First, let's talk about displays. Different displays use different representations of images. A character display represents its image as a sequence of character codes. This is a very compact representation, but not a very flexible one; text is all you can get, and probably in only a limited selection of fonts. A vector display represents its image as a list of vector coordinates. This works very well for certain varieties of line drawings, but not so well for filled areas or text. A bitmap display, on the other hand, produces an image by taking a large matrix of zeros and ones, and putting white where the zeros are and black where the ones are (or vice versa). The great advantage of bitmap displays are their flexibility: you can specify a tremendous number of images by giving even a relatively small array of bits. Cursors and icons are two large classes of prominent examples. Of course, you do have to supply enough memory to hold all those bits. Altos and D-machines store their bitmaps in main storage. An alternative would be to provide a special chunk of memory on the side where the display's image sits; such a memory is often called a frame buffer.

The primary display of the Alto is a bitmap that is 608 pixels wide by 808 pixels high. Such a display is almost large enough to do a reasonable job of rendering a single 8.5" by 11" page of text. The CRT on a D-machine has the long axis horizontal instead of vertical, giving a bitmap display that is 1024 pixels wide by 808 high. It had to be 808 high so that D-machines could emulate Altos, of course. The extra space allows you to have something else on the screen as well as the somewhat scrunched page of text that you are editing.

Ere I leave you with a mistaken impression, let me note in passing that bitmap displays are not the final solution to all of the world's problems. Raster displays that can produce various levels of gray as well as black and white can depict images free of the ``jaggies'' and other artifacts that are inherent in bitmap displays [2]. And, for some purposes, color is well worth its substantial expense.

But now on to mice. A mouse has two obvious propertiesit rolls and it clicks. Inside the machine, the mouse position and the display cursor position are completely unrelated; but most software arranges for the cursor to ``track'' the mouse's movements. The three mouse buttons go by various names; ``left'', ``middle'', and ``right'' is one set of names. The mouse buttons are also called ``red'', ``yellow'', and ``blue'' respectively, even though physically they are nearly always black. These colorful names were proposed at an earlier time when some of the mice had their buttons running horizontally instead of vertically. Using colors (even imaginary ones!) worked better than switching back and forth between the nomenclatures ``top-middle-bottom'' and ``left-middle-right''.

Mice also come in two basic flavors: mechanical and optical. Our current mechanical mice roll on three balls: two small ones, and one large one. Motion of the large ball is sensed by two little wipers inside the mouse, one sensing side to side rolling while the other senses forward and backward rolling. The motion of each wiper drives a commutator, and little feelers slide along the commutator, producing the electrical signals that the listening computer can decode. Building one of these little gadgets is not quite as hard as building a Swiss watch, but it's in the same league. The optical mice are a more recent innovation. An optical mouse lives on a special pad, covered with little white dots on a black background. A lens in the mouse images a portion of the pad onto the surface of a custom integrated circuit. This IC has sixteen light-sensitive regions, some of which notice that they are being shined on by the image of a white dot on the pad. As the mouse slides along the pad on its Teflon-coated underbelly, the images of the white dots move across the IC; it is subtly constructed so as to observe this phenomenon, and take appropriate electrical action. For more details on this interesting application of a custom chip, you might enjoy checking out Dick Lyon's blue-and-white report on the subject [3].

Two's company, three's a network. A collection of machines within reasonable proximity is hooked together by an Ethernet; if that doesn't sound familiar, I know of some blue-and-whites that you might like to browse [4,5]. Ethernets are connected to each other by Gateways and phone lines, which for most purposes allow us to ignore the topology of the resulting network. The resulting network as a whole is called an Internet. Occasionally, it's nice to know where things really are; generally, a map of the internet is posted on some wall in CSL.

Ethernets come in two flavors: old and new. The old one runs at 3 MBits/sec, and should now be referred to as the ``Experimental Ethernet''. The unqualified name ``Ethernet'' should be reserved for the new one, the standardized version used in OSD products; it runs at 10 MBits/sec. Within PARC, there are 8 Ethernets. For starters, each of the three floors contains both a 3MBit net and a 10MBit net. The second floor 3MBit net got too long, so it was split into two. The 8th is a 1.5MBit Ethernet used by the voice project. (The Etherphones use a chip made for the copiers. Did you know that many copiers have an Ethernet in them?)

We all know how uncommunicative computers can be when left to their own devices. That's why we invent careful protocols for them to use in talking to each other. There are two entire worlds of protocols that are spoken on our various Ethernets as well: old and new. The old ones are called PUP-based (PARC Universal Packet) [7]. The new ones, known by the acronym NS (Network Systems) [8, 9], are used in the products sold by OSD. Each protocol world includes a hierarchy of protocols for various purposes such as transporting files, or sending and receiving mail.

To further complicate things, there are two Internets within Xerox: old and new. The new one, the Corporate Internet, is run by the branch of Xerox that provides things like our phone service. It's been growing quite rapidly over the past year. The idea is to use the official products from SDD to support the non-research people within Xerox and to demonstrate how well our equipment really works. The products don't support PUPs, so neither does the Corporate Internet.

