The Star: A Retrospective

Jeff Johnson and Teresa L. Roberts, US West Advanced Technologies William Verplank, IDTwo David C. Smith, Cognition, Inc. Charles H. Irby and Marian Beard, Metaphor Computer Systems Kevin Mackey, Xerox

crox introduced the 8010 “Star” the system were those that office profes- Information System in April of sionals would supposedly need: docu- 1Y8 I. That introductton was an ments, business graphics. tables. personal x databases, and electronic mail. The set important event in the history of personal The Xerox Star has computing because it changed notions of was fixed. always loaded, and automati- how interacttve systems should be de- significantly affected cally associated with data files. sliminat- signed. Several of Star’s designers. some the computer industry. ing the need to obtain. install. and start the of us responsible for the origmal dectpn right application for a given task or data and others for recent improvements, de- In this retrospective, ftle. Users could focus on their work. scribe in this article where Star came from. oblivtous to concepts like softvvare. oper- what is distinctive about it. and how the several of Star’s ating systems. applications. and programs. original design has changed. In doing so. designers describe its Another important assumption was that we hope to correct some misconception\ Star’s users would be casual. occasional about Star that we have seen in the trade important features, users rather than people who spent most of press and to relate some of what we have their time at the machine. This assumption learned from designing it. antecedents, design led to the goal of having Star be easy to For brevity, we use the name “Star” and development, learn and remember. here to refer to both Star and its successor, When Star was introduced in 1981. its Viewpoint. “ViewPoint” refers exclu- evolution, and some bitmapped screen. windows. mouse- sively to the current product. driven interface, and icons were readily lessons learned. apparent features that clearly distin- guished it from other computers. Soon. What Star is however, others adopted these features. Today, windows. mice, and icons are more Star was designed as an office automa- be connected via and vvould share common. However. Star’s clean, consis- tion system. The idea was that profession- access to file servers. printers. etc. tent has much more to do als in a business or organization would Star’s designers assumed that the target with its details than with its gross features. have on their desks and users were interested in getting their work We list here the features that we think make would use them to produce, retrieve, dis- done and not at all interested in computers. Star what it is, categorized according to tribute, and organize documentation, Therefore. an important design goal was to their level in the system architecture: ma- presentations, memos, and reports. All of make the “computer” as invisible to users chine and network, window and file man- the workstations in an organization would as possible. The applications included in ager, user interface. and document editor.

September 1989 0018.91h2/89/0900-(~1 l$Ol.OO 0 1989 IEEE 11 Machine and network level. Impor- board. In tbe sixties and seventies, people for manipulating display memory. tant aspects of Star can be found in the low- invented many different pointing devices: Like the Alto, Star’s display has a reso- est levels of its architecture: the machine the light pen, the trackball, the joystick, lution of 72 pixels per inch. The number 72 and the network of which it is a part. cursor keys, tbe digitizing tablet, the touch was chosen for two reasons. First, there are screen, and the mouse. 72 printer’s points per inch, so 72 pixels Distributed, personal computing. Like other pointing devices, the mouse per inch allows for a smooth interface with Though currently available in a stand- allows easy selection of objects and trig- the world of and typography. alone configuration, Star was designed gering of sensitive areas on the screen. Tbe Second, 72 pixels per inch is a high enough primarily to operate in a distributed mouse differs from touch screens, light resolution for on-screen legibility of a computing environment. This approach pens, and digitizing pads in that it is a rela- wide range of graphics and character sizes combines the advantages and avoids the tive pointing device: the movement of the (down to about eight points - see Figure disadvantages of the two other primary ap- pointer on the screen depends upon mouse l), but not so high as to cause an onerous proaches to interactive computing: time- movement rather than position. Unlike memory burden, which a screen that shared systems and stand-alone personal light pens, joysticks, and digitizing pads, matched the 300 dots-per-inch printer computers. the mouse (and the corresponding pointer resolution would have. Unlike many PC Time-shared systems, dominant through on the screen) stays put when the user lets graphic displays, the pixel size and density tbe sixties and seventies, allow sharing of go of it. are the same horizontally and vertically, expensive resources like printers and To achieve satisfactory mouse-track- which simplifies the display software and large data stores among many users and ing performance, Star handles the mouse improves image quality. help assure the consistency of data that at a very low level. In some workstations, many must use. Timesharing has the disad- the window system handles mouse track- Window and file manager level. Just vantages that all users depend upon the ing, with the result that the mouse pointer above Star’s (not dis- continued functioning of the central com- often jerks around the screen and may even cussed here) are facilities upon which its puter and that system response degrades as freeze for seconds at a time, depending distinctive user interface rests. the number of users increases. upon what else the system is doing. The Persona1 computers, which have re- mouse is a hand-eye coordination device, Windows. Systems now commonly al- placed timesharing as the primary mode of so if the pointer lags, users just keep mov- low several programs to display informa- interactive computing, have the advan- ing the mouse. When the system catches tion simultaneously in separate areas of tage, as one Xerox researcher put it, “of up, the mouse moves beyond the user’s the screen, rather than each taking up the not being faster at night.” Also, a collec- target. We at Xerox considered this unac- entire display. Star was the first commer- tion of personal computers is more reliable ceptable. cial system to provide this capability. than are terminals connected to a central- Star uses a two-button mouse, in con- Some windowing systems allow win- ized computer: system problems are less trast with the one-button mouse used by dows to overlap each other. Other systems apt to cause a total stoppage of work. The Apple and the three-button mouse used by don’t; the system adjusts the size and posi- disadvantages of PCs, of course, are the most other vendors. Though predecessors tion of windows as they are opened and converse of the advantages of timeshar- of Star developed at Xerox Palo Alto Re- closed. Star’s windowing system could ing. Companies that use stand-alone PCs search Center (see “History of Star devel- overlap windows and often did (for ex- usually see a proliferation of printers, in- opment” below) used a three-button ample, property sheets appeared in win- consistent databases, and nonexchange- mouse, Star’s designers wanted to reduce dows overlapping application windows). able data. tbe number of buttons to alleviate confu- However, early testing revealed that users The solution, pioneered by researchers sion over which button did what. Wby stop spent a lot of time adjusting windows, usu- at Xerox (see “History of Star develop- at two buttons instead qf reducing the ally so they did not overlap. Because of ment” below) and embodied in Star, is to number to one, as Apple did? Because this, and because Star’s 17-inch screen connect personal workstations with a lo- studies of users editing text and other ma- reduced the need for overlapping win- cal area network and to attach shared re- terial showed that a one-button mouse dows, the designers decided to constrain sources (file servers, database servers, eliminated button-confusion errors only application windows to not overlap. How- printers) to that same network. at the cost of increasing selection errors to ever, some situations benefit from over- unacceptable levels. lapping application windows. This, added Mouse. An interactive computer system to a subsequent reduction in the standard must provide a way for users to indicate Bitmapped display. Until recently, screen size to 15 inches (with a 19-inch which operations they want and what data most video display terminals were charac- screen optional), resulted in optional con- they want those operations to be per- ter-mapped. Such displays enable vast straints for Viewpoint, Star’s successor, formed on. Users of early interactive sys- savings in display memory, which, when with the default setting allowing applica- tems specified operations and operands memory was expensive, made terminals tion windows to overlap one another. via commands and data descriptors (such more affordable. as text line numbers). As video display ter- In the seventies, researchers at Xerox Integrated applications. “Integrated” minals became common, it became clear PARC decided that memory would get has become a buzzword used to describe that it was often better for users to specify cheaper eventually and that a bitmapped many things. Here, it means that text, operands - and sometimes operations - screen was worth the cost anyway. They graphics, tables, and mathematical for- by pointing to them on the screen. It also thus developed the Alto, which had a mulas are all edited inside documents. In became clear that graphic applications screen 8.5 inches wide and 10.5 inches tall many other systems, different types of should not be controlled solely with a key- and an instruction set specially designed content are edited in separate application

