (ibid., usr/man/man4/bk.4), and entirely out of scope here) — line disciplines, configurable with ioctl(TIOC[GS]ETD), with the "old (standard) terminal driver", "used when using the standard shell sh(1) and for compatibility with other standard version 7 UNIX systems" defined as OTTYDISC (0 by convention), and the "newer terminal drivers, with features for job control", which (must be used when using csh(1)), as NTTYDISC.

This is also the system that originates modern-day process groups — complete with [gs]etpgrp(), ioctl(TIOC[GS]PGRP), &c., cf. jobs(3J) (ibid., usr/man/man3/jobs.3j) — which are fully formed, to ε, with the "new" discipline, and as such so truly remarkably unremarkable that one may as well consult modern credentials(7).

The three major modes — cooked, CBREAK, RAW — are described completely and separately as "all driver functions" available, no input processing, and no processing (sans LFLUSHO), respectively. RAW returns BREAKs as NULs; this is the first time this is documented outside getty(8).

For input, NTTYDISC introduces "a literal-next character ^V" (lnext, as present-day) "which can be typed in both cooked and CBREAK mode preceding any character to prevent its special meaning.", which is preferred to escaping erase kill with a \, retained only for "historical reasons" (but even that sees an update with only the escaped character being echoed). as well as "The word-erase charactrer, normally ^W", which "erases the preceding word", "defined as a sequence of non-blank characters, with tabs counted as blanks" (werase, as present-day, to precision of altwerase). and "Finally, the reprint character, normally ^R, retypes the pending input beginning on a new line.", which happens "automagically in cooked mode if characters which would normally be erased from the screen are fouled by program output."; it's unclear how this is mediated.

In OTTYDISC erase is just echoed and kill is additionally followed by a newline (as-if echok -echoke), "(even if the character is not killing the line, because it was preceded by a '\'!.)" (notably, this isn't documented in any manuals heretofore; it's unclear when exactly this stopped being -echok) — whereas "new terminal driver has several modes for handling the echoing of terminal input, controlled by bits in a local mode word." (notably, this is the first occurrence of the "local flags", so spelled and so separated). These flags are largely as present-day behaviourally, though decidedly not as present-day spelling-wise, and are outlined below.

For output, EOT is not transmitted in cooked mode (as opposed to only if RAW), so "programs using raw or cbreak mode should be careful"; NTTYDISC adds "a output flush character, normally ^O" (discard, as present-day), "which sets the LFLUSHO bit in the local mode word, causing subsequent output to be flushed until it is cleared by a program or more input is typed." (flusho, as present-day) — which happens in cooked and CBREAK modes and input is rprnted after resumption.

If LCASE, under NTTYDISC, upper-case bytes, in addition to the punctuation, are preceded by ‘\’es on ouypuy, which is finally as present-day. A provision for "Hazeltine terminals, which do not understand that ~ has been made into an ASCII character" is made via LTILDE, which outputs ~ (0x7E)s as ` (0x60)s.

When the line is hung up, the "distinguished" (foreground) process group is signalled, as present-day; LNOHANG disables this.

The interface is, expectedly, insane: "there are 4 different structures which contain various portions of the driver data."; struct sgttyb sg_flags returns to being 16-bit by becoming short, with no other layout changes, but CR3 "is suitable for the concept-100 and pads lines to be at least 9 characters at 9600 baud." The former refers to Human Designed Systems Concept-series video terminals; the latter is unclear. This corresponds to no modern mode.

struct tchars is unchanged.

The new "Local mode", a word, applies (except for LNOHANG), to NTTYDISC only, settable with TIOCL[GS]ET, with bit-manipulation ioctl()s TIOCLBI[CS] having the expected result.

The erasure mode tree is slightly different; assuming echo:

LMDMBUF functions like TANDEM, but rather than reacting to water-marks, it does to carrier events, and only does something under the "new" DM-BB driver, which bears a "NB: WE HAVEN'T TESTED dhdm CODE ON VAX.", disclaimer. LETXACK is unused and otherwise undocumented. LPENDIN is a flag set internally by the driver to facilitate reprinting input after TIOCSET[PN] from CBREAK to cooked input (settable by userspace, but not used by the distribution userland).

LINTRUP is unused in the distribution userland except as stty [-]intrup. A mode rather similar to LNOHANG may be found in clocal.

Likewise for the new "Local special chars", TIOC[GS]LTC:

struct ltchars {
	char	t_suspc;	/* stop process signal */
	char	t_dstopc;	/* delayed stop process signal */
	char	t_rprntc;	/* reprint line */
	char	t_flushc;	/* flush output (toggles) */
	char	t_werasec;	/* word erase */
	char	t_lnextc;	/* literal next character */
};

The BUGS concedes to TENEX fans and that paragraph disappears.

This is undocumented, but the "Local special chars" are reset to kernel defaults when a teletype device is opened and doesn't have any open descriptors. This explains why getty (ibid., usr/src/cmd/getty.c) just resets the discipline to OTTYDISC.

stty(1) (ibid., 4.0/usr/man/man1/stty.1), rather than the "current settings", reports:

Character arguments may start with "u" to mean undefined (-1) — the manual describes this as "»u« or »undef«".

raw cbreak are updated to reflect their behaviour in tty(4), and hup -hup are removed (but only from the manual, hup remains unchanged in usr/src/cmd/stty.c). The only new options in the main block are, at last:

A separate chunk of options is added, caveated on applying only to the new driver (with the new csh):

And a similarly-caveatted (and "not normally changed") list of characters:

There is an (undocumented, of course) query: speed, which writes the output speed of /dev/tty to the standard output stream and exits 0 (or writes "unknown" and exits 1 if the speed is unknown).

"crt", additionally to described, clears LPRTERA; the rationale is, naturally, found in tty(4): "At speeds less than 1200 baud, the LCRTERA and LCRTKILL processing is painfully slow" — and can be approximated at a tenth of a second per column erased at 300 baud.

The output format is globally compressed; the first line starts with "net discipline, " if bk(4) is in use, "new tty, " if NTTYDISC, or "old tty, " if "everything". The speeds are printed on the same line (by just dropping the newline from the end of the speed printing function; this, of course, means that if the input and output speeds are different, like in "gspeed" mode, they're written as "input speed 300output speed 9600"; this once more supports the point that split-speed teletypes just weren't in use). Thus the line ends if "everything" or a semicolon separates otherwise.

The mode output is identical (sans "[-]tandem", after "[-]lcase") if "everything", otherwise the "defaults" are EVENP | ODDP, !RAW, CRMOD, ECHO, !LCASE, !TANDEM, (XTABS always reported), !CBREAK, and no delays. If "all" or "everything" the line is broken.