The Research Internet, the ``old'' Internet developed by the research community, is the collection of ethernets, phone lines, and gateways that carry PUPs. (The Research Internet hardware also transports a lot of NS packets just to keep you on your toes.) There are phone lines as far as Japan and England, with major clusters of workstations in El Segundo (think LA) and Rochester (NY). Within the Palo Alto area, 56KB and T1 lines connect various buildings. (T1 is a 1.5 megabit/sec phone line. That's faster than the software in the gateways. Think of it as a very long skinny Ethernet.)

In addition to connecting up all of the personal computers, the network also includes a number of machines generically called servers. Normally, servers have special purpose, expensive hardware attached to them, such as large-capacity disks, or printers. Their purpose in life is to make that hardware available to the local community. We tend to identify servers by function, so we talk about print servers, file servers, name lookup servers, mailbox servers, tape servers, and so on. Many of the protocols for use of the Ethernet were developed precisely so that personal computers could communicate effectively with servers.

The innards of the Alto are wonderfully described in a clear and informative blue-and-white report [10]; I seriously recommend that you read it. In the very unlikely event that you need to know still more about the Alto, you might try looking in the Alto hardware manual <11>. But for our purposes, suffice it to say that the Alto is a 16-bit minicomputer whose primary claim to fame is that it comes equipped with a bitmap display, a mouse, and an Ethernet interface.

The D-machines are a family of personal computers, each member of which has a name starting with the letter ``D''. As long as you don't look too closely, D-machines look a lot alike. In particular, they are all 16-bit computers with a microprogrammed processor that handles most of the I/O as well running the user's programs. And they all generally come equipped with a hard disk, a bitmap display, a keyboard, a mouse, and an Ethernet interface. There are differences of course: in size, in speed, and in flexibility.

All of the local processors come equipped with a hidden button called the ``boot button'' that is used to reinitialize the processor's state. The Alto had just one boot button, hidden behind the keyboard; pushing it booted the Alto. On Dolphins, the situation is only slightly more complex: there are two boot buttons, one at the back of the keyboard, and the other on the processor chassis itself. They perform roughly the same function, but the one on the chassis is a little more potent. On Dorados, there is a lot more going on. There are really two computers involved, the main Dorado processor and a separate microcomputer called the baseboard. It is the baseboard computer's job to monitor the power supplies and temperature and to stage-manage the complex process of powering up and down the main processor, including the correct initialization of all of its RAM's. The boot button on a Dorado is actually a way of communicating with this baseboard computer. You encode your request to the baseboard computer by pushing the boot button repeatedly: each number of pushes means something different. For details, see Ed Taft's memo on the subject <13>. If the baseboard computer of the Dorado has gone west for some reason (as occasionally happens), your only hope is to push the real boot button, a little white button located on the processor chassis itself, far, far away. Just as the boot button on the keyboard is essentially a one-bit input device for the baseboard computer, the baseboard computer also has a one-bit output device: a green light located on the processor chassis. Various patterns of flashing of this light mean various things, as detailed in <13>.

There is one more bit of folklore about booting that I can't resist mentioningevery once in a while, I have to throw in some subtle tidbit to keep the wizards who read this from getting bored. Our subject this time is the ``long push boot''. Suppose that you have been working on your Dorado for a while, and you walk away to go to the bathroom. When you return and reach toward your keyboard, you get a static shock. You are only mildly annoyed at this until you notice that the cursor is no longer tracking the mouse, and the machine doesn't seem to hear any of your keystrokes. The screen looks OK, but the Dorado is ignoring all input. What has probably happened is that the microprocessor in your terminal has been knocked out by the static shock. Yes, Virginia! In addition to the Dorado itself, and the baseboard computer, there is also a microprocessor in your terminal (located in the display housing), which observes your input actions and sends them on to the main processor under a protocol referred to as ``the seven-wire interface''. What you want to do now is to reboot the terminal microprocessor without disturbing the state of the Dorado at allafter all, you were in the process of editing something, and you are now in danger of losing those edits. What you should do is to depress the boot button and hold it down for quite a while (more than 2.5 seconds); and then release it. This is known as a ``long push boot'', and it does just what you want under these conditions: it reboots your terminal without affecting anything higher up.

The first high-level programming language used on the Alto was BCPL, and quite a bit of program writing was done in that environment over the years. By now, however, essentially no new programming is being done in BCPL. The language itself will be around for some time to come, since there are BCPL programs that perform valuable services for us: the print server programs Press and Spruce and the file server program IFS are three important examples.

History: Of the better-known computer languages, BCPL is closest to C. The fundamental data type in BCPL is a sixteen-bit word. There are facilities in the language for building structured data objects including records and pointers. But there is no type-checking in the language at all. For example, if foo is a pointer to a record of type node that includes a field named next, that field is referenced in BCPL by writing

The debugger in the BCPL environment was named ``Swat''. This name is preserved in the local dialect as the name of the bottom of the three unmarked keys at the right edge of the keyboard. Various debuggers may be invoked in various environments by depressing this key, perhaps in conjunction with the left-hand shift key or the control key. (The right hand shift key won't do; it is too close to the swat key itself for comfort!)