12 COMPUTER Figure 1. Viewpoint screen image. Star’s bitmapped display, once unique in the marketplace, is non much more common. Such a display permits WYSIWYG editing, display of proportionally spaced fonts, integrated text and graphics, and graphical user interfaces.

windowa and then cut and paated together. L>csXrol~ r,~~rt~/~llor-. Star. unlike all con- For example, a MacDraw drawing put into ventional 55 sterns and many 1%Indow- and a or Aldus Pagemaker mouse-based ones. uvx an analogy with document can no longer be edited; rather, real office\ to make the system easy to the original must be re-edited with learn. This analogy is called “the Deshtop MacDraw and then substituted for the old metaphor.” To quote from an early article Having uindow s and a mouse doe\ not drawing in the document. about Star: make a system an embodtment of the Desh- Not even Star is fully integrated in the top metaphor. In a s) \- sense used here. For example, though the Even user‘\ initial view of Star 1sthe Deal- tcm, users deal mainly with data file\. original structured graphics editor. the top. which rc\embles the top of im office de\k. oblivious to the existence of program”. together M lth surrounding furnlturr and “invoke a text editor.” the! new one (see “History of Star develop- equipment. It repre\ent\ ;1 uorhmg en\,ron- They do not ment” below). and the table and formula ment, where current protect, and accesrtble “open a document.” The system knows editors all operate inside text files, spread- resource\ re\~de. On the xreen are dl\pla)ed the type of each file and notifies the relc- sheets and freehand drawings are cur- picture\ of famtltar office ohJeCt5. such ;i\ vant application program when one is documents. folder\. file drawers. In-haslets. rently edited in separate application win- opened. and out-haArts. These object\ ilre displayed dows and transferred into documents, as small pictures. or icons. Most systems, including M indowed where they are no longer fully editable. The Desktop i\ the prtncipal Star technque ones. use a Tools metaphor. in which users for realizing the phy\icnl offlce metaphor deal mainly with applications as tools. The tcon\ on it are vlvble. concrete emhodi- User-interface level. Star’s user inter- Users start one or more application pro- ments of the correspondmg phq\ical oh,jects. face is its most outstanding feature. In this Star users are encouraged to think of the ob- grams (such as a or spread- section we discuss important aspects of jects on the Desktop in physical terms. YOU sheet), then specify one or more data files the interface in detail. can move the icon\ around to arrange your to edit with each. Such systems do not ex-

September 1989 I3 plicitly associate applications with data to applications is that users sometimes essary in systems in which selection is tiles. Users bear the burden of doing that - want to operate on a file with a program only approximate. Consider the many and of remembering not to try to edit a other than its “assigned” application. systems in which the object of a command spreadsheet file with the text editor or vice Such cases must be handled in Star in an ad is specified by a combination of the cursor versa. User convention distinguishes dif- hoc way, whereas systems like allow location and the command modifier. For ferent kinds of files, usually with filename you to run almost any file through a wide example, Delete Word means “delete the extensions (such as memo.txt). Star re- variety of programs. Star’s designers feel word that the cursor is on.” lieves users of the need to keep track of that, for its audience, the advantages of Modifiers are unnecessary in Star be- which data file goes with which applica- allowing users to forget about programs cause exact selection of the objects of tion. outweighs this disadvantage. commands is easy. In many systems, the SunView is an example of a window large number of object-specific com- system based upon the Tools metaphor Generic commands. One way to sim- mands is made even more confusing by rather than the Desktop metaphor. Its users plify a computer system is to reduce the using single-word synonyms instead of see a collection of application program number of commands. Star achieves sim- command modifiers for similar opera- windows, each used to edit certain tiles. plicity without sacrificing functionality tions on different objects. For example, -80, Cedar, and various Lisp by having a small set of generic commands depending upon whether the object of the environments also use the Tools metaphor apply- to all types of data: Move, Copy, command is a file or text, the command rather than the Desktop metaphor. Open, Delete, Show Properties, and Same used might be Remove or Delete, Dupli- This is not to say that the Desktop meta- (Copy Properties). Dedicated function cate or Copy, and Find or Search, respec- phor is superior to the Tools metaphor. The keys on Star’s keyboard invoke these tively. Desktop metaphor targets office automa- commands. Each type of data object inter- Careful choice of the generic com- tion and publishing. It might not suit other prets a generic command in a way appro- mands can further reduce the number of applications (such as software develop- priate for it. commands required. For example, you ment). However, we could argue that ori- Such an approach avoids the prolifera- might think it necessary to have a generic enting users toward their data rather than tion of object-specific commands and/or command Print for printing various toward application programs and employ- command modifiers found in most sys- things. Having Print apply to all data ob- ing analogies with the physical world are tems, such as Delete Character, Delete jects would avoid the trap that some sys- useful techniques in any domain. Word, Delete Line, Delete Paragraph, and tems fall into of having separate com- The disadvantage of assigning data files Delete File. Command modifiers are nec-

Direct manipulation thetia “computer-car” has only one control: a keyboard. Direct manipulation requires that distinct functions be in- voked and controlled in spatially dlsttnct locations, in ways that Jeff Johnson and Teresa L. Roberts are speckic and appropriate for the particular function being controlled. Data objects should be selected and operated on Stars Desktop metaphor is based upon the more general by simulated physical contact rather than by abstract verbal principle of “direct manipulation.“l~ What, exactly, is direct reference: “That one” rather than “The one in row 6.” Continu- manipulation? Consider the following passage from a descrip- ous functkms (such as screen brightness and color saturation) tion of Apples : shouM bs controlled via continuous controls such as sltders, Imaginadriving a car that has no steeringwheel, scceterator, knobs, and dials. Discrete functkms (such as character font brakepedal, turn signalmr, or gear selector.In placeof all the family) sh~td be ccntroiled via discrete means such as com- familii manualcontrols, you hava only a fypwrifw kayboa& mands, muMposit& &t&es, or menus. In effect, a dii Anytimeyou want to turn a comer,change lanes, stow down, manipulation system has a dfrent input channel for every speedup,honkyourhom,orbadcup,youhevetotypeacommend saqwnca on the keyboard.lJrMtwat& the car can’t undemtand funckontheusercanhaveftperform. Englicrhsen~.Insteed,youmuethoklCbwnaspedalkeywith Conventional interfaces are indirect in that there is a single, onafi~andlypainaomelettefaandnumbwa,auchaa general interface to all func6onafky (such as a keyboard and WCTLAS5.” whkzhmeans. “ slow to M, turn left, and aocelarata command language or a menu). In other words, there is only to $5.” NodoubtyoucouldlaamtodfiveauchacarifyouhadauffWnf one input channel for atl kinds of input; different kinds of input motivstionanddeterminsUon.Butwhyttother,whsnsomanycars are distinguished linguistk?ally, rather than spatially. use fern&r contmfs?Most peoolewouldn ”t.3 Having a different interface to each functkm may seem to contradict the goal of having a con&tent interface, but in fact Actually, It isn’t famlllarfty that makes real cars easier to does not Slmitar functions should fndeed have sfmflar user in- drive than the hypothegcal “computer car” would be - cars terfaces across contexts. Mrect manipulation requires, how- are certainly not familiar to those who are just learning to drive aver, that dHerent functions shouM have distinct interfaces, them. What makes real cars easier to drive is the dtrecmess of just as a car has distfnct interfaces to its various functions. their controls. Real carshave distfnct interfaces to tha spaed Directness versus indirectness ls not a simpfe dichotomy: control (the accelerator pedal), the dtrec6on control (the steer- we can speak of degrees of directness. Consider a graphics ing wheel), the gears (the gearshi handle), the radio (several ad&or for creating itlustrations. In the foflwving sequence of in- knobs and buttons), etc. Each interface is speciaffy d&gned twfacee, each conteine all of the indir&on of the pmvtous for controlling its respective fun&on. In contrast, tha hypo- oneandaddsanewlevel:

14 COMPUTER sheets. illustration\. directories. etc., but than iz\uinp commanda to the \\ stem to do 11,~ happens “behind the user’s back.” it ih nonetheless unnccej\ary. In Star. u\- it. Star’\ dc\iyners also wanted lo ercploit You needn’t fiddle uith the system to un- ers simply Cop! to ;t printer icon whateve the tremendou\ communication po\\ihili- drr\tand what‘s going on: you can under- they hant to print. Similarly. the Movr ties of the display. The! uanted to mo\e stand by inspection. command is used to invoke Send Mail b> away from strictly verbal communica- Onr of Star‘s designer\ wrote moving ;I document to the out-basket. tion. Therefore. they based the system Of’ course. not everything cm he done heavily upon principle\ that are now When r\cr\thmp ,n u computer \\\tem I\ \~\lblc on the \cri‘en. thr dl\pln> hecomr\ via generic commands. Some operations hnown as direct manipulation and graphi- reallt>, Otqrct\ and acttons can he undrr- a ~31 control. are object-specific. For example, word ‘ stood purely in term\ ot their ett’ect\ upon the might USC italics. but italics are meaning- Star user\ control the system h) manipu- d~\pla\, Thl\ va\tl\ \mlpl~f~e\ under\tand- les\ for a triangle. In Star. object-specific lating fraphical elements on the screen. ,ns and. reduce\ learmn~ time.- operations are provided via selection- elements that represent the state of the 5~4 dependent “soft” function keys and via tern and data created by user\. The system An example of this philosophy IS the fact menus attached to application windou\. does not distlnfuish between input and that. unlike many v,indowbn\ed corn output. Anything displayed (output) b) puter system\ (e\en some developed at nil-lJ(,t nlurli/llrlurroll utfd ,~rtr/‘llic~ul the system can bc pointed to and acted upon Xerox). Star has no hidden menus - all I~.SPI’i,ltofi~c,c~. Traditional computer \yh- b) the u\er (input). When Star display\ a available menus arc marked uith menu terns require users to remember and type a directory. it (unlike MS-DOS and Unix) i\ buttons. great deal ju\t to control the system. This not displaying a list ofthe name\ ofthe files For a more detailed explanation of di- Impedes learning and retention. espe- in the directory. it is displaying the filej rc‘ct manipulation. see the \idebar. cially by casual users. Star‘s deigner\ thcmselvc\ so that the user can manrpulate favored an approach emphasiring rccog- them. User\ of this type of system have the nition over recall. seeing and pointing feeling that they are operating upon the Computer user5 often have di fficultl over remembering and typing. This sug- data directly. rather than through an apent managin_e their files. Before Star existed, gested using menus rather than com- -like fetching a book from a library shelf ~1 secretary at Xerox complamed that she mands. However. the designers wanted to yourself rather than asking someone to do couldn’t keep trach of the files on her disk. go beyond a conventional menu-based It for you. An inspection of her system revealed file, approach. They wanted users to feel that ‘4 relntcd principle is that the grate of the narned~memo. memo 1, memo07 1379. let- they are manipulating data directly. rather system always shous in the display. Noth- ter. etc. Naming thing\ to keep track of

(1) The most direct interface for moving a circle would have method 2 is less direct than method 1. the user directly at the screen and pull the circle to its new The above examples involve an illustration tool and a win- location. dow manager. Such apphcations are actually in a special cate- (2) Introducing a mouse, bitpad, or joystick adds one level of gory with respect to direct manipulation, because the images indirection: moving the mouse, bitpad stylus, or joystick on the on the screen are what the application is intended to manipu- desk moves the pointer on the screen. Some users have diff- late. The purpose of many applications (such as databases, culty with this indirection. command and control, and file management) is to allow users (3) Arrow keys introduce another level - and another kind - to manipulate informatlon that is only represented on the of indirection: the keystroke movements required to move the screen in some way (for example, pictorially or textually). Such screen pointer, and hence the circle, do not resemble the de- applications therefore have one inherent level of indirection. sired movement of the circle. Systems having direct-manipulation user interfaces encour- (4) Typing a command to move the circle is still more indirect age users to think of them as tools rather than as assistants, Though typing a command involves movements (keystrokes), agents, or coworkers. Natural-language user interfaces, we are inclined to thmk of the movements as incidental; they which are inherently indirect, encourage the reverse. As direct- could just as well be speech. Thus, it is no longer a matter of manipulation interfaces become more prevalent and as pro- movement - similar or not - in one place corresponding to gress is made In natural-language understanding and genera- movement in another place; rather, it is the syntax and seman- tion, it will be interesting to see which way users prefer to think tics of the command that determine what happens, about their computers. Differences in directness can be very subtle. Contrast the following two methods of changing the size of a window on the display: References

(1) Grabbing onto a corner of the window and stretching the 1. E. HutchIns, J. Hollan. and D.A Norman, “Direct Manlpulatlon window to the desired size. interfaces,” In User-centered System Desjgn, D.A Norman and S (2) Clicking on the desired window, choosing Resize from a Draper, eds.. Erlbaum Associates, H&dale, New Jersey, 1986. pp. 67-124. command pull-down menu, then pointing to where the win- dow’s new border is to be moved. 2. B. Shnelderman, “Direct Manipulation: A Step Beyond Program- It is sometimes said that mouse-driven user interfaces are mung Languages,” Computer, Vol 16, No. 8, Aug 1983, pp. 57-68.

direct while keyboard user interfaces are indirect. Note, how- 3. L. Poole, “A Tour of the Mac Desktop,” MacWorld, Vol 1, No. 1, ever, that both methods 1 and 2 above use a mouse, yet 1984, pp. 16-21.

