Network Working Group Mike Padlipsky RFC 569/ NET-USING Memo 1 NET-USING NIC # 18972 October 15, 1973
NETED: A Common Editor for the ARPA Network
BACKGROUND
At the recent Resource Sharing Workshop, there was a somewhat surprising degree of consensus on what I had anticipated would the least popular aspect of the my "Unified User-Level Protocol" proposal: A number of the attendees agreed without argument that it would be a good thing to have "the same" context editor available on all Servers -- where "the same" refers, of course, to the user interface. We even agreed that "NETED" seemed to be a plausible common name. In view of the fact that the rest of the proposal didn't seem to capture anybody's imagination, though, it seemed to be a useful notion to separate out the common editor and make it the subject of a stand-alone proposal.
My resolve to come up with the following was further strengthened at the the organizing meeting of the Network User Interest Group, which followed the Workshop. Being primarily concerned with user issues, this group was downright enthusiastic about the prospect of a common editor. Indeed, this proposal has been reviewed by the group and is endorsed by it.
REASONS
The need for a common editor might well be obvious to many readers. They are encouraged to skip this section, which is for the benefit of those who don't already see the light.
In the first place, it's almost axiomatic that to use a time-sharing system, you have to be able to create files (/"datasets"/"segments"). Even if you're only using the Network to send "mail", you'd still like to be able to create a file separately, so as to be able to edit it before sending. And if you want to write a program -- or even make a "runoff" source file -- you simply must be able to use some editor command on the system at hand.
Unfortunately, there are even more editors than there are systems; and each one has it own conventions and peculiarities. So "Network users" (who use several Servers, as opposed to those who use the Network only to access a particular system all the time) are faced with the unpleasant chore of developing several sets of incompatible reflexes if they want to get along. This can certainly be done. It has been by a number of members of the Network Working Group.
NETED SPEC p.2
The real kicker, however, comes when we consider the needs of those users -- who are coming into the Network community in ever-increasing numbers -- who are not professional programmers. They just want to get some work done, "on the Net" (that is, irrespective of which operating system they might be talking to). They are likely to be appalled rather than amused by having to learn a dozen ways of getting to first base. Therefore, it seems clear than not only do we need a common editor, but we also need a simple common editor.
CHOICES
Simplicity is not the only criterion for rejecting the apparently "obvious" choice of either TECO or QED. (That it is a strong factor is indicated by the old test of "Consider explaining it to a naive secretary -- now consider explaining it to a corporation president.") Perhaps even worse is the problem of "dialects". That is, features vary across implementations, and settling on a common set of features (or dialect) is likely to be a very hard task, for programmers tend to get very fond of their familiar goodies. Besides, both TECO and QED have their own strong (/fanatic) advocates, who's probably never be willing to settle for the other one. Further, not every system has both, and implementing the other is a fairly large job even if the NWG could agree on which (and how much).
At any rate, the difficulties seem overwhelming when it comes to choosing a high-powered editor as the common editor. Therefore, I tried to think of a nice low-powered editor, and it suddenly occurred to me that I not only knew of one, but it was even fairly well documented (!). The editor in question is known on Multics as "eds" (the same member of the "ed" family of editors which started on CTSS), a line-oriented context editor (no "regular expressions", but also no line numbers). It is used as an extended example of programming in the Multics environment in Chapter 4 of the Multics Programmers' Manual, which gives an annotated PL/I listing of the actual working program. It is simple to learn and should be quite easy to implement, PL/I version serves as a detailed model with only equivalent system calls and choice of language to worry about. I urge its adoption as the common Network editor, to be known on all participating Servers as "NETED" and/or "neted".
DOCUMENTATION
In view of the fact that if "eds"/NETED is adopted only perhaps a dozen members of the NWG will actually have to implement one, it seems wasteful to distributed some 30 pages of the MPM to everyone -- especially since most of the parties concerned have access to an MPM already. (Another problem solved by not including it here is that of whether I'd be violating copyright by doing so.) The exact reference is pp. 24-54 of Chapter 4 of Part I of the Multics Programmer's Manual.
NETED SPEC p. 3
Anybody who needs a copy can let me know. Although the emphasis in the document is, naturally enough, on the Multics-specific aspects, I believe that the listing is clear enough to serve as a model to implementors without any great difficulty. If we do get to the implementation stage, I'll be glad to try to explain any non-obvious system calls, either individually or in a follow-up memo. But even though we "initiate" where you "open", or we " call los_$read_ptr" where you "IOT TTY" (or something), it shouldn't cause much trouble. For that matter, some implementers might prefer to ignore the existing program and simply work from the function specifications (below).
