Network Working Group M. Padlipsky
Request for Comment: 666 26 November 1974
NIC: 31396
Specification of the Unified User-Level Protocol
After many discussions of my RFC 451, I discovered that the "Unified
User-Level Protocol" proposed therein had evolved into what had
always been its underlying motivation, a common command language.
There are several reasons why this latter approach satisfies the
original goals of the UULP and goes beyond them into even more useful
areas:
1. User convenience. As evidenced by the good response to the common
editor "neted", the Network Working Group has come to acknowledge the
fact that the convenience of non-system programmer users of the
Network must be served. Allowing users to invoke the same generic
functions -- including "batch" jobs -- irrespective of which Server
Host they happen to be using is surely a compelling initial
justification for a common command language. Note that the concern
with generic functions -- which "all" Servers do, one way or another
-- is intended to emphasize the common command subset aspects of the
language, rather than the "linguistic" elegance of it all. The
attempt is to specify an easy way of getting many things done, not a
complicated way of getting "everything" done.
2. "Resource sharing". Another area which is receiving attention in
the NWG of late is that of "automatic" or program-driven invocation
of resources on foreign systems. A common intermediate
representation of some sort is clearly necessary to perform such
functions if we are to avoid the old "n by m problem" of the Telnet
Protocol -- in this case, n Hosts would otherwise have to keep track
of m command languages. For the common intermediate representation
to be human-usable seems to kill two birds with one stone, as
expanded upon in the next point.
3. Economy of mechanism. In RFC 451, I advanced the claim that a
single user-level protocol which connected via socket 1 and Telnet
would offer economy of mechanism in that new responders would not be
required to service Initial Connection Protocols on socket after
socket as protocol after protocol evolved. This consideration still
applies, but an even greater economy is visible when we consider the
context of resource sharing. For if the common command language is
designed for direct employment by users, as the present proposal is,
there is no need for users on terminal support "mini-Hosts" (e.g.,
ANTS and TIPs) to require an intermediary server when all they
actually want is to work on a particular Server in the common
Padlipsky [Page 1]
RFC 666 Unified User-Level Protocol November 1974
language. (This is especially true in light of the fact that many
such users are not professional programmers -- and are familiar with
no command language.) That is, if resource sharing is achieved by an
intermediate language which is only suitable for programs, you would
have to learn the native command language of Server B if you didn't
want to incur the expense of using Server A only to get at generic
functions on Server B. (And you might still have to learn the native
language of Server A, even if the expense of using two Servers where
one would do isn't a factor.)
4. Front-ending. Another benefit of the common command language
proposed here is that it is by and large intended to lend itself to
implementation by front-ending onto existing commands. Thus, the
unpleasant necessity of throwing out existing implementations is
minimized. Indeed, the approach taken is a conscious effort to come
up with a common command language by addition to "native" command
languages rather than by replacement, for the compelling reason that
it would be unworkable as well as ill-advised to attempt to legislate
the richness represented by existing command languages out of
existence. Further, as it is a closed environment, no naming
conflicts with native commands would arise.
5. Accounting and authentication. As evidenced by the spate of RFCs
about the implications of the FTP in regard to both accounting for
use of Network services and authenticating users' identifications
(Bressler's RFC 487, Pogran's RFC 501, and my RFC 505 -- and even
491), this area is still up in the air. The generic login command
proposed here should help matters, as it allows the Server to
associate an appropriate process with the connection while actuating
appropriate accounting and access control as well, if it chooses.
6. Process-process functions. By enabling the invocation of foreign
object programs, the present proposal offers a rubric in which such
process-to-process functions as "parallelism" can be performed. (See
the discussion of the "call" command, below.) Note that the UULP is
not being advanced as a panacea: It is assumed that the actual
transactions carried out are most likely not going to be in the
common command language (although some certainly could be); however,
what is furnished is a known way of getting the presumably special-
cased programs executing elsewhere. Also, it offers a convenient
environment into which can be placed such new functions, which we
would like to have become generic, as Day's File Access Protocol.
All of which seems to be a fair amount of mileage to get out of a
distaste for remembering whether you find out who's logged in by
saying "systat", "users", "s.who:c", "listf tty", or "who"....
Padlipsky [Page 2]
RFC 666 Unified User-Level Protocol November 1974
Context
Although ultimately intended to become the general responder to the
Initial Connection Protocol, the UULP is initially to be a Telnet
Protocol "negotiated option". When the option is enabled, the Server
Host will furnish a command environment which supports the common
conventions and commands discussed herein.