September 1989 I5 them is bothersome enough for program- turn, means that designers need not assign ers to set parameters for the duration of an mers, but completely unnatural for most double-duty (that is, different meanings in interactive session or for the duration of a people. different modes) to physical controls. command, but not for particular data ob- Star alleviates this problem partly by A second way is by allowing applica- jects. For example, headings in Wordstar representing data files with pictures of tions to operate simultaneously. When documents do not “remember” whether office objects called icons. Every applica- using one program (such as a document they are centered or not; whether a line is tion data file in the system has an icon rep- editor), users are not in a mode that pre- centered is determined by how the pro- resenting it. Each type of file has a charac- vents them from using the capabilities of gram was set when the line was typed. teristic icon shape. If a user is looking for a other programs (such as the desktop man- Similarly, directories in Unix do not spreadsheet, his or her eye can skip over ager). “remember” whether files are to be listed mailboxes, printers, text documents, etc. A third way Star avoids modes is by us- in alphabetical or temporal order; users Furthermore, Star allows users to or- ing a noun-verb command syntax. Users must respecify which display order ganize files spatially rather than by dis- select an operand (such as a file, a word, or they want every time they invoke the 1s tinctive naming. Systems having hierar- a table), then invoke a command. In con- command. chical directories, such as Unix and MS- ventional systems, arguments follow DOS, provide an abstract sort of “spatial” commands, either on a command line or in Progressive disclosure. It has been said file organization, but Star’s approach is response to prompts. Whether a system that “computers promise the fountains of concrete. Files can be kept together by uses noun-verb or verb-noun syntax has a utopia, but only deliver a flood of informa- putting them into a folder or simply by lot to do with how mqded it is. In a noun- tion.“4 Indeed, many computer systems clumping them together on the Desktop, verb system such as Star, selecting an ob- overwhelm their users with choices, com- which models how people organize their ject prior to choosing a command does not mands to remember, and poorly organized physical worlds. Since data files are repre- put the system into a mode. Users can de- output, much of it irrelevant to what the sented by icons, and files are distinguished cide not to invoke the command without user is trying to do. They make no presump- by location and specified by selection having to “escape out” of anything or can tions about what the user wants. Thus, they rather than by name, users can use names select a different object to operate on. are designed as if all possible user actions like memo, memol, letter, etc., without Though Star avoids modes, it is not were equally likely and as if all informa- losing track of their files as easily as they completely modeless. For example, the tion generated by the system were of equal would with most systems. Move and Copy commands require two interest to the user. Some systems dimin- As bitmap-, window-, and mouse-based arguments: the object to be moved and the ish the problem somewhat by providing systems have become more common, the final destination. Though less moded default settings of parameters to simplify use of the term “icon” has widened to re- ways to design Move and Copy exist, these tasks expected to be common. fer to any nontextual symbol on the dis- functions currently require the user to se- Star goes further towards alleviating play. In standard English, “icon” is a term lect the object, press the Move or Copy key, this problem by applying a principle called for religious statues or pictures believed to then indicate the destination using the “progressive disclosure.” Progressive contain some of the powers of the deities mouse. While Star waits for the user to disclosure dictates that detail be hidden they represent. It would be more consis- point to a destination, it is in Move or Copy from users until they ask or need to see it. tent with its normal meaning if “icon” mode, precluding other uses of the mouse. Thus, Star not only provides default set- were reserved for objects having behav- These modes are relatively harmless, tings, it hides settings that users are un- ioral and intrinsic properties. Most gra- however, because (1) the shape of the cur- likely to change until users indicate that phical symbols and labels on computer sor clearly indicates the state of the system they want to change them. Implicit in this screens are therefore not icons. In Star, and (2) the user enters and exits them in the design are assumptions about which prop- only representations of files on the Desk- course of carrying out a single mental plan, erties will be less frequently altered. top and in folders, mailboxes, and file making it unlikely that the system will be One place progressive disclosure is drawers are called icons. in the “wrong” mode when the user begins used is in property sheets. Some objects his or her next action. have a large number of properties, many of Few modes.A system has modes if user which are relevant only when other prop- actions differ in effects or availability in Objects have properties. Properties erties have certain values (see Figure 2). different situations. Tesler has argued that allow objects of the same type to vary in For example, on the page layout property modes in interactive computer systems appearance, layout, and behavior. For sheet, there is no reason to display all of the are undesirable because they restrict the example, files have a Name property, char- properties for specifying running header functions available at any given point and acters have a Font family property, and content and position unless the user actu- force users to keep track of the system’s paragraphs have a Justified property. ally specifies that the document will have state to know what effect their actions will Properties may have different types of val- running headers. have.’ Though modes can be helpful in ues: the Name property of a file is a text Another example of progressive disclo- guiding users through unfamiliar proce- string; the Size property of a character sure is the fact that property displays in dures or for handling exceptional activi- might be a number or a choice from a menu; Star are temporary, displayed on demand. ties, they should be used sparingly and the Justified property of a paragraph is ei- In some systems, the properties of the cur- carefully. ther “on” or “off.” In Star, properties are rent selection are displayed at all times, Star avoids modes in several ways. One displayed and changed in graphical forms through codes embedded in the te.xt or in an is the extensive use of generic commands called property sheets. area of the screen reserved for that pur- (see above), which drastically reduces the Property-based systems are rare. Most pose, even though the user usually doesn’t number of commands needed. This, in computer systems, even today, allow us- care.

16 COMPUTER Figure 2. Progressive disclosure. Star’s property sheets, like the rest of the interface, use a principle known as progressive disclosure to avoid overwhelming users with information, Llsually, users don’t need to see an object’s properties: they only need to see and perhaps change its assigned style. Users see an object’s properties only upon request. Also, even when a user sets a property sheet to show an object’s properties, as shown here, some information remains hidden until the user asks to see it. For example, there is no need to clutter the property sheet here with boxes for entering numbers for “Other” values of Line Height. Spacing Before Paragraph, or Spacing After Paragraph until the user actually sets the property to “Other.”

A highlq refined manifestation of Pro- El~r/‘/lcl\i.s o,, ,yf’c’rf ,ppl1rc (Ilid .s~‘,-c’t’,I rect manipulation. is to create the illusion gressive disclosure recently added to (ILcsI,~II. Windows. icons. and property of manipulable objects. It should be clear Viewpoint is stoles. which allows users to sheets are uselrss if users can’t easily dis- that objects can be selected and how to se- regard document content (such as a para- tinguish them from the background or lect them. It should be obvious when the) graph) as having a single style rule instead each other. can’t easily see which labels are selected and that the next action will of a large number of properties. Thus, correspond to which object\. or can’t cope apply to them. Whereas the usual task ot styles hide needless detail from users. with the visual clutter. To assure that Star graphic designers is to present informa- presents information in a maximally per- tion for passive viewing. Star’s designers C‘on.sistcr~c~~.Because Star and all of its celvahle and useful fashion. Xerox hired had to figure out how to present informa- applications were designed and devel- graphic designers to deTermIne the ap- tion for manipulation as well. This show\ oped in-house, its designers had more toll- pearance and placement of screen objects. mo%t clearly in the Desktop icons. uith trol over its user interface than is usually These designers applied various written their clear figure/ground relationship: the the case with computer systems. Because and unwritten principles to the design of icons stand by themselves, with self-con- the designers paid close attention to detail. the window headers and borders, the Desk- tained labels. Windows reveal in their bor- they achieved a very high degree ofconsis- top background. the command buttons, ders the “handles” for scrolling. paging. tency. The left always se- the pop-up menus. the propert) sheets. and window-specific commands, and pop-up lects; the right always extends the selec- the Desktop icons. The most important menus. tion. Mouse-sensitive areas always give principles are l Visual order and user focus. One of the feedback when the left button goes down. most obvious contributions of good but never take effect until the button comes l The illusion of manipulable objects. graphic design is appropriate visual order “P. One goal. fundamental to the notion of d- and focus on the screen. For example, in-

September 1989 I7 Figure 3. Visual order and user focus. The large amount of contrast present on the screens of many window systems (left screen) makes it difftcult to focus on the relevant information. The selection should be the user’s main focus: it is the object of the next operation. The right screen shows how Star/ViewPoint’s screen design focuses attention on the selection.

tensity and contrast, when appropriately applied, draw the user’s attention to the most important features of the display. In some windowing systems, window interiors have the same (dark) color as the Desktop background. Window content should have high intensity relative to the Desktop, to draw attention to what is im- portant on the screen. In Star, window con- tent background is white, both for high contrast and to simulate paper. Star keeps the amount of black on the screen to a minimum to make the selection stand out (see Figure 3). In most window- ing systems, window headers and other areas of the screen are black, making the selection hard to find. This principle is so important that Star’s designers made sure that the display hardware could fill the nonaddressable border of the screen with Desktop grey rather than leaving it black as in most systems. Star also uses icon images that turn from mostly white to mostly black when selected (see Figure 4) and allows at most one selection on the screen at a time.