The default "Local mode"s are all unset; in NTTYDISC, they follow in (mostly) the order listed. The only quirk is that "crt" is coalesced and listed as-if a first-class option if the "crt*" and "prterase" all match, except that it's considered to start to include LCRTERA | LCRTKIL at 600 baud, and if "everything" is listed as "crt: (crtbs crterase crtkill ctlecho)"; the line is broken here if any flags were listed.

In no-argument output, unless under bk(4), non-default Special Control Characters (and, under NTTYDISC "Local special chars") are written in the same format as previous, but only if they differ from the defaults (as described, except for erase which is ^H); otherwise, if "all" or "everything", they're listed in a two-line tabulated format:

erase  kill   intr   quit   stop   eof

This is, perhaps, best illustrated with an example; for a VT05 in a reasonable stty 1200 vt05 new crt kill ^X rprnt u susp u nohang mode, stty all yields:

speed 150 baud; nl nl1 tab1 cr1 ff1 prterase
erase #

4.1 Berkeley distribution

Adds

Which are equivalent to [-]ixany and [-]noflsh.

Computer Science Division, Department of Electrical Engineering and Computer Science, University of California, 4.1BSD, https://archive.org/details/The_CSRG_Archives_CD-ROM_1_August_1998_Marshall_Kirk_McKusick, Marshall Kirk McKusick, The CSRG Archives, CD-ROM 1, Berkeley Systems 1978-1986, 4.1/4.0.upgrade, sys/h/ioctl.h, sys/dev/ttynew.c.

4.2 Berkeley Software Distribution

getty, taking the table name, reaches its full potential by iterating through a (fully-formed) termcap at /etc/gettytab, which includes a default with the (now actual) hostname — UNIX Programmer's Manual, 4.2 Berkeley Software Distribution, Virtual VAX-11 Version, https://archive.org/details/The_CSRG_Archives_CD-ROM_1_August_1998_Marshall_Kirk_McKusick, Marshall Kirk McKusick, The CSRG Archives, CD-ROM 1, Berkeley Systems 1978-1986, 4.2, usr/man/man8/getty.8, usr/man/man5/gettytab.5, Computer Science Division, Department of Electrical Engineering and Computer Science, University of California, Berkeley, California 94720, August, 1983, getty(8), gettytab(5). The latter's BUGS note that:

The relevant revolutionary new part of getty(8) is that can hang up if the login name is not obtained in time (to property).

All mentions of stty()/gtty() are removed from ioctl(2) (ibid., usr/man/man2/ioctl.2), which just directs the reader to the section-4 page they're interested in; stty(3C) (ibid., usr/man/man3/stty.3c), where C is for Compatibility, is described as defunct and obsolteded by ioctl(2).

LINTRUP is "replaced by a call to fcntl and the SIGIO signal." — Samuel J. Leffler, Bug fixes and changes in 4.2BSD, https://archive.org/details/The_CSRG_Archives_CD-ROM_1_August_1998_Marshall_Kirk_McKusick, Marshall Kirk McKusick, The CSRG Archives, CD-ROM 1, Berkeley Systems 1978-1986, 4.2, usr/doc/uchanges/3.t, Berkeley, California 94720, July 28, 1983, Section 2.. The equivalent call is F_SETFL with FNDELAY | FASYNC.

tty(4) (4.2 Berkeley Software Distribution, usr/man/man4/tty.4) expands on LDECCTQ as "Only ^Q restarts output after ^S, like DEC systems"; LNOFLSH remains entirely undocumented.

struct ltchars t_dstopc is renamed to t_dsuspc (it's unclear why it hasn't been that from the beginning), and t_werasec to t_werasc (to fit in the 8-byte limit of older compilers, presumably).

stty(1) (ibid., usr/man/man1/stty.1) sees the addition of

The DEC operating systems are pretty much any of them, to an extent: TENEX doesn't count, since that's originally BBN, but Third Berkeley Software Distribution says that "many prefer control X instead of '@'" — TENEX uses ^X: Theodore H. Myer, Jogn R Barnaby, and William W. Plummer, TENEX Executive Manual, TENEX Executive Language, Manual for Users, Bolt Beranek and Newman Inc., http://bitsavers.org/pdf/bbn/tenex/TenexExecRef_Apr73.pdf, p. 18, 50 Moulton Street, Cambridge, Massachusetts 02138, January, 1971; revised April, 1973, II. Executive Language Structure, C. Editing and Errors in TENEX commands, 4. (CTRL)-X, Rubout. And single-character erasure echoes a \ followed by the erased column every time. ^R reprints the line after CR.

TOPS-20 is almost there: DEL is erase, ^U is kill — TOPS-20 System Manager's Guide, Version 6.0 Interim Release Draft, http://bitsavers.org/pdf/dec/pdp10/TOPS20/TOPS-20_V6_System_Managers_Guide_Interim_Dec84.pdf, p. 2-9, Digital Equipment Corporation, December 1984 (first printing October 1976), Creating the TOPS-20 File System, Step 13, DELETE Key.

VMS sees the full range here: ^O, ^S, ^Q, ^R, ^U, all as-described, and ^X which is equivalent to ^U in the UNIX context — VAX/VMS I/O User's Guide, AA-D028A-TE, http://bitsavers.org/pdf/dec/vax/vms/1.0/AA-D028A-TE_VAX_VMS_IO_Users_Guide_Aug78.pdf, pp. 2-5, 2-6, 2-7, Digital Equipment Corporation, Maynard, Massachusetts, August 1978, Chapter 2, Terminal Driver, 2.2.6 Control Characters and Special Keys, Table 2-1, Termianl Control Characters.

Per the juices (4.2 Berkeley Software Distribution, usr/src/bin/stty.c), 19200 38400 speeds are added as aliases for exta extb and are written as such, rather than "0", in the current-option listing, "[-] extack" is #ifdefed away, and [-]noflsh added. Default characters are now part of the global <sys/ttychars.h>, as present-day, and with present-day names; the surprising difference is that erase kill intr got DECified into ^?, ^U, ^C (from #, @, ^? in the kernel and ^H, @, ^? in userspace (stty, reset — Seventh Edition, Virtual VAX-11 Version, November, 1980, usr/src/cmd/reset.c)).