In a sense, the UULP is a "selector". That is, the common command
subset includes commands to exit from the common command environment
and enter various other environments, along the lines of CCN's
current Telnet Server. To exit from the UULP environment to the
"native" command processor, the UULP command is "local" (see also the
discussion of Case, below). Note that all commands terminate in
Telnet "Newline" (currently cr-lf), unless altered by the "eol"
command (below); internal separator is space (blank). (Entrance into
other environments -- such as the FTP Server -- is discussed below.)
There are two reasons for introducing a mechanism other than the
apparently natural one of simply de-negotiating the option: First, it
is bound to be more convenient for the user to type a command than to
escape to his User Telnet program to cause the option disabling.
Second, it is hoped that eventually the UULP will be legislated to be
the default environment encountered by any Network login, in which
case the natural way to enter the Server's "native" command
environment would be by UULP command.
Note: all UULP commands discussed herein are listed in Appendix 1,
categorized as to optionality, with brief descriptions given. The
appendix may be taken as a first-pass UULP Users' Manual.
Responses
Any optional commands which are not supported by a particular Server
are to be responded to by a message of the form "Not implemented:
commandname.", where the variable is the name of the command which
was requested. Note that throughout this document, all literals must
be sent exactly as specified, so as to allow for the possibility of
Servers' being driven by programs (including "automata" or "command
macros") in addition to "live" users.
In general, the view has been taken here that a small number of
literal, constrained responses is superior to a vast variety of
numerically coded responses in which text may vary. Again, the
motivation is to achieve an economy of mechanism. For on the coded
model, there must be a coordinator of code assignments, which is just
as well avoided. Further, as has been experienced in the use of the
FTP, when there are many codes there are many ambiguities. (The
sender may have a perfectly valid case for choosing, say, 452, while
Padlipsky [Page 3]
RFC 666 Unified User-Level Protocol November 1974
the receiver may have an equally good interpretation of the codes'
definitions for expecting, say, 453.) Experience with a related
"error table" mechanism on Multics also bears out the assertion that
coded responses create both managerial and technical problems. A
final objection to numeric codes might be considered irrelevant by
live some, but I think that the aesthetics of the situation do merit
some attention. And when the common command language is being
employed by live users, it seems to me that they would only be
distracted by all those numbers flying around. (Nor can we assume
that the numbers could be stripped by their "User UULP", for one of
the basic goals here is to make it straightforward enough for a user
at a TIP to deal with.)
Arguments
During the review process, it became evident that some global
comments on arguments were in order. Two areas in particular appear
to have led to some confusion: the strategy of specification of
arguments on the command line, and the question of "control
arguments". On the first score, the goal of "front-endability" must
be recalled. Consider two native implementations of a particular
command, one of which (A) expects to collect its arguments by
interrogation of the user, and the other of which (B) expects to
receive them on invocation (being invoked as a closed subroutine).
Now, it is easy to imagine that a "Server UULP" could feed the
arguments to A as needed without requiring A to be rewritten, but it
is quite difficult to see how B could be made to interrogate for
arguments without extensive rewriting. Therefore, a "least common
denominator" approach of specifying arguments in advance incurs the
minimum cost in terms of reworking existing implementations.
On the second score, I have borrowed a notion from the Multics
command language's convention called "control arguments" because it
seems to be quite convenient in actual practice. The key is that
some arguments are meant as literals, usually specifying a mode or
control function to the command, while others are variables,
specifying something like a particular file name or user identifier.
A common example is a "mail" command, where the variables are the
user identifiers and the Host identifiers, and the "control argument"
is the designator that user identifiers have ceased and Host
identifiers have begun. The convention used here is to begin the
control argument with a hyphen, as this character never seems to be
used to begin variable arguments. Thus, we use "-at" in the mail
example. Although it is not a deep philosophical point, this
approach does relieve argument lists of order-dependency, and feels
right to me.
Padlipsky [Page 4]
RFC 666 Unified User-Level Protocol November 1974
Case
Although it appears to have been legislated out of existence by the
specification of the Network Virtual Terminal's keyboard in the
Telnet Protocol, the question of what to do about users at upper-
case-only terminals remains a thorny one in practice. There are two
aspects to consider: the alphabetic case of commands, and the ability
to cause "case-mapping" in order to allow lower-case input. Some
Servers have no local problems with the first aspect, as they operate
internally in all upper-case or all lower-case and merely map all
input appropriately. (Problems do arise, though, when one is using
the User FTP on such a system to deal with a mixed-case system, for
example.) Other Servers, however, attach the normal linguistic
significance to case. (E.g., Smith's name is "Smith" -- not "SMITH",
and not "smith".) To minimise superfluous processing for those
Servers which are indifferent to case, all UULP commands are to be
recognized as such whether they arrive as all upper-case or all
lower-case. (They will be shown here as all lower merely for typing
convenience.) Note that arbitrarily mixed case is not recognized, as
it is an unwarranted assumption about local implementation to suppose
that input will necessarily be case-mapped.