l Revealed structure. Often, the more powerful the program used, the greater the distance between intention and effect. If only effect is displayed and not intention, the user’s task of learning the connection is much more difficult. A good graphical interface can make apparent to the user these connections between intention and effect, that is, “revealed structure.” For Figure 4. Visual order and user focus. Four candidate sets of icons were designed example, there are many ways to deter- and tested for Star. A representative sample from each set is shown here. In Star, mine the position and length of a line of text the icon selected by the user is indicated by inverting its image. Candidate icon on a page. It can be.done with page margins, sets in which the images are mostly white allow icons to stand out when selected. paragraph indentations, centering, tabs, The set that best satisfies thii criterion, the one on the upper left, was chosen. blank lines, or spaces. The WYSIWYG, or

I8 COMPUTER “what you see is what you get,” view of all these would be identical. That would be enough if all that mattered to the user was the final form on paper. But what will hap- pen if characters are inserted‘? If the line is moved to another page, where will it land? WYSIWYG views are sometimes not enough. Special views are one method of reveal- ing structure. In Star, documents can show “Structure” and/or “Non-Printing Char- acters” if desired (see Figure 5). Another convenient means for revealing structure is to make it show up during selection. For example, when a rectangle is selected in a graphics frame, eight control points high- light it, any of which can attach to the cur- ...... f...... sor during Move or Copy and can land on gTl-tc peace sign is 3 styljzetj reii- grid points for precise alignment. The ditim.of the, control point highlighting allows a user to : foo!print 0f.a it-.CJ ‘.’'- C .tt-te inter- distinguish a rectangle from four straight : na?imal bird of Iines; both might produce the same printed ] pe:?ce, ...... I...... effect but would respond differently to editing.

l Consistent and appropriate graphic -.,.,...... __ ...... _...... - ...... _ ...... _._.. ‘ vocabulary. Property sheets (see Figure 2) present a form-like display for the user to Figure 5. Revealed structure. At the top is the WYSIWYG view of mixed text and specify detailed property settings and ar- graphics. The middle two panels show that structure is revealed when an object is guments to commands. They were de- signed with a consistent graphic vocabu- selected. When a line segment is selected, its control points are shown. When text is selected, the text string is revealed. The bottom panel shows the effect of the lary. All of the user’s targets are in boxes: Show Structure and Show Non-Printing Characters commands, which is to re- unchangeable information such as a prop- erty name is not. Mutually exclusive val- veal the location of embedded graphics and text frames (dotted lines) and “new ues within choice parameters appear with paragraph” and Space characters. boxes adjacent. Independent “on/off” or state parameters appear with boxes sepa- rated. The current settings are shown in- ground texture so that they have a consis- verted. Some of the menus display graphic tent quality appearance. symbols rather than text. Finally, there are text parameters consisting of a box into Document editor level. At the top level which text or numbers can be typed, cop- of Star’s architecture are its applications, ied, or moved, and within which text edit- the most prominent of which is the docu- ing functions are available. ment editor.

l Match the medium. It is in this last prin- ciple that the sensitivities of a good WYSIWYG document editor. Within the graphic designer are most apparent. The limits of screen resolution, Star docu- goal is to create a consistent quality in the ments are displayed as they will print, in- graphics that is appropriate to the product cluding typographic features such as bold- and makes the most of the given medium. face, italics, proportional spacing, vari- Star has a large black and white display. able font families, and superscripts, and The solutions the graphics designers de- layout features such as embedded graph- vised might have been very different had ics, page numbers, headers, and footers. d the display had grey-scale or color pixels. This is commonly referred to as “what you Figure 6. Match the medium. Many. A common problem with raster displays see is what you get,” or WYSIWYG. graphic refinements were made during is “jaggies”: diagonal lines appearing as Star adheres to this principle even in the design process. For example, the staircases. With careful design, jaggies domains where other WYSIWYG docu- turned corner of the document icon can be avoided, for example, by using only ment editors do not. For example, mathe- was moved to the top so that the three vertical, horizontal, and 4%degree matical formulas are created and edited in lines of label would line up with the la- angles. Also important is controlling how documents using a WYSIWYG editor that bels of other icons. Also, icons were the edges of the figures interact with the has knowledge built into it about the ap- carefully sized and positioned against texture of the ground. Figure 6 shows how pearance and layout of mathematical sym- the gray background to create edges are carefully matched to the back- bols. A square root sign has a slot for an smoother lines.

September 1989 19 expression and grows when the expres- ible with each other in data format as well influenced Star’s user interface as a sion becomes large (see Figure 7). In most as user interface. as well as its graphics applications. systems, mathematical formulas are cre- Star, in contrast, assumes that the pri- ated either by putting together special mary use of the system is to create and NLS. Also in the sixties, Douglas Engel- characters to make mathematical symbols maintain documents. The document edi- bart established a research program at or by using a special in-line notation (such tor is thus the primary application. All Stanford Research Institute (now called as sqrt(sigma( 1, n, (x*3)/2))) to represent other applications exist mainly to provide SRI International) for exploring the use of the formula that will eventually be printed. or manipulate information whose ulti- computers “to augment the knowledge Formulas created with such systems usu- mate destination is a document. Thus, worker” and human intellect in general. ally require several print-edit cycles to get most applications are integrated into the He and his collegues experimented with right. document editor (see “Integrated appli- different types of displays and input de- cations” above), operating within frames vices (inventing the mouse when other Extended character set for multilingual embedded in documents. Those applica- pointing devices proved inadequate) and capability. Star uses 16- character tions that are not part of the document developed a system commonly known as codes, in contrast to most of the computer editor support transfer of their data to NLS.* industry, which uses seven- or eight-bit documents. NLS was unique in several respects. It character codes (for example, ASCII or used CRT displays when most computers EBCDIC). The Star character set is a su- used teletypes. It was interactive (i.e., on- perset of ASCII. The reason for a 16-bit History of Star line) when almost all computing was code is a strong market requirement for development batch. It was full-screen-oriented when enhanced multilingual capabilility com- the few systems that were interactive were ing from Xerox’s subsidiaries in Europe Having described Star and Viewpoint, line-oriented. It used a mouse when all and Japan. Most systems provide non- we will describe where they came from and other graphic interactive systems used English characters through different how they were developed. Figure 8 graphs cursor keys, light pens, joysticks, or digit- fonts, so that the eight-bit “extended” this history, showing systems that influ- izing tablets. Finally, it was the first sys- ASCII codes might be rendered as math enced Star and those influenced by it. tem to organize textual and graphical in- symbols in one font, Greek letters in an- formation in trees and networks. Today, it other font, and Arabic in yet another. This Pre-Xerox. Although Star was con- would be called an “idea processor” or a has the effect that when any application ceived as a product in 1975 and was re- “hypertext system.” loses track of font information while han- leased in 198 1, many of the ideas that went dling the text (which happens often in into it were born in projects dating back The Reactive Engine. While Engelbart some systems), a paragraph of Arabic may more than three decades. et al. were developing ideas, some of turn into nonsensical Greek or math sym- which eventually found their way into bols or something else, and vice versa. Memex. The story starts in 1945, when Star, , then a graduate student, Star uses 16-bit character codes to per- Vannevar Bush, a designer of early calcu- was doing likewise. His dissertation, The mit the system to reliably handle European lators and one of President Franklin D. Reactive Engine, contained the seeds of languages and Japanese, which uses many Roosevelt’s science advisors, wrote an many ideas that he and others later brought thousands of characters. All Star and article describing his vision of the uses of to fruition in the Smalltalk language and Viewpoint systems have French, German, electronics and information technology. programming environment, which, in Italian, Spanish, and Russian language At a time when computers were new, turn, influenced Star. Like the designers of capabilities built in. The Japanese lan- room-sized, and used only for military NLS, Kay realized that interactive appli- guage capability was developed as part of number-crunching, Bush envisioned a cations do not have to treat the display as a the original Star design effort and released personal, desktop computer for non-nu- “glass teletype” and can share the screen in Japan soon after Star’s debut in the merical applications. He called it the with other programs. United States. Since that time, many more Memex. Due to insufficient technology characters have been added, covering Chi- and insufficient imagination on the part of Xerox PARC. In 1970, Xerox estab- nese, Arabic, Hebrew, and nearly all Euro- others, Bush’s ideas languished for 15 lished a research center in Palo Alto to pean languages. years. explore technologies that would be impor- As explained in several articles by Joe tant not only for the further development of Becker, the designer of Star’s multilin- Sketchpad.In the sixties, people began Xerox’s then-existing product line (copi- gual capabilities, handling many of the to take interactive computing seriously. ers), but also for Xerox’s planned expan- world’s languages requires more than an One such person was . He sion into the office systems business. The expanded character set.5 Clever typing built an interactive graphics system called Palo Alto Research Center was organized schemes and sophisticated rendering al- Sketchpad that allowed a user to create into several laboratories, each devoted to gorithms are required to provide a multi- graphical figures on a CRT display using a basic and applied research in a field related lingual capability that satisfies customers. light pen. The geometric shapes users put to the above goals. The names and organi- on the screen were treated as objects: after The document is the heart of the world being created, they could be moved, cop- and unifies it. Most personal computers ied, shrunk, expanded, and rotated. They *The actual name of the system was On-Line System. and workstations give no special status to could also be joined together to make A second system called Off-Line System was abbrevi- ated FLS, hence NLS’s strange abbrewiation.NLS is any particular application. Dozens of ap- larger, more complex objects that could now mark&d by McDomtell Douglas under the name plications are available, most incompat- then be operated upon as units. Sketchpad Augment.