4.3 Berkeley Software Distribution

No longer "must" one "call UNIX from an appropriate terminal"; instead, a phone number is necessary only "If you will be accessing UNIX remotely". A direct connection "generally runs at 9600 or 19200 baud", and modem speed is "typically 300, 1200, or 2400 baud" — UNIX User's Reference Manual (URM), 4.3 Berkeley Software Distribution, Virtual VAX-11 Version, https://archive.org/details/The_CSRG_Archives_CD-ROM_1_August_1998_Marshall_Kirk_McKusick, Marshall Kirk McKusick, The CSRG Archives, CD-ROM 1, Berkeley Systems 1978-1986, 4.3, usr/man/man0/intro.ms, Computers Systems Research Group, Computer Science Division, Department of Electrical Engineering and Computer Science, University of California, Berkeley, California 94720, April, 1986, Introduction to User's Reference Manual, How to get Started, Logging in.. For logging out, the logout csh(1) primitive is recommended above EOFing the shell (ibid., Logging out.). The How to communicate through your terminal. section is adamant that DEL and ^H have replaced @# by default — there is no evidence of this: they remain ^H and ^U. And provides an ominous note:

ttys(5), rather than being a trivial columnated list, is now fully comment-capable and paraeterised with regards to what to run with full cmdline, where, what type of terminal is on it, and arbitrary parameters to the programs run; this includes running xterms; this is as present-day in most Berkeley derivatives — UNIX Programmer's Reference Manual (PRM), 4.3 Berkeley Software Distribution, Virtual VAX-11 Version, https://archive.org/details/The_CSRG_Archives_CD-ROM_1_August_1998_Marshall_Kirk_McKusick, Marshall Kirk McKusick, The CSRG Archives, CD-ROM 1, Berkeley Systems 1978-1986, 4.3, usr/man/man5/ttys.5, Computers Systems Research Group, Computer Science Division, Department of Electrical Engineering and Computer Science, University of California, Berkeley, California 94720, April, 1986. tty(4) (ibid., 4.3/usr/man/man4/tty.4) finally stops mentioning bk(4) (ibid., 4.3/usr/man/man4/bk.4) (though it's not fully removed, only marked "(obsolete)" with a BUGS section that lays out why it's not fit for purpose), and the listing is reworded to "there are two line disciplines available for use with terminals" which makes much more sense, given that bk(4) is (unsuitable) for uploading data from microprocessors; ominously, "Other disciplines may exist for special purposes, such as use of communications lines for network connections.".

TIOCNOTTY, detaching the caller from its controlling terminal, available since 4.2 Berkeley Software Distribution, is finally documented. eof is finally documented as one of the inputs that end lines in cooked mode in-line. The DECified defaults are propagated to the description. TIOCHPCL is now "usually done on outgoing lines and dialups.", and the userland agrees with this.

Rather than always outputting even parity, odd parity may be generated by setting ODDP !EVENP.

TIOCSTOP and TIOCSTART ioctl()s appear to emulate pressing stop start. It's unclear if TIOCSTIing them would suffice.

One new "Local mode" bit appears:

Window size provisions appear fully-formed as present-day, "to accommodate terminals and workstations with variable-sized windows": SIGWINCH, TIOC[GS]WINSZ, and

struct winsize {
	unsigned short	ws_row;	/* rows, in characters */
	unsigned short	ws_col;	/* columns, in characters */
	unsigned short	ws_xpixel;	/* horizontal size, pixels */
	unsigned short	ws_ypixel;	/* vertical size, pixels */
};

The new "Local mode"s are prescribed as such

size is implemented (ibid., usr/src/bin/stty.c), like "speed", by re-opening /dev/tty and exiting 0; the purpose of this is unclear.

The window size is written as ", ws_row rows, ws_col columns" after the speed, if "all" or "everything".

UNIX Programmer's Manual, CB Version

Sees a truly unique Logging in. section, alluding to well-developed port-specific getty tables:

A new Logging out, method of sitting around for a while is noted, based on (lack of) teletype activity. As compared to Sixth Edition (from which a "large part of the present manual's contents is descended" (CB-UNIX Programmer's Manual, cbunix_intro.pdf, ACKNOWLEDGEMENTS, p. iii)), ^S can "temporarily stop output", which "is useful with CRT terminals to prevent output from disappearing before it can be read.", and is only resumed with a ^Q — these are as-present-day stop start defaults in -ixany mode. Additionally,

2.3 is, of course, a late version of the system, but, in the words of an unnamed "retired CB/UNIX developer";

The core difference there is that that file is /dev/ln — this follows the console being ln00 and the dial-in lines being ln*. dh (4) (CB-UNIX Programmer's Manual, cbunix_man4.pdf, p. 2) is extended to also cover DZ11s (also incessantly called "DZ-11"s): "(For DZ-11 lines, output speed is always the same as input speed. The 200 speed and the two externally clocked speeds (exta, extb ) are missing on the DZ-11.)". The DZ11 supports a different set of speeds:

And the "Transmitter clock is always on." (ibid., Table 3-3 LPR Bit Functions, p. 3-8). How this is negotiated is not noted, but CB-UNIX Source Code, cbunix7.pdf, pp. 64-72 (dz.c), bears:

/* this array maps the standard unix transmit and receive speeds
 * into the speeds used int he dz11 hardware. If a speed is selected
 * that is not available on the dz11 it will be ignored.
*/

char dzspeed[]
{
        0000,           /*  not available */
        0000,           /*    50 baud */
        0001,           /*    75 baud */
        0002,           /*   110 baud */
        0003,           /* 134.5 baud */
        0004,           /*   150 baud */
        0000,           /*  not available */
        0005,           /*   300 baud */
        0006,           /*   600 baud */
        0007,           /*  1200 baud */
        0010,           /*  1800 baud */
        0012,           /*  2400 baud */
        0014,           /*  4800 baud */
        0016,           /*  9600 baud */
        0000,           /*  not available */
        0000,           /*  not available */
};

        lpr = ((dzspeed[tp->t_speeds.lobyte]<<8) | ((dev.d_minor&07) |
                        RECV_ON));

The kl (4) (CB-UNIX Programmer's Manual, cbunix_man4.pdf, p. 6) ln00 name is "for historical reasons only": the canonical keying for "the real system console" is systty, and the current console if going single-user — "Init (1M) and reboot (1M) talk to this device" — is syscon, which telinit (1M) links to the invoking teletype when passed runlevel S.

This nomenclature is matched with line(1) (ibid., cbunix_man1_03.pdf, p. 100) — "get line identification" — which "gives the name of the user's line in the form lnnn in the range 00 to 99" (or "lnxx if the standard input file is not a typewriter"); the familiar tty(1) (ibid., cbunix_man1_09.pdf, p. 21) interface is also available, writing the full path and accepting -s, as present-day.