Mesa is a strongly typed, PASCAL-like implementation language designed and built locally. It first ran on Altos. Herein, I shall call that system Alto/Mesa. Dolphins and Dorados (but not Dandelions) can run Alto/Mesa by impersonating an Alto at some level. More recent instances of Mesa now run on all of our D-machines under the Pilot operating system. In passing, I should observe that Pilot is an operating system written in Mesa by folk in SDD. It is a heavier-weight operating system than the Alto OS, providing its clients with multiprocessing, virtual memory, and mapped files.

History: Alto/Mesa programs do not use the Alto OS at all, mostly because Mesa and BCPL have rather different philosophies about the run-time world in which they exist. So the first thing that a Mesa program does when running on an Alto is to junta away almost all of the OS, and set about building a separate Mesa world. It is a considerable nuisance for Mesa and BCPL programs to communicate, since their underlying instruction sets are completely different. So, most of the important OS facilities, such as the file system, had to be re-implemented directly in Mesa. Mesa's memory management strategies replace the revolutionary tactics of ``junta'' and ``counter-junta'' with the relative anarchy of segment swapping.

A fair amount of software was written in Alto/Mesa, but little new programming is being done in that environment; that is, Alto/Mesa isn't quite as dead as BCPL, but it is getting there. Perhaps the crown jewels of Alto/Mesa are the systems Laurel, Grapevine, Mockingbird, and Griffin. You will be hearing more about the former two in the section on electronic mail; to find out more about the latter two, check out their entries in the Glossary.

The Pilot version of Mesa is the home to lots of active programming in several locations. First, it is the system in which the Star product was and is being implemented by OSD. The programmers in OSD have developed a set of tools for programming in Mesa variously called the ``Tools Environment'' or ``Tajo''. This body of software may soon be marketed under the name ``the Xerox Development Environment'' and is currently being released to numerous Universities in the most recent incarnation of the University grant program.

Although Mesa programs look a lot like PASCAL programs when viewed in the small, Mesa provides and enforces a modularization concept that allows large programs to be built up out of smaller pieces. These smaller pieces are compiled separately, and yet the strong type checking of Mesa is enforced even between different modules. The basic idea is to structure a system by determining certain abstract collections of facilities that some portions of the system will supply to other portions. Such an abstraction is called an ``interface'', and it is codified for the compiler's benefit in a Mesa source file called an ``interface module''. An interface module defines certain types, and specifies a collection of procedures that act on values of those types. Only the procedure headers go into the interface module, not the procedure bodies (except for INLINE's, sad to say). This makes sense, since all the interface module has to do is to give the compiler enough information so that it can type-check programs that use the abstraction.

Having specified the interface, some lucky hacker then has the job of implementing itthat is, of writing the procedure bodies that actually do the work. These procedure bodies go into a different type of module called an ``implementation module''. An implementation module is said to ``export'' the interface that it is implementing; it may also ``import'' other interfaces that it needs to do its job, interfaces that some other program will implement.

In simple systems, each interface is exported by exactly one module. In such a system, there isn't much question about who should be supplying which services to whom. In fact, in these simple cases, the binding, that is, the resolution of imports and exports, can be done on the fly by the loader. But in more complex cases, there might be several different modules in the system that can supply the same service under somewhat different conditions, or with somewhat different performance. Then, the job of describing exactly which modules are to supply which services to which other modules can become rather subtle. A whole language was devised to describe these subtle cases, called C/Mesa. The Binder is the program that reads a C/Mesa description, called a config, and builds a runnable system by filling imports request from exports according to the recipe.

The Mesa language is described in the recently updated manual published by the Systems Development Department of OSD [18]. Older blue-and-white versions still abound, and may be of use if the newer version is not available [19]. You may also be interested in Jim Morris's comments on how programs should be structured in Mesa [20].

Smalltalk was developed by the folk in the System Concepts Laboratory (formerly known as the Software Concepts Group). The Smalltalk language is the purest local embodiment of ``object-oriented'' programming:

Genealogists will recognize that ideas from both Simula and Lisp made their way into Smalltalk, together with traces of many other languages.

For some years now, the folk in SCL have been working at trying to get the Smalltalk language and system out into the great wide world. The first public event that came out of this effort was the August 1981 issue of Byte magazine; it was devoted to Smalltalk-80, including a colorful cover drawing of the now famous Smalltalk balloon. In addition, the SCL folk have written several books about Smalltalk, and they are planning to license the system itself to various outside vendors. The first of the books, entitled Smalltalk-80: The Language and Its Implementation, emerged from the presses at Addison-Wesley in 1983 [21]. Other books include Smalltalk-80: The Interactive Programming Environment, and Smalltalk-80: Bits of History, Words of Advice.