On the second aspect, any Server which does distinguish between
upper- and lower-case in commands' arguments (a.k.a. parameters) must
furnish a UULP "map" command as specified in Appendix 2 in order to
support logins from upper-case-only terminals attached to User Hosts
which either do not support the Telnet Protocol's dictum that all 128
ASCII codes must be generable, or support it awkwardly. This seems a
simpler and preferable solution than the alternative of legislating
that upper-case Network-wide personal identifiers (and perhaps even
Network Virtual Path Names) be pre-conditions to a usable common
command subset. (As noted below, these latter concepts will fit in
smoothly when they are agreed upon. The point here, though, is that
we need not deprive ourselves of the benefits of a UULP until they
are agreed upon.)
User Names
As implied above, the various Servers have their various ways of
expressing users' names. Clearly, the principle of economy of memory
dictates that there should be a common intermediate representation of
names in and for the Network. It is probably also clear that this
representation will be based upon the Network Information Center's
"NIC ID's". However, it is unfortunately amply clear than an
acceptable mechanism for securing up-to-date information cannot be
legislated here - much less a mechanism for securely updating the
implied data base. Therefore, at this stage it seems to be the
Padlipsky [Page 5]
RFC 666 Unified User-Level Protocol November 1974
sensible thing to specify only the UULP syntax for conveying to the
Server the fact that it is to treat a user name as a Network-wide
name rather than as a local name, and let the supporting mechanisms
evolve as they may.
The prefacing of a name with an asterisk ("*") denotes a Network-wide
name. (Such names may be either all upper-case or all lower-case, as
with UULP commands' names.) The name "*free" is explicitly reserved
to mean that (in the context of logging in) a login is desired on a
supported or sampling account, if such an account is available. The
response if no such account is available is to be "Invalid ident:
*free." When Network-wide names are generally available Servers will
either map them into local names or cause them to be registered as
local names as they prefer. The point is that a Network-wide name
will be "made to work" by the Server in the context of the UULP.
Special Characters and Signals
Another area in which the facts of life must outweigh the letter of
the Telnet Protocol if the user's convenience is to be served is that
of "erase" and "kill" characters. It is possible that User Telnets
will uniformly facilitate the transmission of the Telnet control
codes for generic character erase and generic line kill. It is
certain, however, that User Telnets will differ -- and users will, if
they use more than one User Telnet, be again placed in the
uncomfortable position of having to develop too many sets of
reflexes. Therefore, the UULP will optionally support the following
commands: "erase char" and "kill char", where char is a printable
ASCII character (to avoid possible conflicts with "control
characters" which are recognized in the innermost areas of particular
operating systems). Presumably, unwary users can be instructed not
to choose an alphabetic, so as to avoid being placed in a position
where they cannot invoke certain commands (erase and kill themselves,
for example, in which case they couldn't be changed).
These commands are supplements to the related Telnet control codes,
and have the same meanings. The point here is that it may be far
more convenient for a user to be able to say "erase #" and get the
"#" to be recognized as the erase character by the Server than for
the user to get his User Telnet to send the Telnet equivalent. The
commands are designated as optional because they may lead to severe
implementation problems on some Servers, and because the equivalent
functions do, after all, exist in Telnet.
Note: the erasing is assumed to be performed "as early as
possible". That is, the sequence "erase x" "erase x" should come
out equivalent to "erase x" "erase" -- the second appearance of
Padlipsky [Page 6]
RFC 666 Unified User-Level Protocol November 1974
"x" resulting in the erasing of the space in the command line.
Presumably, this is a sufficiently uncommon path that anomalous
results would be tolerated by the user community, but the intent
ought to be clear.
The Telnet "synch" and "break" mechanisms are, by their very nature,
best left to Telnet. End of line, however, might well be a different
story. Therefore, as a potential convenience, the UULP optionally
supports "eol char" to ask the Server to treat char as the end of
line character thenceforth. To revert to Telnet Newline, "eol"
(i.e., no argument, current terminator).