Opening tty (4) "causes a wait to take place at the first read or write.", rather than waiting "until a connection is established.". The description for raw mode is extended; rather being "awakened on each character":

The start stop-and-friends spiel is reproduced here entirely, but "A BREAK character is treated like and ESC character when not in raw mode." only; this makes sense.

getty (1M) (ibid., cbunix_man1_03.pdf, p. 40) — "the second process in the series, init-getty-login-shell, that ultimately connects a terminal user with UNIX" — copies issue (5) (ibid., cbunix_man5.pdf, p. 32) to the calling teletype before the prompt, as present-day, and sees a primary SYNOPSIS of:

-h suppresses the default behaviour of forcing "a hangup on the line by setting the speed to zero before"hand. It's unclear how this is supposed to work with DZ11s, given the above (it is: it doesn't, see above and note that RECV_ON is 010000 (bit 12)). line is the filename under /dev, speed is the gettydefs (5) (ibid., p. 26) entry, itself nothing more than the extraction of the built-in tables, and so tabulating the label, flags when reading the username (with hupcl -echo raw implied), final flags, speeds, the message, and the label to go to on BREAK. type describes "what type of terminal is connected to the line in question":

The interesting differences in the username adaptation mode are:

And

The BUGS mention that none of the characters mentioned above (or "'^D'") is quotable.

gtty(1) (CB-UNIX Programmer's Manual, cbunix_man1_03.pdf, p. 58) appears, and bears recalling in its entirety:

And these options, each taking a numeric argument:

"the terminal type" may also be specified, but "If this is done it must be the first option."; depending on the system configuration, the following may be meaningful (this is implied to not be an exhaustive list):

"When a terminal type other than »none« is in use, several more flags become valid:"

stty:o(2) (ibid., cbunix_man2_04.pdf, pp. 23-28) — where the :o denotes being for compatibility with "release 1 of CB-UNIX" (intro(2) (ibid., cbunix_man2_01.pdf, OLD C COMPILER, p. 4)) — is noted as superseded by ioctl(2) (ibid., cbunix_man2_02.pdf, pp. 1-7) but nevertheless contains a more cohesive overview of the system's general teletype system (though, notably, the mode layout and nomenclature outside of the core XTABS-EVENP spectrum are completely different between the two). Line disciplines 1, 2 are "reserved for project specific line disciplines", and [5, 9] "for future common line disciplines". Of those, 0 (STD_LTYPE) is the "normal" full-duplex discipline; 4 (HF_LTYPE) is the half-duplex line discipline — at last! — both of which "require no additional information". It's unclear how HF_LTYPE differs from setting HDPLX ("For those communicaitons controllers with the capability, disable reception during transmission.").

The interesing discipline is 3 (TRANS_LTYPE):

Backronymising, it may be that "[-]tstread" may actually be TranSparenT READ, otherwise it's unclear what it could be, aside maybe from a murky description of the equally-murky NOSLEEP:

stty:o(2) has an explicit NLDELAY mode, ioctl(2) has a range of CR[0123] NL[0123] TAB[01] FF[01] BS[01] with the would-be TAB2 bit defined as NOAL ("no allocation", presumably), since XTABS is unmoved from 2. "[-]xclude" corresponds directly to XCLUDE, described identically with addition of "Cleared upon last close.".

TANDEMI ("response to xon/xoff", [-]ixon) is contrasted with TANDEMO (but only in ioctl(2)) — "tranmission of xon/xoff": "turns off the keyboard when there are too many characters in the terminal hardware queue" — which itself is [-]ixoff.

erase kill are changeable, layout compatible with Sixth Edition. No other character is, hence STDTTY, which is exactly as-described (erase _ kill $ eol / eol2 !) with the addition of icrnl and a proverbial "kill2 &" (in addition to the primary DEL) (though, curiously, the flag itself does not (appear to, CB-UNIX Source Code, cbunix10.pdf, pp. 36-58 (tty.c), esp. ttyinput() (p. 45); the manual itself would suggest it does) imply the erase kill keyings, so those need to be set separately). This is likely [-]sccmod as Switching Control Center MODe.

ioctl(DIOCSETT) "is used to specify the type of CRT connected to a line":

/*
 * Terminal types
 */
#define	TERM_NONE	0	/* tty */
#define	TERM_TEC	1	/* TEC Scope */
#define	TERM_V61	2	/* DEC VT61 */
#define	TERM_V10	3	/* DEC VT100 */
#define	TERM_TEX	4	/* Tektronix 4023 */
#define	TERM_D40	5	/* TTY Mod 40/1 */
#define	TERM_H45	6	/* Hewlitt-Packard 45 */
#define	TERM_D42	7	/* TTY Mod 40/2B */

The video-terminal-specific modes are as such:

"vrow" is additionally noted as "for split screen operation". The actual semantics of this are unnoted.

ioctl(VIOC[GS]ETD) is described as "Get/set versatec parameters.". Votrax International, Inc. manufactured voice synthesisers, originally the VS4 "Votrax", and is attested as discipline 5 (VOT_LTYPE) (CB-UNIX Source Code, cbunix3.pdf, p. 95 (votrax.h); ibid., cbunix10.pdf, pp. 65-69 (vs.c)).

No use of TM_SNL is attested, especially in ibid., cbunix7.pdf, pp. 59-63 (ds40.c), which is supposed to host it.

The lost-media <sys/crtctl.h> is also shed light upon via the emulation therein:

char ds40map[] {
        LCA,    NONE,           /* Load Cursor Address (simulated) */
        CUP,    SDESC|'7',      /* Move Cursor Up */
        CDN,    SDESC|'B',      /* Move Cursor Down */
        CRI,    SDESC|'C',      /* Move Cursor Right */
        CLE,    010,            /* Move Cursor Left */
        NL,     NONE,           /* Special DS40 newline */
        CRTN,   SDESC|'G',      /* Return Cursor to start of line */
        HOME,   SDESC|'H',      /* Home Cursor */
        CS,     SDESC|'R',      /* Clear Screen */
        EEOP,   SDESC|'J',      /* Erase to end of page */
        EEOL,   NONE,           /* Erase to end of line */
        IL,     SDESC|'L',      /* Insert Line */
        DL,     SDESC|'M',      /* Delete Line */
        IC,     SDESC|'^',      /* Insert Char */
        DC,     SDESC|'P',      /* Delete Char */
        STB,    SDESC|'3',      /* Start Blink */
        SPB,    SDESC|'4',      /* Stop  Blink */
        STAB,   SDESC|'0',      /* Set Single    Tab */
        ATAB,   SDESC|'1',      /* Set Column of Tabs */
        CTAB,   SDESC|'2',      /* Clear Tabs */
        USCRL,  SDESC|'S',      /* Scroll Up */
        DSCRL,  SDESC|'Y',      /* Scroll Down */
        ASEG,   SDESC|'U',      /* Advance Segment */
/*      KBL,    NONE,           /* Keyboard Lock */
/*      KBU,    NONE,           /* Keyboard Unlock */
        UVSCN,  NONE,
        DVSCN,  NONE,
        0,
};

Where NL is handled as \33G\33B, regardless of any mode. Mayhap it was optional in a previous version of the system?