LIMITATIONS
It became abundantly clear during the course of the review of this document by the User Interest Group that the limitations of NETED must be acknowledged (even insisted upon) and explained here. In the first place, it must be emphasized that it is not being proposed as "THE" Network editor. Rather, it is an insistently simple-minded editor for two major reasons: 1) it is meant for use mainly by non-professional programmers, and 2) more important still, it is meant to be extremely easy to implement. Therefore, it seems far more important to go with the published program, with all its warts, than to specify the addition of new, undebugged features. The idea is to make it implementable in man-days by an average to average-plus programmer instead of in man-weeks by a superstar programmer.
In the second place, the very act of adding new features is fraught with peril. To take some examples from the comments I received during the review phase: In the first draft, I inadvertently failed to document the mechanism by which the ability to "go backwards" (i.e., to reverse the direction of the current-line pointer described below) is actuated. Several reviewers argued strongly for the inclusion of such a mechanism; but, not knowing it was already "in" the code I argued -- successfully -- for leaving it out, on the grounds that we should stick to what's in the existing code, which is known to work as published. Even what to call such a new request would have been debatable -- should it be "-" and become the only non-alphabetic name? should it be "b" for "bottom"? should "n" (for "next") become "+"? And so on. Although this particular issue turned out be a false alarm, I've left it in to emphasize the sort of pitfalls we can get into by haggling over particular "features". Another familiar feature is some sort of "read" request so that the file name need not be specified as an argument to the command. Then, of course, one would also want a "create" or "make" request. And perhaps a file delete request? It keeps going on and on. The point, I think, is that if we allow ourselves to go into "tinker mode" we could spend as many years striving to achieve consensus on what features to add as we've spent on Telnet or FTP ... and still not please everyone. Therefore, I urge the NWG to accept the contention that having a working model to use as
NETED SPEC p. 4
a pattern is more important than any particular additional features (even though I myself find "=" for "what's the current line's number?" annoying to live without).
RESPONSES
For the benefit of those who don't want to plow through the functional spec, this seems to be a good spot to indicate what appropriate responses to this proposal would be. Ideally, I'd like to hear from a responsible system programmer at, say, TENEX, CCN, UCSD, UCSB, AMES-67, one of the DEC 10/50 Hosts, and from any of the experimental Servers who happen to be interested, that they think it's a fine idea and why don't I log in next week to try their NETEDs. Next most desirable would be agreement in principle followed by specific inquiries about "eds". I would hope that haggling over specific features wouldn't occur (as we're not trying to do a definitive editor, just an easy, commonly implemented one based on an existing implementation), but unfortunately I can't legislate such haggles out of existence. At the very least, I'd hope to either hear or see reasoned arguments why it's not worth doing. As usual, phone, mail "mail" ("map.cn" in sndmsg, or "map cn" in "mail" on Multics) or RFC's are the assumed media for responding.
USAGE
(The following is intended to serve double-duty, as both a functional spec now and -- with the addition of some examples -- a "users' manual" later. So if it seems to "tutorial", I'm sorry. And if it doesn't seem tutorial enough -- assuming the addition of examples -- please let me know.)
As is typical of "context editors," the NETED command is used both for creating new files and for altering already existing files -- where "files" are named collections of character encoded data in the storage hierarchy of a time-sharing system. Consequently, NETED operates in two distinct "modes" -- called "input mode" and "edit mode".
When NETED is used to create a file (that is, when it is invoked from command level with an argument which specifies the name of a file which does not already exist in the user's "working directory"), it is automatically in input mode. It will announce this fact by outputting a message along the lines of "File soandso not found. Input." Until you take explicit action to leave input mode, everything you type will go into the specified file. (Actually, it goes into a "working copy" of the file, and into the real file only when you indicate a desire to have that happen.) To leave input mode, type a line consisting of only a period and the appropriate new-line character: ".<NL>", where <NL> is whatever it takes to cause a Telnet New-Line to be generated from your terminal
NETED SPEC p. 5
After leaving input mode, you are in edit mode. Here, you may issue various "requests" which will allow you to alter the contents of the (working) file, re-enter input mode if you wish, and eventually cause the file to be stored. Note that edit mode is entered automatically if the argument you supplied to NETED specified an existing file. Regardless of how it was entered, being in edit mode is confirmed by NETED's outputting a message of the form "Edit". Editing is performed relative to (conceptual) pointer which specifies the current line, and many requests pertain to either moving the pointer or changing the contents of the current line. (When edit mode is entered from input mode, the pointer is at the last line input; when entered from command level, the pointer is at the "top" of the file.)