20 COMPUTER Here is an elementary integral:

I Figure 7. WYSIWYG formula editing. Mathematical formulas are edited in Star in a highly WYSIWYG fashion, in contrast to most systems, in which formulas are specified via in-line expressions or by constructing them from pieces in a special character font.

Memex

ess

Macintosh

:ript

5 Metaphor

Cognition V k MCAE Mac II Viewpoint 2.0 System

Figure 8. How systems influenced later systems. This graph summarizes how various systems related to Star have influenced one another over the years. Time progresses downwards. Double arrows indicate direct successors(i.e., follow-on versions). Many “influence arrows” are due to key designers changing jobs or applying concepts from their graduate research to products.

September 1989 21 cated to a vision ot personal computers in a distributed environment. In tact. the) coined the term “” in 1973. long before iiiicrocompt~t~r~ started vv hat has been called the “personal com- puter revolution.” One result of thc search for 3 ncvv q- preach M ;I\ the Alto (set‘ Figure 9 1, The .-\I10 v.;15 ;I that had a remov- ahlc. 2.5.megabyte hard dish path tllopp> dishs did not exist ;II the time) and 17X to 256 hilob) tcs of mcmor~. L’nlil\e rno\t machines of its day. the Alto ;IIW had a rnIcropro~rammable instruction set. ii “full-page“ t t OJ x W/1 inch. 600 x 800 pixel) bitmapped graphic display. about 50 hitobhtes of high-speed dispta) nlt‘n~- orb. and a mouse. The first Alto became operational in 1972. At tirst. onl! a halt-doren OI- w Al- tos were built. Atter software that ex- plaited the Alto‘s capahilitics hccamc av ailabtc. d~~~~a~idfor them frcvv trcmcn- dousl). spreading be! and PARC into Xerox as a \v hole and even to external cus- tamers. EventualI!. Xerox built more than a thousand Altos.

Fti~r-rrcjr. Another product of the neu approach v+ as the Ethernet. With its stair- dardired. layered communications proto- cots. Ethernet provided a uay of connect- ing computers much more tlcxibt) than previousI> possible. Soon after the first Altos uere built. they uere networked together. Eventually. the network greu to thousands of worhstations (Altos and Alto successors) within Xerox’s uorlduidc organization.

Sn~ni/rcl/l\. Alan Kay mas OIW of the main advocates of the Alto. His Learning Rc- search Group began using the Alto to build Figure 9. The . The Alto, developed at Xerox PARC in the seventies. prototypes for a personal computing s\ 4- was a prototype for Star. Both its hardware design and the many programs writ- tern “of the future” - a portable machine ten for it by PARC researchers strongI> influenced Star’s designers. that vvoutd provide not canned apptica- tions but rather the building blocks neces- sar:, for users to build the tools and appli- cations they needed to solve their own in- zation of the labs have changed o\er the slogan “The best ua\ to predict the future formation processing problems. The tech- years. but the research topic\ habe stayed is to invent it.” After some initial ewperi- nologies needed to build a lap computer the same: materials science. laser physics. ments with time-shared systems. thej with the power of the envisioned system integrated circuitry. computer-aided de- began searching for a new approach to (called the “DlnaBook”) here unavait- sign and manufacturing, user intertacea computing. able at the time and still are. (not necessaril! to computers). and con- Among the founding members of PARC The prototypes developed by Kay’s puter science (including netuorking. da- was Alan Kay. He and his colleagues were group evolved into the Smalttalk language tabases. operating systems. languages acquainted with NLS and liked its nov>el and programming env,ironment. Smatt- and programming en\ ironments. praph- approach to human-computer interaction. talk further promoted the notion of per- its. document production systems. and Soon. PARC hired several people who had sonal computing; pioneered complete, artificial intelligence). worked on NLS. In I97 I. the center signed interactive programming environments; an agreement with SRI licensing Xerox to and refined and soliditied concepts of ob- Alto. PARC researchers were fond of the use the mouse. Kay and others were dedi- ject-oriented programming that had been