LISP is the standard language of the Artificial Intelligence community. Pure LISP is basically a computational incarnation of the lambda calculus; but the LISP dialects in common use are richer and bigger languages than pure LISP. Interlisp is one dialect of LISP, an outgrowth of an earlier language called BBN-LISP; for more historical details, read the first few pages of the Interlisp Reference Manual [22]. One of the biggest strengths of Interlisp is the large body of software that has developed to assist people programming in Interlisp. Consider the many features of Interlisp: an interpreter, a compatible compiler, sophisticated debugging facilities, a structure-based editor, a DWIM (Do What I Mean) error correction facility, a programmer's assistant, the CLISP package for Algol-like syntax, the Masterscope static program analysis database, and the Transor LISP-to-LISP translator, to name a few.

Interlisp itself has been implemented several times. Interlisp-10 is the widely-used version that runs on PDP-10's. Interlisp-D is an implementation of Interlisp on the D-machines [23], produced by folk at PARC. In the process of building Interlisp-D, the boundary between Interlisp and the underlying virtual machine was moved downward somewhat, to minimize the dependencies of Interlisp on its software environment; that is, functions that were considered primitive in Interlisp-10 were implemented in Lisp itself in Interlisp-D. But the principal innovations of Interlisp-D are the extensions that give the Interlisp user access to the personal machine computing environment: network facilities and high-level graphics facilities (including a window package) among them.

By the way, Interlisp has the honor of being the first system (to my knowledge) to use the prefix ``Inter-''. This prefix has become quite the rage of late: Internet, Interpress, Interscriptyou get the general idea.

Back in 1978, folk in CSL began to consider the question of what programming environment we would use on the emerging D-machines. A working group was formed to consider the programming environments that then existed (Lisp, Mesa, and Smalltalk) and to form a catalog of programming environment capabilities, ranked by both by value and by cost. A somewhat cleaned-up version of the report of that working group is available as a blue-and-white for your perusal [24]. After pondering the alternatives for a while, CSL chose to build a new programming environment, based on the Mesa language, that would be the basis for most of our programming during the next few years. That new environment is named ``Cedar''. A recent blue-and-white report presents an introductory ``tour'' through the Cedar programming environment [40].

Cedar documentation is in a constant state of flux; indeed, it might be said that Cedar as a whole, not only its documentation, is in a constant state of flux. Much of the documentation for the current release is accessible through a ``.df'' file named Documentation.df <25>. Hardcopies of this packet of stuff, entitled ``The Cedar Manual'', are produced from time to time, and distributed to Cedar programmers.

The programming language underlying Cedar is essentially Mesa with garbage collection added. Now, adding garbage collection actually changes things quite a bit. First of all, it changes programming style in large systems tremendously. Without garbage collection, you have to enforce some set of conventions about who owns the storage. When I call you and pass you a string argument, we must agree whether I am just letting you look at my string, or I am actually turning over ownership of the string to you. If we don't see eye to eye on this point, either we will end up both owning the string (and you will aggravate me by changing my string!) or else neither of us will own it (and its storage will never be reclaimeda storage leak). Once garbage collection is available, most of these problems go away: God, in the person of the garbage collector, owns all of the storage; it gets reclaimed when it is no longer needed, and not before. But there is a price to be paid for this convenience. The garbage collector takes time to do its work. In addition, all programmers must follow certain rules about using pointers so as not to confuse the garbage collector about what is garbage and what is not.

Thus, programs in the programming language underlying Cedar look a lot like Mesa programs, but they aren't really Mesa programs at all, on a deeper level. To avoid confusion, we decided to use the name ``Cedar'' to describe the Cedar programming language, as well as the environment built on top of it. Cedar is really two programming languages: a restricted subset called the safe language, and the unrestricted full language. Programmers who stick to the safe language can rest secure in the confidence that nothing that they can write could possibly confuse the garbage collector. Their bugs will not risk bringing down the entire environment around them in a rubble of bits. Those who choose to veer outside of the safe language had better know what they are doing.

Those who want to know more about Cedar are once again encouraged to dredge up a copy of the Cedar Manual <25>. It includes documentation on how Cedar differs from Mesa, annotated examples of Cedar programs, manuals for many of Cedar's component parts, a Cedar catalog, and lots of other good stuff. By the way, the most authoritative source for what the current Cedar compiler will do on funny inputs can be found in a document called the Cedar Language Reference Manual, also known by the acronym CLRM. This is logically part of the Cedar Manual, but it is currently bound separately, and only available in draft form. The CLRM suggests a particular design philosophy for building a polymorphic language that is a superset of the current Cedar, since that is the direction in which the authors of the CLRM, Butler Lampson and Ed Satterthwaite, wanted to nudge the Cedar language.

When programming in the Alto world, or in current Cedar, you are dealing with two different types of file systems: local and remote. The local file system sits on your machine's hard disk. Remote file systems are located on file servers, machines with big disks that are willing to store files for you. Local file systems have several unpleasant characteristics in comparison with the remote systems: they are small, and they aren't very reliable. Both of these problems have consequences.