NETED's edit mode requests follow, in order intended to be helpful. Two important reminders: the requests may only be issued from edit mode, and each one "is a line" (i.e., terminates in a new line / carriage return / linefeed is appropriate to the User Telnet being employed). SYNTAX NOTE: If the request takes an argument, there must be at least one space (blank) between request's name and the argument.
For unsigned m, the n(ext) request causes the pointer to be moved "down" m lines. If m is negative, the pointer is moved "up" m lines. If m is not specified, the pointer is moved one line. If the end of the file is reached, an "End of file reached by n m" message is output by NETED; the pointer is left "after" the last line.
The l(ocate) request causes the pointer to be moved to the net line containing the character string "string" (which may contain blanks); the line is output. If no match is found, a message of the form "End of file reached by l string" will be output (and the pointer will have returned to the top of the file). The search will not wrap around the end of the file; however, if the string was above the starting position of the pointer, a repetition of the locate request will find it, in view of the fact that the pointer would have been moved to the top of the file. To find any occurrence of the string -- rather than the next occurrence -- it is necessary to move the pointer to the top of the file before doing the locate (see following request).
The i(nsert) request cause a line consisting of string (which will probably contain blanks) to be inserted after the current line. The pointer is moved to the new line. Edit mode is not left.
The p(rint) request causes the current line and the succeeding m - i lines to be output. If m is not specified, only the current line will be output. End of file considerations are the same as with "n".
The c(hange) request is quite powerful, although perhaps a bit complex to new users. In the line being pointed at, the string of characters s1 is replaced by the string of characters s2. If s1 is void, s2 will be inserted at the beginning of the line; if s2 is void, s1 will be deleted from the line. Any character not appearing within either character string may be used in place of the slash (/) as a delimiter. If a number, m, is present, the request will affect m lines, starting with the one being pointed at. All lines in which a change was made are printed. The pointer is left at the last line scanned. If the letter "g" is absent (after the final delimiter) only the first occurrence of s1 within a line will be changed. If "g" (for "global") is present, all occurrences of s1 within a line will be changed. (If s1 is void, "g" has no effect.) NOTE WELL: blanks in both strings are significant and must be counted exactly. End of file considerations are the same as with "n".
The d(elete) request causes m lines, including the current one, to be deleted from the working copy of the file. If m is not specified, only the current line is deleted. The pointer is left at a null line above the first undeleted line. End of file considerations are the same as with "n".
Write out the working copy into the storage hierarchy but remain in NETED. (Useful for those who fear crashes and don't want to lose all the work performed.)
Write out the working copy into the storage hierarchy and exit from NETED.
Additional specs:
a. On Multics, type-ahead is permitted. This approach is recommended for all versions of NETED, but is of course not required as various Servers' NCP Implementations may prohibit it; however:
b. If an error is detected, the offending line is output, and pending typeahead (if any) must be discarded (to guard against the possibility of the pending request's being predicated on the success of erroneous request).
c. The command is not reinvokable, in the sense that work is lost if you "quit" out of it via the Telnet Interrupt Process command or its equivalent; indeed, quitting out is the general method of negating large amounts of incorrect work and retaining the original file intact.
(When the time comes, I'll be glad to furnish examples for the users' manual version; but for now, that's all there is.)
NOTE
It really does work, unsophisticated though it may be. I think that it's sufficient to get new users going, and necessary to give them a fighting chance. It would even be of utility within the NWG, for those of us who don't like having to learn new editors all the time. If anybody wants to try it, I'll make a version available to "anonymous users" (see the Multics section in the Resource Notebook if you don't already know how to get in our sampling account), under the name "neted". (*) (If you do try it, please delete files when done with them.)
______________
(*) Knowledgeable Multics users with their own accounts can instead link to >udd>cn>map>neted. It is also there under the names "eds" if you want to save typing a couple of characters.