Prompts
Another aspect in which Servers vary while being the same is how they
indicate "being at command level". Some output "ready messages";
others, "prompt characters". For the UULP, where some functions will
be performed by means of a command's logging in to another system,
the ability to specify a known prompt character is extremely
desirable. The UULP command is "prompt char" where char is the
character which is to be sent when the user's process (on the Server)
is at command level. It is explicitly permitted to prefix char to a
line consisting of a "native" prompt or ready message. Also, this
command is explicitly acknowledged to be permissible prior to login.
(Again, warning must be made of the bad results which can ensue if an
alphabetic character is chosen.)
Note: "prompt", "eol", "erase", and "kill" may all be re-invoked
with a new value of char in order to change the relevant setting;
all may be turned off by invocation with no argument.
Login
Perhaps the stickiest wicket of them all is the attempt to specify a
generic login, but here we go. The UULP login command is "login
userident", where userident is either a locally-acceptable user
identifier or a Network-wide identifier as discussed above. Note
that for utility in contexts to be discussed later, the locally-
acceptable form must not contain spaces. Servers may respond to the
login attempt with arbitrary text (such as a "message of the day"),
but some line of the response must be one of the following: a prompt
(as discussed above; indicating, in the present context, successful
login); "Password:"; or "Invalid ident: userident." When passwords
are required, it is the Server's responsibility either to send a mask
or to successfully negotiate the Hide Your Input option.
Padlipsky [Page 7]
RFC 666 Unified User-Level Protocol November 1974
Note that "login *free" is specifically defined to require no
password. (If a "freeloader" has access to a User Telnet and has
learned of the "*free" syntax, it is fruitless to assume that he
couldn't have also read the common password.) If a password must be
given, acceptable responses are arbitrary text containing a line
beginning either with a prompt or with "Login unsuccessful." or with
"Account:". If an account is requested, the responses must be either
the "Login unsuccessful" message or the text containing a prompt
already described. If any errors occur during the login sequence,
users are to re-try by starting from the login command. (I.e., it is
not required that the Server "remember" idents or passwords.)
It is explicitly acknowledged that an acceptable response to "login
*free" is "Limited access only." (followed by a prompt). This is
intended to warn (human) users that the free account on the Server in
question exists only to allow such functions as accepting mail and
telling if a particular user happens to be logged in. (For
objections to "loginless" performance of such tasks, see RFC 491.
Note also that nothing here says that a Server must do anything other
than return a prompt in response to "login *free" in the event that
loginless operation is natural to it.) Given the UULP login
discipline and the "prompt" command, it is reasonably straightforward
for a program to login on a free account and perform one of these
functions, for if the login command succeeded, the program will "see"
a guaranteed prompt character.
To make life simpler for those Hosts which normally have some sort of
"daemon" process service mail and the like, a further expansion to
login is in order. The point here is that some Hosts may not know
what sort of process to pass an unqualified "login *free" to, whereas
they'd be sure what to do with an explicit request to process mail,
do a who command, or set up console to console communications.
Therefore, UULP "login" will allow a "control argument" (as discussed
above) of either "-mail", "-who", or "-concom", and the respective
UULP commands involved must use the respective strings in any login
line they transmit. Again, nothing is being said about what a Server
has to do with the information, but some Servers need/want it.
Usage Information
Most Servers offer some sort of on-line documentation, from calling
sequences of commands to entire users' manuals. There are two sorts
of information of interest in the UULP environment: "normal" system
information, and information about the particular Server's UULP
implementation. To learn how to get descriptions of "native"
commands, the UULP command is "help -sys" (abbreviation: "?"). Note
that "-sys" is viewed as a "control argument" and as such prefaced by
Padlipsky [Page 8]
RFC 666 Unified User-Level Protocol November 1974
a hyphen ("-") to facilitate distinction from other sorts of name
(e.g., command names). To get a description of the Server's UULP
implementation, "help -uulp". To get a description of a particular
UULP command's implementation, "help comname". To be reminded of how
to use the help command, "help".
Note: as with command names and Network-wide user names, control
arguments may be either all upper-case or all lower-case.
It is specifically acknowledged that "No peculiarities." is an
appropriate response to "help comname" if nothing of interest need be
said about the Server's implementation of the UULP command in
question. (After all, we're sparing users the necessity of studying
a dozen or so users' manuals; the least they can do is to read the
UULP command list.) Appropriate information for less taciturn Hosts
to furnish would be such data as local command invoked (if such be
the case), argument syntax (e.g., pathname description, or name of
help file about pathnames), "To be implemented.", or even "Not to be
implemented."