Most systems including Cedar keep at least one or more old versions. Some text editors maintain one backup copy of each file being edited as a separate file, whose name ends with a dollar sign. That is, the backup copy of ``foo.bravo'' is stored in the file ``foo.bravo$''. Remote file servers also maintain multiple versions of files, and letting old versions of things accumulate is one easy way to overflow your disk usage allocation.

No disk is completely reliable. Our remote file servers have automatic backup facilities that protect us from catastrophic disk failures. But the local file systems have no such automatic protection. Since this protection isn't provided automatically, it behooves you to adjust your behavior appropriately: make sure that, on a regular basis, backup copies of the information on your local disk are put in some safe place, such as on a remote file server where suitable precautions are constantly being taken by wizards to protect against disk failure. Doing this is one facet of what is meant by the phrase Living Cleanly, which deserves its own section.

In the outside world, document production systems are usually de-coupled from text editors. One normally takes the text that one wants to include in a document, wraps it in mysterious commands understood by a document processor, feeds it to that processor, and puzzles over the resulting jumble of characters on the page. In short, one programs in the document processor's language using conventional programming toolsan editor, a compiler, and sometimes even a debugger. Programmers tend to think this is neat; after all, one can do anything with a sufficiently powerful programming language. (Remember, Turing machines supply a sufficiently powerful programming language too.) However, document processors of this sort frequently define bizarre and semantically complex languages, and one soon discovers that all of the time goes into the edit/compile/debug cycle, not careful prose composition.

History: Bravo is the editor and typesetter in the Alto world, and it represented a modest step away from the programming paradigm for document production. A single program provided both the usual editing functions and a reasonable collection of formatting tools. You can't program Bravo as you would a document ``compiler'', but you can get very tolerable results in far less time. The secret is in the philosophy: what you see on the screen is what you get on paper. You use the editing and formatting commands to produce on the screen the page layout you want. Then, you tell Bravo to ship it to a print server and presto! You have a hardcopy version of what you saw on the screen. Sounds simple, right?

Of course, it isn't quite that easy in practice. There are dozens of subtle points having to do with fonts, margins, tabs, headings, and on and on. Bravo was a success because most of these issues are resolved more or less by fiatsomeone prepared a collection of configuration parameters and a set of forms that accommodated most document production. Many of the configuration options aren't even documented, so it is hard to get enough rope to hang yourself. The net effect is that one spent more time composing and less time compiling.

In Bravo's wake, several new editors of unformatted text appeared: the Laurel editor, and the editor in the Tools Environment are prominent examples. The Laurel editor is particularly noteworthy in that it pioneered the development of a modeless (or at least less modal) user interface for an editor. The Star product editor and Tioga are more recent local editors in the full Bravo tradition: they can handle formatting and multiple fonts. Tioga is the editor within Cedar, and its user interface is very close to the widely beloved Laurel modeless interfacetry going back to Bravo after using Tioga for a while, and see how horrible it feels to have to remember to type ``i'' and ``ESC'' all the time. Tioga shows formatted text on the screen. To get a hardcopy of that text, the current path involves running a companion program called the TSetter, which will compose your pages for printing and send them to a print server.

In general, our printers are built by taking a Xerox copier and adding electronics and a scanning laser that produce a light image to be copied. There are many different types of such printers, and there are multiple instances of each printer type as well. There are also many different programs that would like to produce printed output. The Press print file format was our first answer to the problem of allowing every printing client to use every printer. Press files are the Esperanto of printing. Most print servers demand that the documents that you send to them be in Press format. This means you have to convert whatever you have in hand (often text) to Press format before a server will deign to print it.

Press file format <30> is hairy, and some print servers don't support the full generality of Press. Generally, however, such servers will simply ignore what they can't figure out, so you can safely send them any Press file you have.

A Press file can ask that text be printed in one of an extensive collection of standard fonts. Unfortunately, you must become a wizard in order to print with your own new font. You can't use a new font unless it is added to the font dictionary on your printer, and adding fonts to dictionaries is a delicate operation: a sad state of affairs. If the Press file that you send to a printer asks for a font that the printer doesn't have, it will attempt a reasonable substitution, and, in the case of Spruce, tell you about the substitution on the break page of your listing. If you have chronic font difficulties of this sort, contact a wizard.

There is a new print file format under development, called Interpress. The print servers that are part of the Star product speak a dialect of Interpress. A print file in Interpress format is called a master. Our local plans for printing Interpress masters involve converting them first into a printer-dependent print file in so-called PD format (with conventional extension ``.pd''). From there, a relatively simple driver program on each printer should be able to produce the final output.

PARC has a variety of printers available for your hardcopy needs. We have high volume printers for quantities of text, listings, and documentation; we have slower printers with generally higher quality for more complex files; and we have very slow printers for extremely high quality. All of our current printers except Platemaker offer 384 spots per inch and share a common font dictionary. We use two different software systems for printing Press files, both running on Altos: one is called Spruce, and the other is called (confusingly) Press. Spruce offers speed and spooling, but it can only image characters and rules, and not too many of them. This makes it limited in graphics applications. Furthermore, Spruce is limited to the particular sizes of fonts that it has stored in its font dictionary: it does not know how to build new sizes by converting from splines. Press is slower, but can handle arbitrary bitmaps, and can produce odd-sized fonts from splines.