22 COMPUTER extant only in vestigial forms in previous Microsoft Word, a direct descendent of chine. Laurel and Hardy were instrumen- systems. Most importantly for Star, BravoX. Another member of the BravoX tal in getting nonengineers at Xerox to use Smalltalk demonstrated the power of gra- team, Tom Malloy, went to Apple and e-mail. The use of e-mail spread further phical, bitmapped displays; mouse- wrote LisaWrite. with the spread of Star and Viewpoint driven input; windows; and simultaneous throughout Xerox. applications. This is the most visible link Draw, Sil, Markup, Flyer, and Doodle. between Smalltalk and Star, and is perhaps Star’s graphics capability (its provisions OfficeTalk. One more Alto program why many people wrongly believe that for users to create graphical images for that influenced Star was OfficeTalk, a Star was written in Smalltalk. incorporation into documents, as opposed prototype office automation system writ- to its ) owes a great ten by Clarence (“Skip”) Ellis and Gary Pygmalion. The first large program to deal to several graphics editors written for Nutt. OfficeTalk supported standard of- be written in Smalltalk was Pygmalion, the the Alto and later machines. fice automation tasks and tracked jobs as doctoral thesis project of David C. Smith. Draw, by Patrick Beaudelaire and Bob they went from person to person in an or- One goal of Pygmalion was to show that Sproull, and Sil (for Simple Illustrator) ganization. Experience with OfficeTalk programming a computer does not have to were intellectual successors of Suther- provided ideas for Star becauseof the two be primarily a textual activity. It can be land’s Sketchpad (see above): graphical systems’ similar target applications. accomplished, given the appropriate sys- object editors that allowed users to con- tem, by interacting with graphical ele- struct figures out of selectable, movable, Summing up. The debt that Star owes to ments on a screen. A second goal was to stretchable geometric forms and text. In the Alto and its software is best summed up show that computers can be programmed turn, Star’s graphic frames capability is in by quoting from the original designers, in the language of the user interface, that is, large measure an intellectual successor of who wrote in 1982: by demonstrating what you want done and Draw and Sil. Markup was a bitmap graphics editor Alto served as a valuable prototype for having the computer remember and repro- Star. Alto users have had several thousand duce it. The idea of using icons - images (that is, a paint program) written by Wil- work-years of experience with them over a that allow users to manipulate them and in liam Newman for the Alto. Flyer was an- period of eight years, making Alto perhaps so doing act upon the data they represent- other paint program, written in Smalltalk the largest prototyping effort in history. came mainly from Pygmalion. After com- for the Alto by Bob Flegel and Bill Bow- There were dozens of experimental pro- grams written for the Alto by members of the pleting Pygmalion, Smith worked briefly man. These programs inspired Doodle, a Xerox Palo Alto Research Center. Without on the NLS project at SRI before joining paint program written for a later machine the creative ideas of the authors of these sys- the Star development team at Xerox. by Dan Silva. Doodle eventually evolved tems, Star in its present form would have been into Viewpoint’s Free-Hand Drawing impossible. . In addition, we ourselves pro- grammed various aspects of the Star design , , and BravoX. At the same application. Silva went on to write De- on the Alto. time that the Learning Research Group luxepaint, a paint program for PCs. was developing Smalltalk for the Alto, Star. To develop Star and other office others at PARC, mainly . Fancy graphics capa- systems products, Xerox created the and , were writing an ad- bilities in a workstation are of little use Systems Development Department. SDD vanced document editing system for it without hard-copy capability to match it. was staffed by transferring people from called Bravo. Because it made heavy use of Laser printing, invented at PARC, pro- other parts of Xerox, including PARC, as the Alto’s bitmapped screen, Bravo was vided the necessary base capability, but well as by hiring from outside. Thus, con- unquestionably the most WYSIWYG text computers needed a uniform way to de- trary to what has often been stated in the editor of its day, with on-screen underlin- scribe output to laser printers. For this pur- industry press, Star was not developed at ing, boldface, italics, variable font fami- pose, developed the Press PARC, but rather in a separate product- lies and sizes, and variable-width charac- page-description language. Press was development organization. ters. It allowed the screen to be split, so heavily used at PARC, then further devel- When SDD was formed, a decision was different documents or different parts of oped into Interpress, Xerox’s commercial made to use Mesa, an “industrial- the same document could be edited at once, page-description language and the lan- strength” dialect of Pascal conceived at but did not operate in a windowed environ- guage in which Star encodes printer out- SRI and further developed at PARC, as the ment as we use the term today. Bravo was put. Some of the developers of Interpress primary product . widely used at PARC and in Xerox as a later formed Adobe Systems and devel- SDD took over development and mainte- whole. oped Postscript, a popular page descrip- nance of Mesa from the Computer Science From 1976 to 1978, Simonyi and others tion language. Laboratory at PARC, freeing CSL to de- rewrote Bravo, incorporating many of the velop Mesa’s research successor, Cedar. new user-interface ideas floating around Laurel and Hardy. A network of per- PARC at the time. One such idea was sonal workstations suggests electronic Star hardware. Star is often discussed modelessness,promoted by Larry Tesler3 mail. Though electronic mail was not in- as if it were a computer. In fact, Star is a and exemplified in Tesler’s prototype text vented at PARC, PARC researchers body of software.* However, using the editor, Gypsy. Simonyi et al. also added (mainly Doug Brotz) made it more acces- styles, enhancing users’ ability to control sible to nonengineers by creating Laurel, the appearance of their documents. The a display-oriented tool for sending, re- new version was called BravoX. ceiving, and organizing e-mail. The expe- *The official name for Star was the Xerox 8010 Infor- mation System. The machine was called the 8CNlO Se- Shortly thereafter, Simonyi joined Mi- rience of using Laurel inspired others to ries Network Systems Processor. Originally, “Star” crosoft, where he led the development of write Hardy for an Alto successor ma- was only an internal name.

September 1989 23 name Star to refer to the machine is under- ing what might have seemed like new op- machines, development aids for Mesa standable since the machine was designed erations to be handled by existing com- programmers were lacking. in conjunction with the software to meet mands. When the 8000 Series workstation be- the needs of the software design. This is in When SDD was formed, it was split be- came available, the systems group within sharp contrast to the usual approach, in tween two locations: Southern California SDD began working on a suitable develop- which software is designed for existing (El Segundo) and Northern California ment environment. Known internally as computers. (Palo Alto). Few people were willing to Tajo and externally as Xerox Develop- The 8000 Series workstation was based transfer one way or the other, leaving SDD ment Environment (XDE), the completed upon a microcoded proc&sor designed with the choice of losing many competent development environment and the numer- within Xerox especially to run the object engineers or being creative. SDD’s man- ous tools written to run in it were quickly code produced by the Mesa compiler. Be- agement took the creative route: they put adopted by programmers throughout sides being microprogrammed to run the Ethernet to work, attaching the devel- SDD. Star’s later improvements adopted Mesa, the processor provided low-level opment machines at both sites to a net- many good ideas from Tajo. operations for facilitating display opera- work, connecting the two sites with a 56- tions. For example, the bitblt operation for kilobit-per-second leased line, encourag- ViewPoint. Though Star’s introduc- manipulating rectangular arrays of screen ing heavy use of electronic mail for work- tion at NCC ‘81 was lauded in the industry pixels is implemented as a single instruc- related communication, and developing press, initial sales were not what had been tion. As sold, the machine was configured tools for facilitating distributed, multi- hoped. Almost immediately, efforts were with at least 384 kilobytes of real memory party development. launched to improve its performance, ex- (expandable to 1.5 megabytes), a local As might be expected from Star’s ori- tensibility, maintainability, and cost. hard disk (10, 29, or 40 megabytes), a 17- gins, most of the design and prototyping inch display, a mechanical mouse, an work was done in Palo Alto, whereas most Viewpoint software. Even before Star eight-inch floppy disk drive, and an Eth- of the implementation was done in El was released, the implementors realized ernet connection. The price was initially Segundo. Though this split was handled that it had serious problems from their $16,500 with software. creatively, some of Star’s designers now point of view. Its high degree of integra- Even though the machine was designed believe it caused problems not overcome tion and user-interface consistency had to run Star, it also ran other software. In by extensive use of e-mail. For example, been achieved by making it monolithic: addition to selling it as the 8010 “Star” the implementors did not benefit from the system “knew” about all applica- workstation, Xerox sold it as a server ma- much of the prototyping done at PARC. tions, and all parts of the system “knew” chine and as an and a Smalltalk The development process has been re- about all other parts. It was difficult to cor- workstation. counted in detail elsewhere6 and will not be rect problems, add new features, and in- repeated here. Suffice it to say that the Star crease performance. The monolithic Star software. Alhough Star incorpo- development effort architecture also did not lend itself to dis- rated ideas from a number of predecessors, tributed, multiparty development.