"Mail"
Even though a separate mail protocol is being evolved for general
purposes, the UULP needs to address this topic as, by virtue of being
login based, it allows systems which do access control and sender
authentication on mail to make these abilities available to users
within its framework of generic functions. Therefore, to read one's
mailbox, the UULP command is "readmail". To have "live" input
collected and sent to a local user, "mail userident"; to a remote
user, "mail userident -at hostname", where the arguments have the
"obvious" meanings. To send a previously-created file, "mail -f
filename userident -at hostname". Several useridents may be
furnished; the delimiter is space (blank). Similar considerations
apply to hostnames. If both are lists, they sould be treated
pairwise. (A more elaborate syntax could be invented to deal with
the desire to send to several users at a given host and then to other
users at other hosts, but it seems unnecessary to do so at this
point, for multiple invocations would get the job done.)
The mail command prefaces the message with a line identifying the
sender (Host and time desirable, but not mandatory). For "live"
collection, the end of message is indicated by a line consisting of
only a period (".") followed by the regnant line terminator (usually
the Telnet Newline, but see also the discussion of the eol command).
If remote mail is not successfully transmitted, it is to be saved in
a local file and that file's name is to be output as part of the
failure message. ("Queueing" for later transmission is admired, but
Padlipsky [Page 9]
RFC 666 Unified User-Level Protocol November 1974
not required.) The transmission mechanism will follow the general
mail protocol. Note that when invoked with a "-at" clause, the mail
command will send "login *free -mail" to the remote Host(s), followed
by a mail command with no "-at" clause.
A desirable, but not required, embellishment to "readmail" would be
the accepting of a Host name ("-at hostname") to cause the local Host
to go off to the named Host (via "login *free -mail") and check for
mail there. Several hostnames could, of course, be specified. A
further embellishment, which would probably be quite expensive, would
be to accept "-all" as a request to check all Hosts (or, perhaps, all
Hosts known to have a free account for the purpose) for mail.
Direct Communication
The ability to exchange messages directly with other logged in users
is apparently greatly prized by many users. Therefore, despite the
fact that there is a sense in which this function is not within the
purview of the UULP, we will address it, after a digression.
Digression: The UULP assumes that there can be straightforward
"front ends" at the various Servers which translate generic
function calls in a common spelling to calls for specific, pre-
existing "native" functions. In the area of console to console
communications, however, this premise does not really hold. The
problem is that both major "native" implementations known to the
author are seriously flawed. The TENEX "link" mechanism is both
insecure (you've got no business seeing everything I type even if
I'm careless enough to let you) and inconvenient (why should I be
forced to remember that pesky semi-colon? how do I get back into
phase after I've forgotten one?). It is also likely to be
extremely difficult to simulate on systems which do not force
Network I/O through local TTY buffers, even if the user interface
were not subject to criticism. The Multics "send_message"
mechanism, on the other hand, has a more sophisticated design, but
is absurdly expensive. Therefore, the UULP mechanism to be
described assumes that, for this function, new local
implementations will be developed to support it.
To permit console to console communications: "concom -on"; to refuse,
"concom -off". Default is off. To enter message-sending mode:
"concom userident -at hostname" ("-at" clause is optional). To exit
from message-sending mode, type a line consisting of only a period
(cf. Mail, above). While in message-sending mode, each line will be
transmitted as a unit. The first message sent by concom must be
prefaced by an identifying line, beginning "From:" and containing an
appropriate address to which to reply. The closing period-only line
Padlipsky [Page 10]
RFC 666 Unified User-Level Protocol November 1974
should be transmitted, so as to allow the other concom to close as
well. Acceptable error response is "Not available: userident."
(which neither confirms nor denies the existence of the particular
user -- a matter of concern on the security front). The command
must, of course, do whatever is necessary to transmit the messages;
i.e., if locally invoked, access the local mechanism, and if invoked
for remote communications, access the remote Host's concom command
(via "login *free -concom"). Thus, a user at a TIP would use the
local form of concom on the Host of the other party if this is
convenient, or would use the remote form on his "usual" Server if the
direct use is inconvenient for some reason (such as having no account
there, say).
The prerequisites for establishing communications are to find out if
the user is logged in, and what "address" to use if so. The
mechanism for gathering this information is an expanded "who"
command. (Note that "who" is the UULP command to invoke the generic
who's logged in function, with no constraints on format of reply.)