For comparison, for the VT100 (ibid., cbunix10.pdf, pp. 70-74 (vt100.c)), the tables alone are at least twice that.

A unique feature of CB-UNIX is ioctl(DIOCSETS):

UNIX User's Manual, Release 3.0

Rattles off 110, 150, 300 and 1200 baud as "Common terminal speeds", with 2400, 4800, and 9600 "occasionally" available. The phone number spiel is generalised to

A record low of "two ways to log out" is recorded, by losing direct-login (ibid., Logging out.).

The "On CB-UNIX only" note rings true — ^S/^Q are the only flow control method, but now in ixany mode (ibid., How to communicate through your terminal., para. 4).

The individual drivers are rich: dj(4) (ibid., p. 560) for DJ11s (all together now: incessantly called "DJ-11"s) — "Line speeds and other characteristics are not programmable but are set by switches in the hardware in groups of 4 lines. Only parameters such as character delays and mapping can be altered."† — dz(4) (ibid., p. 566) for DZ11s (dz), DZ11/KMC11s (dzk), and DH11s (dh) (yes.) — the DZ11 note is exactly the same as on CB-UNIX; "The behavior of dzk lines is indistinguishable from that of dz lines, except that on the dzk backspace delays are implemented using fill characters (rubouts) instead of timed delays.", which corresponds to hard [-]ofill based on the driver‡; "Note that the DH-11 is considered obsolete and is not supported on the VAX-11/780." — same old kl(4) (ibid., p. 571), though marked (PDP-11 only) in the header — on /dev/console, "Since they run at a constant speed, attempts to change the speed are ignored.".

init(8) (ibid., p. 681) distinguishes entries by arbitrary {state, id}, with no relation to line state; if no /etc/inittab (inittab(5), ibid., p. 618) is present or init is the only process, it runs a shell on /dev/console — this is described as more of a rescue state than a distinct single-user mode.

getty(8) (ibid., p. 674) sees a SYNOPSIS of

The text itself is a generalisation of the Sixth Edition into a mimickry of the Seventh Edition format. Rather than just the "Terminet 300 terminal" being put in full-duplex by the "login:" messages, they may additionally "take the DASI terminals out of the plot mode, or put a TELETYPE® Model 37 into full-duplex.". Returning "the GSI terminal to non-graphic mode" is lost.

The type table is compiled-in, but this is, at least, listed in the BUGS. A special (undisclaimed with the "(may vary locally)" for the others) type of ‘!’ type "tells getty to update /etc/utmp and exit". What this update does is unspecified. The types recognised in the distribution Release 3.0 are:

The source itself is derived from Sixth Edition (provided to me by my good friend Juliano: UNIX System III Source Code, https://vetusware.com/download/UNIX%20System%20III%20Source%20Code%20SYSIII/?id=11576, src/cmd/getty.c. Don't be a cop if you're affiliated with whoever bought the rights from SCO Group) and modes ‘-’ and ‘1’ agree completely, the ‘0’ loop is extended with the 1200-baud mode, and the new ‘3’ just starts there, the ‘2’ 9600-baud mode is replaced, all up from there is new.

The utmp(5) (Release 3.0, p. 631) processing is done as the first thing and consists of setting the entry for name to {current-time, ∅} in /etc/utmp and appending the same to /usr/adm/wtmp.

name is ignored if the standard input stream is a teletype, and the name of the attached teletype used instead; this is, presumably, to be used in conjunction with ‘!’, explained as

/* If the term type arg is special, exit only,
   this is so you can update utmp without spawning
   an open line
*/

tty(4) (Release 3.0, pp. 585-592) sees, fore-mostly, a fleshed-out controlling teletype and process group system: /dev/tty is "a synonym for the control terminal associated with the process group of that process, if any", and "The very first terminal file opened by the process group leader of a terminal file not already associated with a process group becomes the control terminal of that process group,", "A process can break this association by changing its process group using setpgrp(2)." — this corresponds directly to final present-day semantics and this is where nigh-ubiquitous techniques like double-forked daemonisation take their root.

"A line is delimited by a new-line (ASCII LF) character, an end-of-file (ASCII EOT) character, or an end-of-line character." — this fore-shadows an almost-normal interface, but one would be remiss to not note that EOT is End Of Transmission, hence why it wouldn't be outlandish for a teletype to hang up on its receipt.

‘@’ "optionally outputs a new-line character." and the table of "Certain characters" is re-tagged in terms of the logical actions, rather than the raw (now-default) values, and includes all characters with special meaning:

<sgtty.h> as the container for the structure used by the "primary calls" <termio.h>:

#define	NCC	8
struct	termio {
	unsigned	short	c_iflag;		/* input modes */
	unsigned	short	c_oflag;		/* output modes */
	unsigned	short	c_cflag;		/* control modes */
	unsigned	short	c_lflag;		/* local modes */
	char		c_line;		/* line discipline */
	unsigned	char	c_cc[NCC];	/* control chars */
};

The attentive reader may note that these are all as present-day, sans the lack of foreground/background process groups (signals are targeted as for SIGINT) and START/STOP always being stripped from the input (nowadays they're passed through if not IXON). The less-obvious flags are described below the original table; they are mostly found with mildly-evolved wording in IEEE Std 1003.1-202x (“POSIX.1”), Draft 2.1 , where IUCLC is lost. "The initial input control value is all bits clear.".
"The c_oflag field specifies the system treatment of output", and is outlined below.

They just don't miss! In IEEE Std 1003.1-202x (“POSIX.1”), Draft 2.1, OPOST is the only truly portable flag, all others are shaded XSI (X/Open Systems Interfaces — extension), and, symmetrically, OLCUC is lost (Release 3.0 even points out that "This function is often used in conjunction wiht IUCLC."). The entire description for the now-XSI parts lives verbatim in the standard. "The initial output control value is all bits clear.".
"The c_cflag field specifies the hardware control of the terminal", and is outlined below.