The imaging group is developing Interpress printing capabilities. Printing ``.pd'' files is now an option on most Press printers (that is, on printers running the program Press as opposed to Spruce). Just ship your ``.pd'' file to the printer in the standard way: it is smart enough to figure out whether what you have sent it is in PD or Press format, and it will invoke PDPrint or Press as appropriate. Documentation on these two printing programs is available, by the way <31, 32>. PD printing should not be undertaken without consultation with a wizard.

Dover printers run Spruce for high volume printing, producing a page per second. one Dover named Clover, is found in room 2106; another, named Menlo, is in room 2305. Samples of the Dover font dictionary may be found next to Clover and Menlo. Instructions for modifying the queue and generally running these Spruce printers are to be found next to their Alto terminals.

Lilac is our color Press printer and may be found in 2106 with Clover. It is a three color, composite-black machine; it generally produces good quality output, but is occasionally temperamental. Anyone interested in color printing or the state of Lilac should join the distribution list LilacLovers^.pa.

In CSL, we have an assortment of black and white Press printers, answering variously to the names of RockNRoll and Stinger. These printers are Ravens (Raven is a Xerox product). The print quality is normally excellent. Instructions for interpreting status displays are posted locally.

Our best quality printer is Platemaker, which is normally operated at 880 spots per inch, but can be run up to 2200 spi; it is not normally useful to go beyond 1600. Platemaker uses a laser to write on photographic paper or film. Color images can be done in individual separations, which are then merged using the Chromalin process. The Platemaker printing process is used for final prints of fine images or for printing masters for publication.

We rely very heavily on an electronic mail system. We use it for mail and also for the type of announcement that might, in other environments, be posted on a physical or electronic bulletin board. In our environment, a physical bulletin board is pretty useless, since people spend too much of their days staring at their terminals and too little wandering the halls. Electronic bulletin boards might work satisfactorily. But a bulletin board, being a shared file to which many people have write access, is a rather tricky thing in a distributed environment. It probably presupposes a distributed transactional file server, for example. Mumble. For whatever reason, the fact remains that we don't have an electronic bulletin board facility at the moment. As a result, announcements of impending meetings, ``for sale'' notices, and the like are all sent as messages directed at expansive distribution lists. If you don't check your messages once a day or so, you will soon find yourself out of touch (and saddled with a mailbox full of obsolete junk mail). And conversely, if you don't make moves to get on the right distribution lists early, you may miss lots of interesting mail. This business of using the message system for rapid distribution of announcements can get out of hand. One occasionally receives notices of the form: ``meeting X will start in 2 minutesall interested parties please attend''.

Grapevine is the PUP-based distributed transport mechanism that delivers the local mail [33]. When talking to Grapevine, individuals are referred to by a two-part name called an ``R-name'', which consists of a prefix and a registry separated by a dot; for example, ``Ramshaw.pa'' means Ramshaw of Palo Alto. In addition to delivering the mail, Grapevine also maintains a distributed database of distribution lists. A distribution list is also referred to by an R-name, whose prefix conventionally ends in the character up-arrow, as in ``CSL^.pa''. Distribution lists are actually special cases of a construct called a Grapevine ``group''. Groups can be used for such purposes as controlling access to IFS directories. When you send a message, your mail sending program hands it to any convenient Grapevine server. That server expands any distribution lists. Expanding a distribution list may require contacting another server to access registration data not contained locally. The server then forwards the message to each server that needs a copy. Most users have two servers that can store their mail. If the primary server for a user is down, the message will be sent to that users secondary server. In general, if a server crashes, the mail on it is trapped, but everything else should continue to work reasonably smoothly.

There is a program named Maintain that allows you to query and update the state of the distribution list database. In fact, there are two versions of Maintain: the documented one with the unfortunate teletype-style user interface is used from within Laurel or the Mesa Development Environment <34>; the undocumented one with the futuristic menu interface is used from within Cedar. Some distribution lists are set up so that you may add or remove yourself using Maintain. If you try to add yourself to Foo^.pa and Maintain won't let you, the proper recourse is to send a message to the distribution list Owners-Foo^.pa, asking that you please be added to Foo^.

At the moment, Grapevine pretty much has a monopoly on delivering the mail. But there are several different programs that give users access to Grapevine's facilities from different environments. From an Alto, one uses Laurel, which is mentioned elsewhere as a pioneer of modeless editor interfaces. Even if you aren't a Laurel user, I recommend that you read Chapter 6 of the Laurel Manual [35], which is an enlightening and entertaining essay on proper manners in the use of the mail system. In the Mesa Development Environment, the program Hardy provides services analogous to Laurel's. From within Cedar, most folk use Walnut, whose documentation appears as part of the Cedar Manual <25>. Walnut represents a step towards the future in some respects, since Walnut uses Cypress, the Cedar database management system, to store your mail in a database. Access to Grapevine from within Cedar can also be had without the database frills through a program called Peanut, which stores your messages in a structured Tioga document instead of in a database. Finally, in case travel should take you away from your multi-function personal workstation, there are servers on the Internet known by the name ``Lily'' to whom you can connect from any random teletype in order to peruse the mail sitting in your Grapevine mailbox.