The syntax is "who userident -at hostname", where both arguments may
be multiple. If no "-at" clause, then check local Host only.
Response must begin "From hostname: userident:" followed by either an
appropriate address (e.g., "ll" if local "concom" uses TTY numbers
and userident is logged in on TTY ll), or "Not available."
As with mail, a "-all" embellishment might be pleasant. Note that
the search for the specified user(s) -- whether or not "-all" is used
-- still assumes that a "login *free -who" login will be used on the
appropriate remote Host(s), followed by "who userident". This is why
responses to the expanded who command must be so rigidly specified.
Note also that regardless of whether the inquiry is made in terms of
Network-wide or local user name, the response must be appropriate for
use in "concom".
"Good" concom implementations will presumably do an expanded who
command automatically, so as to spare the user the necessity of
having to do it separately. Indeed, the -concom control argument to
login is defined to imply the ability to do a who as well as a concom
to cater to this possibility. It is tempting to legislate that such
an approach be the rule, but the implementation implications are not
quite clear enough to do so. The implicit who should be viewed as a
strong hint to implementers, though.
File Creation and Manipulation
The common command subset must furnish the ability to create and
manipulate files. Creation is necessary in order to send mail on the
one hand, and to produce source files for subsequent compilation on
the other hand. Manipulation (such as copying, renaming, typing out,
Padlipsky [Page 11]
RFC 666 Unified User-Level Protocol November 1974
and the like) is necessary both as a convenience aspect for users who
seek to operate only in the common command language and as a means of
performing desired batch functions (see below). For file
manipulation commands, the user could enter the File Transfer
Protocol environment. However, the FTP user interface is constrained
by a very high degree of program-drivability. It is also lacks
abbreviations and suffers from the lack of mnemonicity dictated by
limiting command names to four characters. Further, some valuable
functions (such as causing a file to be typed out) are not dealt
with. Therefore, various UULP file manipulation commands are given
in Appendix 1. They need not be addressed in detail here. However,
some context would be useful:
The file manipulation commands assume that all Servers have some
notion roughly corresponding to "the user's working directory". All
file names, whether the yet to be invented Network Virtual Pathname
or the "local" variety, are taken to refer to files in this directory
unless otherwise indicated. That is, the user should not have to
furnish "dsk:" or the like; it is taken as given that when he refers
to file "x" he means "the file named 'x' in my current working
directory" and the Server "knows" what that means.
At the present stage of development of the UULP, it does not seem
fruitful to go into a reasoned explication of the following
statement. For now, suffice it to say that those file manipulation
commands (a copy of a foreign file, for example) which need to employ
the FTP do employ the FTP and let it go at that. As the context and
implications of the protocol become more widely understood, the
detailed implementation notes will be added to the file commands --
and refined for the other commands, doubtless. In a way, the common
file commands may be viewed as a kind of "User FTP" of known human
interface when they deal with foreign files. (And, of course, until
there's a Network virtual pathname, the issue doesn't really arise.)
I expect that an "identify" command might be desirable, so that UULP
commands which have to access other Servers in turn on behalf of the
specific current user can have the necessary login information
available to them. Such a command is included in Appendix 1, but
should rank as speculation for now.
On the topic of file creation, matters are rather complicated. It is
clear that the ability to create files in the UULP environment is
extremely desirable. It is also clear that using mail to a fake
address to get the file created, then renaming the "unsent mail" file
is too byzantine to expect users to do. Unfortunately, it is not
clear exactly what the alternative is. That is, it's fairly clear
that we need a common editor, but it's not at all clear which editor
it should be.
Padlipsky [Page 12]
RFC 666 Unified User-Level Protocol November 1974
Two widely-known editors come to mind: TECO and QED. However, not
everybody has them. Even if everybody did, the "dialects" problem is
bound to be a large one. Even if all the relevant system programmers
could agree, there remains the question of whether the intended user
population would be willing to bother learning a language as complex
as TECO or QED. Therefore an optional UULP command to be called
"neted" is proposed. (See also RFC 569.) This editor is a line-
oriented context editor (no "regular expressions", but also no line
numbers). It is copiously documented in Chapter 4 or the Multics
Programmers' Manual, including an annotated listing of the (PL/I)
source code. A simple user's guide has been prepared (see Appendix
3). Several implementations already exist, and commitments have been
made for more. It may also be repugnant to some of the system
programmers who would be called upon to implement it -- which is why
it is optional, until and unless higher authority makes it mandatory.