"For any particular hardware, impossible speed changes are ignored.", and note that this is a global (RX and TX) speed — this further supports the hypothesis that no-one has ever actually used uneven speeds.

The same deal as for c_iflag applies; the speeds are a separate table with specialised setters (see below), but are otherwise identical, except EXTA, EXTB are spelled as B19200, B38400 with the expected descriptions; IEEE Std 1003.1-202x (“POSIX.1”), Draft 2.1 has an additional explicit side-effect for CLOCAL as functioning as-if O_NONBLOCK: this is attested in the drivers and makes logical sense, but not laid forth. "The initial hardware control value after open is B300, CS8, CREAD, HUPCL.".
"The c_lflag field of the argument structure is used by the line discipline to control terminal functions. The basic line discipline (0) provides" flags outlined below.

Same deal here, but this is the first field that's missing flags found in IEEE Std 1003.1-202x (“POSIX.1”), Draft 2.1 (IEXTEN and TOSTOP). ISIG is, naturally, not yet SUSP-capable. The standard defines ECHOE as its function — erasing the last character — rather than the output sequence. ECHOK is allowed to do either [-] echoke mode — this system reasonably does -echoke since it doesn't have the flag. XCASE is lost.

The only weirdness here is that in the !ECHO ECHOE case, in Release 3.0 "the erase character is echoed as ASCII SP BS". It also says "Note that an escape character preceding the erase or kill character removes any special function." (and likewise for EOT); this is allowed for in modern standards, cf. IEEE Standard Portable Operating System Interface for Computer Environments, Approved August 22, 1988. ECHONL is noted to be "useful for terminals set to local echo (so-called half duplex).". "The initial line-discipline control value is all bits clear.".

The "lost" flags all last appear in Version 2 of the Single UNIX Specification (“SUSv2”) and are only defined for the POSIX locale. This is a bit of an issue when the kernel has to process them (alternatively, "mangle my preferred non-seven-bit-ASCII encoding").

The ISIG "special input functions may be disabled individually by changing the value of the control character to an unlikely or impossible value (e.g. 0377).". Admittedly, the common spelling is _POSIX_VDISABLE nowadays, and that value is usually handled explicitly, but yeah. There's also nothing preventing this working from other remappable values in ICANON, this is likely just suboptimal phrasing location.

"The MIN and TIME values are stored in the position for the EOF and EOL characters respectively.", and to this day IEEE Std 1003.1-202x (“POSIX.1”), Draft 2.1 requires that indices for all Special Control Characters are unique, "except that the VMIN and VTIME subscripts may have the same values as the VEOF and VEOL subscripts, respectively.".

Quite funnily, there is a marked loss of function against Seventh Edition, since START and STOP can't be remapped. There is also a provision for this (but a lesser one, in that they may just be ignored when setting, but must be otherwise provided).

Admittedly, there is also a loss of ioctl()s: since all parameters are integrated into one complete package, the struct tchars TIOC[GS]ETC requests are lost entirely, the new spellings (well, the old ones are still supported as <sys/ttold.h> compat shims, but)

TIOCHPCL is removed, since it's once again provided by HUPCL.

The system ships with only one line discipline, 0, imaginatively branded

/*
 * Line discipline 0
 */

A "synchronous terminal interface" is included as st(4) (ibid., p. 582), described as "a pseudo-device driver that enables a UNIX system to communicate with a TELETYPE® Model 40/0 ASCII synchronous terminal." utilising vpm(4) ("The Virtual Protocol Machine"; ibid., 594), itself a KMC11 (this time spelled correctly) driver and compiler toolchain that "translates a high-level language description protocol into an intermediate language that is interpreted by an interpreter running in the KMC11." — this machinery is entirely unrelated to the teletype driver and line disciplines, and driven entirely from user-space (though a default is provided the, cf. st(1M) (ibid., p. 343) and UNIX System III Source Code src/cmd/st/, in particular mod40.r therein; the corresponding st(4) driver lives in src/uts/pdp11/pwb/st.c and src/uts/vax/pwb/st.c, the former branded "Synchronous terminal interface for the PDP 11." and the port "TTY 40/4 driver for VAX 11/780").

ioctl(2) (Release 3.0, p. 446) is a stub even more so than the Seventh Edition one, giving only the prototype and errors.

login(1) (Release 3.0, p. 240) — "sign on" — is delightfully uneventful; passphrases may now expire, delightfully termed "password aging", and the "»dialup« password" becomes "a second »external« password", with a prompt of "External security:", and "If you do not complete the login successfully within a certain period of time (e.g., one minute), you are likely to be silently disconnected.", which goes hand-in-hand with getty's delay.

stty(1) (ibid., pp, 348-350) sees a SYNOPSIS of

It's odd that this is the first system noting that "many combinations of the options make no sense, but no sanity checking is performed". The settable options follow. ← indicates the argument is identical to present-day (⇐ with notes); → indicates the description is the same as in Seventh Edition (to the precision of whitespace, punctuation, and link to tty(4)).

The "134.5" spelling of "134" is retained, but undocumented.
"DATA-PHONE®" is a line of Bell modems. The only spelling change from "[-]hup" being trademarkisation of "dataphone" is, well.
[-]iuclc is predicated on iexten since its loss in Version 3 of the Single UNIX Specification (“SUSv3”).
"[-]lfkc" is the only psycho mode here. There's never been any mention of it anywhere before.
"[-]isig" is the same to the precision of Release 3.0 not having job control. Modern [-]echoe implementations are usually smarter than described and do process tabs normally, and \-escaping is thankfully lost. Embedded backspaces in -echoctl mode are usually sufficiently confusing, though.
"echok" corresponds to echok -echoke mode.
"evenp" and "parity" are documented as parenb cs7 but are actually also -parodd, so as present-day.
"ek" uses CERASE and CKILL macros, so it is the same, but the system defaults are different.
"sane" is -ignbrk brkint ignpar -parmrk -inpck istrip -inlcr -igncr icrnl -iuclc ixon -ixoff opost -olcuc onlcr -ocrnl -onocr -onlret -ofill -ofdel nl0 cr0 tab0 bs0 ff0 vt0 cs7 cread parenb -parodd -clocal isig icanon -xcase echo -echoe echok -echonl -noflsh and all Special Control Characters default. This corresponds to, going from Release 3.0 to present-day, a loss of ignpar -parmrk -inpck istrip ixon cs7 parenb -parodd -clocal -echoe and gain of -ixany imaxbel -iutf8 echoe echoke -tostop echoctl -echoprt -flusho iexten -extproc.
The terms are the same as in Seventh edition, but the short forms are lost.
Due to the VMIN/VTIME VEOF/VEOL overlap, "raw" (all Input Attributes clear, -opost cs8 -parenb -isig -icanon -xcase) sets min 1 time 1 (or, since no parsing provisions are made for min time, "min ^A time ^A") "-raw"/"cooked" (brkint ignpar istrip icrnl ixon opost cs7 parenb isig icanon) set eof eol to default. The latter two (sans cs7) is as present-day.