Mail systems here understand about Arpanet recipients; mail for these folks is currently sent through a Dorado-based mail gateway to the Arpanet. There is also a mail gateway between Grapevine and the NS world. You can get a message to/from NS land and the Arpanet by going through both gateways. Remote login and file transfer to other Arpanet machines are available through PARCVAX, (or VAXC) a Vax 11/785 running Unix 4.2 BSD.

History: In the BCPL world, the primary facility for packaging up a group of related files and either handing them out or storing them for later recall was the dump file. A dump file, given conventional extension ``.dm'', is simply the concatenation of the dumpees, together with enough header information to allow the dumpees to be pulled apart again. Dump files have fallen out of favor.

In the Alto/Mesa world, and more strongly, in the Cedar world, a collection of software called ``DF files'' has grown up that attacks the problem of describing and packaging systems, detailing their interdependencies, and controlling the versions of things. You can find out a lot about DF files by reading Eric Schmidt's dissertation [36]. You can find out how to use DF files by reading the reference manual for DFTool, the Cedar tool that deals with DF files <37>. All that I will try to do here is to give you some idea of what DF files are good for, and how, in a general sense, they are used. One way or another, all Cedar programmers must make their peace with DF files: they perform valuable functions, and they have no current competition.

In the simplest case, a DF file just consists of a list of the names of a related set of files. At this level, a DF file is something like a dump file: given the DF file, you can get at each of the files that it describes. Of course, you want to be sure that you get the right versions of the described files, so just having the DF file list their names isn't quite enough. If there were an Internet-wide notion of version number that made sense, we could get around this problem by specifying the version number along with the file name. But there isn't. The closest thing to a Internet-wide unique identification stamp that we have is the create date of the file. Thus, what a DF file really contains is a list of file names and associated create dates.

The Bringover function of the DFTool will retrieve to your local file system the set of files described by a particular DF file. This would be something of a challenge unless the DF file included some hint to Bringover concerning where in the great, wide Internet the correct versions of these files might be found. So DF files do indeed include such hints: in particular, they include specifications of remote directories on which to look. These directories are just hints, in the sense that Bringover will always verify by checking the create date that it is getting you the correct version of the specified file. If Bringover can't find the correct version on the specified directory, it will issue a sprightly error message. Bringover has lots of bells and whistles. For example, you can point it at either a local or a remote DF file; you can ask it to retrieve just a selected subset of files to your local disk, rather than the entire set described by the DF file; or you can individually consider the files one by one, deciding which you would like to retrieve and which you wouldn't.

Suppose that I am working on a collection of files, such as the sources for this Briefing Blurb. I have made a DF file that describes them, and I can use Bringover to retrieve them from their remote and permanent home to my local file system, where I can edit them. The next thing that I need is a service that is symmetric to Bringover: after doing my editing, I want to put the new versions back on the remote file server, along with a new DF file that describes the new versions. This function is performed by the SModel button in DFTool. I run SModel, and point it at the old DF file. SModel considers each file in turn, and looks to see if I have edited it; that is, it looks to see if the create date of that file in my local file system is now different than the create date stored in the old DF file. If so, SModel deduces that I have edited the file. It stores the new version that I have made out onto the remote directory. After doing this for each file in turn, SModel writes a new version of the DF file itself, filling in all of the create dates correctly to describe the new version of the entire ensemble. If the DF file describes itself, as most DF files do, SModel is smart enough to make sure that the new version of the DF file is stored out to the remote server as well. SModel also has lots of bells and whistles, but let's not go into them.

If that were the whole story, mere mortals could figure out DF files without straining their brains. But there's more. So far, we have only discussed DF files as descriptions of ensembles of files. In fact, these ensembles are often components of large programs. And this has consequences.

First, there are two distinctly different reasons that you might have for retrieving a program: you might want to change it, or you might just want to run it. In the latter case, you don't need to bring over all of the sources; all that you need is the runnable ``.bcd''. We could handle this by having two DF files: one for the users and the other for the maintainers. But that would be a disaster: the two DF files would never agree! Instead, each DF file distinguishes between files that it ``exports'' and the rest. The exported files are the ones that users need, while maintainers are assumed to need the entire ensemble. Bringover can be warned that you are a user rather than a maintainer by specifying Access: public.

Secondly, some programs are going to depend upon other programs: that is, the programs themselves will be ``users'' (``clients'' is a better word here). This suggests that one DF file should be able to contain another DF file. In fact, there should be several different kinds of containment, corresponding to such phrases as:

You get the point? For more details on the ways that these things are done (``includes'' and ``imports''), check out the reference manual.