Other Protocols
The nominal initial impetus for proposing a UULP was to allow new
Network user protocols to be invokable through a common mechanism,
rather than requiring a new responding mechanism to be built for a
new contact socket for each new protocol. Although this goal has
been shunted into the background by the admission of the true goal of
the UULP, it has not been dropped completely. Therefore, to enter
the FTP Server environment, the UULP command is "ftp"; to enter the
RJE Server environment, the UULP command is "rje". Exit is as per
the respective protocols. (Where possible, exit should be back to
the UULP environment.)
Invoking Foreign Programs
There are two broad contexts in which it is desirable to cause a
specific local program to be invoked from the common command
environment: The User side of the connection may itself be a program,
and the desired Server side program a specifically cooperating one;
this is the more sophisticated context, of course. The less
sophisticated context assumes that the User side is a "live" user,
and the desire is to invoke a compiler or an object program the user
has already compiled in the common language -- again as a convenience
to the user so that he may operate in a sort of "Server-transparent"
mode. (The latter case also covers "batch" use of the Server; see
below.) In both contexts, the important role of the UULP is to
specify the mechanisms through which the particular programs may be
invoked, irrespective of the idiosyncrasies of the Servers' command
languages.
Padlipsky [Page 13]
RFC 666 Unified User-Level Protocol November 1974
Programming languages are much too big a problem to tackle here.
However, assuming that a user somehow manages to create a source
program, he still wants some commonality of spelling in invoking the
appropriate compiler, or even the object program. As an optional but
strongly recommended UULP command, then, "call name" should invoke
object program name (where the named program may be a "native"
command with arguments specified as appropriate). The values "pl1",
"-basic", "-fortran", "-lisp", etc., should be recognized as
requesting the invocation of the appropriate language processor (to
operate on a named source file or interpretively/interactively if no
source file was named), with "reasonable" defaults in effect. Note
that this all is meant to imply that "native" commands are not
directly invokable from the UULP environment (other than by "call"),
to avoid potential naming conflicts between system commands and new
UULP commands.
Note that the "call" command in the UULP environment constitutes a
rubric for "parallel" computation, given any ad hoc convention for
the return of completion information. (Writing on the Telnet
write socket plus 2 would seem appropriate, provided the initiator
has the ability to "listen" for the rfc; but even a response in
the data stream as a special-cased program is assumed on the
"user"side anyway.)
Other Matters
The topic of "batch" mode merits some attention. As with the file
manipulation commands, more consultation is necessary for a firm
spec. However, I suspect that a "-batch" control argument to login
should initiate batch mode processing by the Server, and given the
call and identify commands all we might then require is a convention
for designating the output file in order to return it via a copy
command in the "job" itself (if output is to be returned rather than
stored at the Server). Of course, -batch will probably need some
substructure as to password and timing matters. More details will
emerge in this area in future iterations.
An admittedly fictionalized scenario might look like this:
login Me -batch -pw xxx -shift 3
copy *452<me>source.text source.pl2
call -pl2 source
call source input output
identify Me2 yyy
copy output *555>root>Me>output452
logout
Padlipsky [Page 14]
RFC 666 Unified User-Level Protocol November 1974
where user "Me" wants the Server receiving the commands (either
directly from him at a TIP or perhaps from some other Server on which
he has created a file containing them) to set up a batch job for him,
with password "xxx", to be run on Shift 3 (whenever that is). The
job first copies file "source.text" from directory "<me>" on Host 452
into local file "source.pl2", then compiles it with the local PL2
compiler, executes it (assuming a "Not found" response would go into
a known file if compilation had failed) with specified arguments
(presumably the names of files for input and output), then copies the
"output" file to Host 555's file hierarchy at the indicated place,
using the user identifier "Me2" and the password "yyy". It's not
elegant, but it ought to work.
Finally, on the topic of logging out, the UULP command is "logout".
The Server must close the Telnet connection after doing whatever is
appropriate to effect a logout. To retain the Telnet connection,
"logout -save". Having the Server close is viewed as a convenience
for the user, in that it spares him the necessity of causing his User
Telnet to close. It is also desirable for program-driven
applications, so as not to leave the connections "dangling" and not
to require possibly complex negotiations with the User side to break
the connection.
APPENDIX 1. THE COMMON COMMAND SUBSET
Syntax Opt
I. "Set-up" Commands
login id arg
The id may be Network-wide or Host-specific.
"*free" is reserved.
The arg may be "-mail", "-who", "-concom",
"-batch", or may be absent.