The Control Assignments table reads, for "control-character c":

It's interesting to note here that "^-" — 0377 (UNIX System III Source Code, src/cmd/stty.c) — is explicitly described as "undefined", rather than just "unlikely". No special provisions are made for 0xFF in the kernel, it is just very unlikely — enough to be used as an undefined value. The gain of "^?" matches that of Seventh Edition, Virtual VAX-11 Version, December, 1979. It's unclear what "escaped from the shell" means — ^ doesn't appear to be special to to sh(1) (Release 3.0, pp. 237-333).

The -g format is the same as this implementation's (Saved Attributes), and read back the same way.

The first line of the no-argument format starts with

speed 300 baud; evenp

speed 300 baud; -parity

The second line starts with "line = c_line;" (if not 0) and continues with all Special Control-Characters separated with semicolons, as, given intr ^- quit ^? erase ^E kill K eof ^@ eol 0xD8 (which is uninputtable, but):

intr <undef>; quit = DEL; erase = ^e; kill = K; eof = ^`; eol = -X;

The third line is the Input Modes, with defaults of -ignbrk -brkint inpck -ignpar -parmrk istrip -inlcr -igncr -icrnl -iuclc ixon ixany -ixoff (ignbrk hides the value of [-]brkint; -inpck – of [-]ignpar; -xon – of [-]ixany).

The fourth line is the Output Modes: either -opost, or flags in manual order, default off (except for ofill ofdel rendered as "del-fill" and ofill -ofdel rendered as "nul-fill") and 0 delays hidden. These (and two others, all listed below) are the only modes accepted as input but not noted in the manual.

The fifth line is the Local Modes, with defaults of isig icanon -xcase, [-]echo [-]echoe [-]echok always written, then of -echonl -noflsh,

The -a format is similar, but arranged differently: the first line lists the speed, line discipline, and Special Control Characters; the second the Control Modes, third – Input Modes (broken into fourth before [-]ixon), fifth – Local Modes, sixth – Output Modes, all in manual order with no aggregation.

UNIX User's Manual, Release 3.0, add'l bibliog.

†:
DJ11 asynchronous 16-line multiplexer maintenance manual, EK-DJ11-MM-003, 4th Printing, http://bitsavers.org/pdf/dec/unibus/EK-DJ11-MM-003_DJ11_Maint_Man_Aug76.pdf, p. 1-5, Digital Equipment Corporation, Maynard, Massachusetts, August 1976, General Description, 1.6 Specifications, Performance, Receivers and Transmitters, Data Rate.

‡:
This configuration is called out explicitly in the manual:

UNIX System User's Manual, System V; UNIX Administrator's Manual, Release 5.0

Yes, there's two of them, and yes, they're branded differently; this coincides with the Bell System break-up, and the User's Manual title page has a beautifully-crafted "we're kinda mad about this but don't wanna be openly super mad about it" blurb — UNIX System User's Manual, System V, Western Electric Company, Incorporated., 301-905 Issue 1, http://bitsavers.org/pdf/att/unix/System_V_Release_1/301-905_UNIX_System_V_Release_1_Users_Manual_Jan83.pdf, p. 1, January 1983.

The manual sections are helpfully rearranged: i.a., devices ("Special Files") now live in section 7 — UNIX System Administrator's Manual, Release 5.0, Bell Laboratories, Inc., 301-926 ISS. 1, http://bitsavers.org/pdf/att/unix/System_V_Release_1/301-926_UNIX_System_V_Release_1_Administators_Manual_Jun82.pdf, p. 3, June 1982, Introduction.

The haphazard assembly of random uploadable code is productised as stermio(7) (ibid., p. 234) — "general synchronous terminal interface" — which roughly simulates the normal teletype interface, alongside stgetty(1M) (ibid., p. 160) and stlogin(1) (UNIX System User's Manual, System V, p. 351). We are, thankfully, saved from a hypothetical ststty(1), but not scat(1) (ibid., p. 305).

login(1) (ibid., pp. 231-232)'s "»external« password" reverts to its CB-UNIX spelling (or, more likely, the cross-pollination goes the other way, or, likelier still, since CB-UNIX is noted as a source system for Release 3.0 and post-dates it, both).

The same holds for getty(1M) (UNIX System Administrator's Manual, Release 5.0, pp. 116-117), they're both dated March 13, 1981 (with header for "CB-UNIX 2.3") in the CB-UNIX Programmer's Manual, so it's best-described in terms of the differences from there. "[ -t timeout ]" "specifies that getty shold exit if the ope non the line succeeds and no one types anything in the specified number of seconds." — this is essentially a port of Release 3.0 delay. The type list is (unexpectedly?) shorter:

^U is also understood when reading the username, and kill adjusted. If the "»standard« ESS2" (note the odd updated branding) line editing characters are seen, "it arranges for this set of characters to be used for these functions.", rather than setting an exotic mode, since they all do map here, as intr & erase _ kill $ eol / eol2 ! — this checks out in the source, likewise provided to me by my good friend Juliano: UNIX SYSVr1 Source Code, https://vetusware.com/download/UNIX%20SYSVr1%20Source%20Code%20r1/?id=11570, sysv-pdp11_usr-src/cmd/getty.c.

dz(7) (UNIX System Administrator's Manual, Release 5.0, p. 199) is marked "(DEC only)"; the "(3B20S only)" tn4(7) (ibid., p. 246) and tn74(7) (ibid., p. 257) appear, identical except the former is eight-line and the latter – two-line: "The c_cflag items of B200, EXTA, and EXTB are not available.", which sounds an awful lot just like the DZ11.

tty(4) is split into tty(7) (ibid., p. 252) — "controlling terminal interface" — with the /dev/tty description paragraph and termio(7) (ibid., pp. 238-244) contains the rest, unchanged.