In case you still aren't convinced that things are complicated, it is now time to mention the fact that DF files are used for yet another purpose: they describe components of the Cedar release. During the Cedar release process, all of the new versions of Cedar components, which are sitting out on development directories, must be checked for consistency, and then moved en masse to the official release directory. And an entire new set of DF files must be produced, describing the released version of the system (as opposed to the development version). This means, among other things, that some DF files must specify two different remote directories: the development directory and the release directory. In addition, there is a third DFTool function, called Verify, whose purpose is to perform certain consistency and completeness checks on a DF file. By insisting that all component implementors have successfully run Verify on their components, the Cedar Release Master ensures that the release process has at least a fighting chance of succeeding.

DF files grew up over time in response to a mixed bag of needs. As they became more popular, features were added one by one to make them more useful in these varying contexts. The resulting system as a whole is rather hard to grok, but I hope that this introduction has given you a leg up on the problem, at least.

CSL has a weekly meeting on Wednesday afternoons called Dealer, starting at 1:15. The name comes from the concept of ``dealer's choice''the dealer sets the ground rules and topic(s) for discussion. When someone says she will ``give a Dealer on X'', she means that she will discuss X at some future weekly meeting, taking about 15 minutes to do so (plus whatever discussion is generated). Generally, such discussions are informal, and presentations of half-baked ideas are encouraged. The topic under discussion may be long-range, ill-formed, controversial, or all of the above. Comments from the audience are encouraged, indeed, provoked. More formal presentations occur at the Computer Forum on Thursday afternoons; the Forum is not specifically a CSL function, and it is open to all Xerox employees, and sometimes also to outsiders. Dealers are also used for announcements that are not appropriate for distribution by electronic mail. Members of CSL are expected to make a serious effort to attend Dealer.

On occasions of great festivity, Dealer is replaced by a picnic on the hill (that is, Coyote Hill, across Coyote Hill Road), with Mother Xerox picking up the tab.

The CSL Archives (not to be confused with the Cedar Archive System) are a collection of file cabinets and 3-ring binders that provide a continuing record of CSL technical activities. The archives are our primary line of defense in legal matters pertaining to our projects. They also make interesting reading for anyone curious about the history of any particular project.

There is also an institution known as the CSL Notebook, which exists to make all of the potentially interesting documentary output of CSL folk easily accessible to all CSL folk. Trip reports, design documents, immigration manuals (like this one): they should all be submitted to the CSL Notebook <38>. If you thought that it was worth writing down, it is pretty likely that there are other folk in CSL who would consider it worth reading, and submitting it to the CSL Notebook is one easy way to get it read. (I believe that likely looking submissions to the CSL Notebook are considered for entry into the CSL Archives as well.)

CSL has 3 coffee alcoves where tea, cocoa, and several kinds of coffee are available. All coffee drinkers (not just the secretaries or some other such barbarism) help out by making coffee. If you are about to consume enough coffee that you would leave less than a full cup in the pot, it is your responsibility to make a fresh pot, following the posted instructions. There are lots of coffee fanatics around, and they get irritated beyond all reason if the coffee situation isn't working out smoothly. For those coffees for which beans are freshly ground, the local custom is to pipeline grinding and brewing. That is, you are expected to grind a cup of beans while brewing a pot of coffee from the previous load of ground beans. This speeds up the brewing process for everyone, since a load of ground beans isat least, had better bealways ready when the coffee pot runs out.

Be warned as well that some lab members are unbelievably picky about the state of their offices. The convention is that any Alto in an empty office is fair game to be borrowed. Private Dolphins and Dorados may be borrowed only by prior arrangement with their owners, because of the problems of sharing disk space. If you use someone's office for any reason, take care to put everything back exactly the way it was. Don't spill crumbs around, or leave your half-empty cocoa cup on the desk, or forget to put the machine back in the state that you found it, or whatever. Of course, lots of people wouldn't mind even if you were less than fanatically careful. But some people do mind, and there is no point in irritating people unnecessarily.

When you pick up your output from a printer, it is considered antisocial merely to lift your pages off the top of the output hopper, and leave the rest there. Take a moment to sort the output into the labelled bins. Sorting output is the responsibility of everyone who prints, just as making coffee is the responsibility of everyone who drinks (coffee). Check carefully to make sure that you catch every break page: short outputs have a way of going unnoticed, and hence being missorted, especially when they come out right next to a long output in the stack. The rule for determining which bin is to use the first letter that appears in the name on the break page. Thus, ``Ramshaw, Lyle'' should be sorted under ``R'', while ``Lyle Ramshaw'' should be sorted under ``L''. A trickier question is what to do with output for ``Noname'', or the like. Following the rule would suggest filing such output under ``N'', but that doesn't seem very helpful, since the originator probably won't find it. Check the contents and file it in the right box if you happen to recognize whose output it is; otherwise, either leave it on top of the printer or stick it back in the output hopper.