Result is to be either logged in or passed off to appropriate daemon.
prompt char
Specifies that char is to become or
precede the normal prompt message.
Acceptable prior to login.
erase char X
Specifies that char is the erase character.
Invocation with no argument reverts to default.
kill char X
Padlipsky [Page 15]
RFC 666 Unified User-Level Protocol November 1974
Specifies that char is the kill character.
Invocation with no argument reverts to default.
eol char X
Specifies that char is the newline character.
Invocation with no argument reverts to default.
local
Enter the local command environment.
ftp
Enter the FTP environment.
rje
Enter the RJE environment.
logout
Logout and sever the Telnet connection.
logout -save
Logout but keep the Telnet connection.
map
Apply the case-mapping conventions of Appendix 2.
Required on Hosts to which case is significant.
identify id arg X
Specifies that id is to be used as the user
identifier in any "fanout" logins required.
If arg is specified, it is to be either the
password to be used in such logins or "-pw", in
which case the Server will furnish a mask or negotiate the Hide Your
Input Telnet option; if no arg, then no password is to be furnished
on fanout logins.
Default id is "*free".
II. Communications Commands
readmail
Type out "mailbox".
readmail (id) -at host X
Type out "mailbox" on remote Host host.
Multiple Hosts may be specified,
separated by spaces (blanks).
Padlipsky [Page 16]
RFC 666 Unified User-Level Protocol November 1974
Implies ability to change working directory
at host to directory implied by known
user identifier, or (optionally) by id.
readmail -all XX
Search for mail.
Extremely optional.
mail id
Collect input until line consisting of
only a period (".") for mailing to local
user specified by id.
mail -f file id
Send contents of specified file to specified
local user.
mail id -at host
Collect input until line consisting of
only a period (".") for mailing to remote
user(s) at specified Host(s). Both id and
host may be multiple, separated by spaces.
(If multiple, they should be taken pairwise.)
mail -f file id -at host
Send contents of specified file to specified
remote user(s).
who
The generic who's logged in command.
who id
Is id logged in? Constrained responses.
who id -at host
Is the specified user logged in at the
specified host. Constrained responses.
concom -on
Enable console to console communications.
concom -off
Disable console to console communications.
concom id
Send messages to specified local user
until line consisting of only a period (".").
Padlipsky [Page 17]
RFC 666 Unified User-Level Protocol November 1974
concom id -at host
Send messages to specified remote user.
III. File Commands
type path
Type out the contents of the specified file.
Pathname may be local or Network-wide.
Default to current working directory.
listdir
List the contents of the current working directory. (Local format
acceptable.)
listdir path
List the contents of the specified directory.
rename old new
Change the specified file's name as indicated.
addname old new X
Give the specified file the specified extra name.
delete path
Get rid of the specified file.
("Expunge" if necessary.)
copy from to
Make a copy of the file specified by the first pathname at the second
pathname.
link from to X
If your file system has such a concept, make a "link" between the two
pathnames. If no second argument,
use same entry name in working directory.
status path st X
If your file system has such a concept, give status information about
the specified file or directory.
changewd path X
If no argument, return to the "home" directory.
typewd
Type out the pathname of the current working directory.
neted path
Padlipsky [Page 18]
RFC 666 Unified User-Level Protocol November 1974
See Appendix 3.
IV. Invoking "Native" Programs
call name (args)
Invoke the specified program with the
specified arguments (if any).
The following names are reserved to indicate the
invocation of the corresponding language processor: "-pl1", "-basic",
"-fortran", "-lisp".
(If no source file indicated, invoke "interpretively" if possible.)
V. On-line Documentation
help name
Type out information about the specified UULP command. If name is
"-sys", type out information about how to use the local system's help
mechanism; if
"uulp", about the local system's UULP implementation. If no name
given, describe the command itself.
APPENDIX 2. MAP COMMAND CONVENTIONS
This appendix will eventually contain the case-mapping conventions
detailed in RFC 411.
APPENDIX 3. EDIT COMMAND REQUESTS
This appendix will eventually contain descriptions of the neted
command requests (a draft of which now exists), or a reference to the
Resource Notebook version, if that gets published first. For now, it
should be sufficient to point out that the requests are basically
locate, next, top, change, save, and quit -- i.e., it's the "old-
fashioned" flavor of context editor.
[Optical character recognition and initial proofreading performed
11/20-21/04 by The Author. A few original typos were corrected; some
may remain.]
Padlipsky [Page 19]