Of course, the header itself does change, if only to accomodate the STDTTY replacement — now <sys/termio.h>, VEOL2 is allocated as c_cc[6] (UNIX SYSVr1 Source Code, sysv-pdp11_usr-src/uts/sys/termio.h). Video-terminal-specific flags also appear — the list of terminals is taken verbatim, down to "Hewlitt", from CB-UNIX, as are the flags, and argument format (this is also copied verbatim, save for the new ioctl() names):

/*
 * structure of ioctl arg for LDGETT and LDSETT
 */
struct	termcb	{
	char	st_flgs;	/* term flags */
	char	st_termt;	/* term type */
	char	st_crow;	/* gtty only - current row */
	char	st_ccol;	/* gtty only - current col */
	char	st_vrow;	/* variable row */
	char	st_lrow;	/* last row */
};

stty(1) (UNIX System User's Manual, System V, pp. 355-357) is unchanged, save for the addition of some stermio(7)-specific modes. These are outlined below, but they aren't "actually" local, since that interface is much simpler (if insane).

More importantly, a copy-paste error that meant -lcase included -xcase but -LCASE didn't is fixed — (UNIX SYSVr1 Source Code, sysv-pdp11_usr-src/cmd/stty.c).

This does leave the system in an odd state where VEOL2 is supported (ibid., sysv-pdp11_usr-src/uts/pdp11/io/tt?.c), but only settable by getty, and not even observable with stty. This must be a fun mode to try to get out of if you trigger it by accident.

UNIX System V Release 2.0

The only way you'd know that these are AT&T System V Release 2 UNIX is by process of elimination, since that's the largest release number that appears on exactly one page out of the entire 530-page volume — Steven V. Earhart, UNIX Programmer's manual, AT&T, CBS College Publishing's UNIX System Library, Select Code 230-031, ISBN 0-03-009317-1, Volume 1: Commands and utilities, http://bitsavers.org/pdf/att/unix/System_V_Release_2/UNIX_Programmers_Manual_Vol_1_Commands_and_Utilities_1986.pdf, pp. 77-78, 1986, convert(1).

Rather than saying that @ "»kills«" all prior characters on the line, it now "cancels" them (ibid., Introduction, How To Get Started, How to communicate through your terminal, para. 3, p. vii).

stermio(7) and friends appear to be gone, and the individual driver pages are lost — Steven V. Earhart, UNIX Programmer's manual, AT&T, CBS College Publishing's UNIX System Library, Select Code 230-033, ISBN 0-03-009313-9, Volume 3: System administration facilities, http://bitsavers.org/pdf/att/unix/System_V_Release_2/UNIX_Programmers_Manual_Vol_3_System_Administration_Facilities_1986.pdf, pp. vii-viii, 1986, Table of Contents.

termio(7) (ibid., pp. 119-129) now has "SWTCH" allocated as c_cc[7] with no default listed, but the summary describes it as

ISIG and NOFLSH also govern swtch p.a. to intr quit.

A c_cflag LOBLK ("Block layer output.") is added, described as

The SEE ALSO defers to shl(1) (UNIX Programmer's manual, Volume 1, 1986, pp. 397-398) — "shell layer manager", no-argument SYNOPSIS,

There's a prompt, so this is naturally conversational. A summary of meaningful commands (for the full inglory, see the original) is included below, with their job control equivalents. For the purposes of this analysis, one could consider a ^Z here and ^Z, bg on a modern ("BSD-style") system the same fundamental operaition; things diverge from there.

Hidden in the create description is the blurb that "A maximum of seven layers can be created.". This is by far the worst "job control" scheme ever devised; its only "redeeming" quality is that sxt(7) (UNIX Programmer's manual, Volume 3, 1986, pp. 116-118) provides almost-general virtual teletypes.

getty(1M) (ibid., pp. 64-66) appears to lose both the ESS and ‘\b’/‘^U’ processing, but doesn't, it just stops being documented. Also undocumented is that it starts setting xcase for lines it deems upper-case-only — this is apparent in the source, similarly provided to me by my good friend Juliano: UNIX SYSVr2 Source Code, https://vetusware.com/download/UNIX%20SYSVr2%20Source%20Code%20r2/?id=11571, src/cmd/getty.c.

It's important to note that the manuals post-date the system by approximately two years — the kernel branding is for 1984 (ibid., src/uts/vax/os/main.c) and other, themselves unsourced, secondary sources agree.

login(1) (UNIX Programmer's manual, Volume 1, 1986, pp. 232-234) gains a fun facility where if the shell passwd(5) field is *, it chroots to the home directory and reexecs /etc/login,

The aforementioned swtch/eol[2] precedence is UNIX SYSVr2 Source Code, src/uts/vax/io/tt0.c, ttin(), if (flg) while (nchar--).

termio(7) Extends "or the timeout value TIME has expired" to "has expired between characters"; ibid. and tt[y1].c corroborate that this just a wording change.

Ibid., src/uts/sys/termio.h concretifies speeds:

#define	B19200	0000016
#define	B38400	0000017
			/* EXTA and EXTB should be deleted when */
			/* cleaned out of the source */
#define	EXTA	0000016
#define	EXTB	0000017
			/* they are aliases for b19200 and B38400 */

stty(1) (UNIX Programmer's manual, Volume 1, 1986, pp. 420-423) loses the stermio(7) references (but not the "st*" modes), gains the LOBLK proxy, below, adds SWTCH notes to "isig (-isig)", "noflsh (-noflsh)", and "raw (-raw or cooked)" and a swtch one to Control Assignments.

In "hupcl (-hupcl)" "a DATA-PHONE® data set connection" is reduced to "DATA-PHONE® connection".

The description for "clocal (-clocal)" starts with an 'n', for "n assume…". Unclear why.

UNIX SYSVr2 Source Code, src/cmd/stty.c lines up with this, except all mentions of LOBLK and VSWTCH are predicated on not being built for the PDP-11. Presumably due to memory constraints? But enforcing this here is odd.

The default for no-argument output is "-loblk".

UNIX® System V User's Reference Manual; AT&T System V.3 Administrator's Reference Manual

In this case the only way to figure out that this 350-page volume is for AT&T System V Release 3 UNIX is in the SEE ALSO section in UNIX® User's Reference manual, Prentice-Hall C and UNIX® Systems Library, ISBN 0-13-940487-2 025, http://bitsavers.org/pdf/att/unix/System_V_Release_3/UNIX_System_V_Users_Reference_Manual_1987.pdf, pp. 116-117, AT&T, 1987, 1986, getopts(1).

The "valid user name" requirement is changed to "valid login id" — the purpose of this is unclear, unless the editor was smoking on that psychoanalysis mid — and the "common"/"occasionally available" dichotomy is hegelised by commonalising all of them and removing 110, 150, and 300 baud, all of which between ten- and four-fold slower than the slowest listed, now 1200, 2400, 4800, and 9600 baud (ibid., Introduction, How To Get Started, Logging In, paras. 1-2, p. 4).