it still required a mammoth design effort to l involved developing new network This created pressure to rewrite Star. pull all of those ideas - as well as new protocols and data-encoding schemes Bob Ayers, who had been heavily involved ideas - together to produce a coherent when those used in PARC’s research in the development of Star, rewrote the design. According to the original design- environment proved inadequate; infrastructure of the system according to

ers, “. . . it was a real challenge to bring l involved a great deal of prototyping the more flexible Tajo model. He built, on some order to the different user interfaces and user testing; top of the operating system and low-level

on the Alto.“’ About 30 person-years went l included a late redesign of the proces- , a “toolkit” for building into the design of the user interface, func- sor; Star-like applications.

tionality, and hardware. l included several software redesigns, In the new infrastructure, transfer of To foster uniformity of specifications rewrites, and late additions, some data between different applications was as well as thoughtful and uniform design, based on results from user testing, handled through strict protocols involv- Star’s designers develo+d a strict format some based on marketing considera- ing the user’s selection, thus making appli- for specifications. Applications and sys- tions, and some based on systems con- cations independent from one another. tem features were to be. described in terms siderations (see below); The object-oriented user interface, which

of the objects that users would manipulate l included a level of attention to the re- requires that the system associate applica- with the software and the actions that the quirements of international custom- tions with data files, was preserved by software provided for manipulating ob- ers unmatched in the industry; and having applications register themselves

jects. This “objects and actions” analysis l left much of what was in the Star Func- with the system when started, telling it was supposed to occur at a fairly high level, tional Specification unimplemented. which type of data file they correspond to without regard to how the objects would and registering procedures for. handling actually be presented or how the actions Tajo/XDE. Since the machine upon keyboard and mouse events and generic would actually be invoked by users. A full which Star ran was developed in parallel commands. User-interface consistency specification was then written from the with the software, it was not available was fostered by building many of the stan- “objects and actions” version. This ap- early-on for use as a software development dards into the application toolkit. The de- proach forced designers to think clearly platform. Early prototyping and develop- velopment organization completed the about the purpose of each application or ment was done on Altos and successor re- toolkit and then ported or rewrote the exist- feature and fostered recognition of similar search machines developed at PARC. ing applications and utilities to run on top operations across specifications, allow- Though the Mesa language ran on these of it.

24 COMPUTER Other software changes included . stylesheets. for facilitating control of even than M ord-processing in expanding document appearance. the personal computer re\ olution beyond . the addition of several applications enpineers and hobb\,ists into business. and utilities. including a Free-Hand Lessons from E\entualll realizing that Star’s closed- Drawing prograrn and an IBM PC ne\s M as a problem. Xerox replaced it M ith emulation application: experience VieuPotnt. a more “open” sl3tcrn that . optional window tiling constraints. so altoNs user\ to pick and choose applicu- that users can have overlapping bill So what have we learned from all this’? ttons that the) need. including a spread- dews if desired: We believe. the following: sheet and IBM PC software. Apple Corn- . rcdexigned screen graphics (icons. puter learned the \ame lesson \\ith its LIU windous. property sheets. command Pay attention to industr! trends. computer and similarI> replaced it with a buttons. and menus) to accommodate Partly out of excitement over what they cheaper one ha\ ing it rnorc open dtx are a smaller screen and to meet the de- were doing. PARC researchers and Star’s architecture: the Macintosh. mands of a more sophisticated public: designers didn’t pa> enough attention to

and the “other” personal computer rrvolu- linofi your competition. Star’s initial . impro\ ‘ed performance tion occurring outside of Xerox. B) the per-workstation price uas near that ot late seventies. Xerox had its o&n powerful time-shared . dedicated To the fact that the new \y\- technical tradition (mouse-driven. nrt- uord-processors. and other Glared con- tern has 3 substantial miprovemcnt over WOI -hcd workstations uith large bitmap- puting facilrties. Star was. hour\er. corn the old. the name uas changed from Star to ped screens and multIpIe. simultaneou\ peting for desktop \pace with microcon- Viewpoint. VieM Point I .O was released in applications). blinding Star’\ designers to puter-based PCs. VleNPoint has cor- 19x5. the need to approach the marhet with rccted that problem: The 6085 co\ts about cheap. stand-alone PCs. The result uas a the same as its competition. I ic~f’oi~l Irrrr-tlw~ur~ In addition to re- product that uas highly unfamiliar to its vising the software. Xerox brought the intended customers: businesses. Noua- Establish firm performance goals. hardu are up to date by designing a com- days. of course. such systems are no longer Star‘s designers should have established plctelq neu cchicle for VtewPoint: the unusual. performance goals. documented them 111 60X5 workstation. The new machine was DeLcloping Star and Viewpoint In- the functional specifications. and stuck to de\ipned to take advantage of advances in Lolved developing several enabling tech- thern as the! dehcloped Star. Where per- integrated circuitry. reduction5 in rnern- nologies. for networking. cornmunicat- formance goats couldn‘t be met. the corrc- ory costs, new disk technologies. and new ing with servers. describing to laser spending functionality should have been standards in keyboard design. as well a5 to printer,. and developing software. At the cut. provide IBM PC compatibility. The 60X5 time they were developed, these technolo- In lieu of speed. the designer\ should workstation has a Mesa processor plus an gies were unique in the industry. Xerox habe made the user interface more respon- optional IBM-PC-compatible processor. elected to keep them proprietary for fear of \ive. Designing systems to handle user one megabyte of real memory (expand- losing its competitive advantage. U’lth input more intelligentI> can mahc them able to 3 megabytes). a hard disk (I(1 to X0 hindsight. ue can say that it might habe more responsive without necessaril! megabytes). a choice of a IS- or a 19.inch been better to release these technologies mahing them execute functions taster. display. an optical mouse. a ne\* key- into the public domain or to market them The) can operate as>,nchronouslq u Ith board. a 5%mch floppy disk drive. and. of early. so that they rnight have become in- respect to user input. making use of bach- course. an Ethernet connection. The base dustry standards. Instead. alternative ap- ground processes. keeping up uith impor- cost was initially $6.340 with the Vieu- proaches developed at other companies tant user action\. dela),ing unimportant Point software. Like the 8010. the 6085 is have become the industry standards. tasks (such as refreshing irrelevant areas sold as a vehtcle for Interlisp and Smalltalk Xerox’s current participation In the devel- of the screen) until time permits. and ship- as well as for Viewpoint. opment of various industry standards indi- ping tasks called for by early user actlons cates its desire to reverse this trend. but rendered moot by later ones. Vieu- Rcc,cr~t l’iclw,Poinf c~/~on~q~~s.The re- Point nob makes use of background proc- cently released Viewpoint 2.0 adds man\ Pay attention to what customers esses to increase its responsiveness. features relevant to desktop publishing. want. The personal computer revolution These include has shown the futility of trying to antici- Avoid geographically split develop- pate all of the applications that customers ment organizations. Having a develop-

l Xerox ProIllustrator. a new vector WIII want. Star should have been designed ment organization split between Palo Alto graphics editing application designed from the start to be open and extensible by and El Segundo was probably a mlstahe. mainly for professional illustrators: users, as the Alto was. In hindsight, exten- less for reasons of distance per se than for

l Shared Books, support for groups of sibility was one of the keys to the Alto’s lack of shared background in “PARC- users working on multipart docu- popularity. The problem wasn’t that Star style” computing. However. the adverse ments; lacked functionality, it was that It didn’t effect of sheer distance on communrcation

l a Redlining feature. for tracking dele- have the functionality customers wanted. was certainly a factor. tions and insertions in documents: An example is the initial lack of a spread-

l cursor keys, for moving the insertion sheet application. The designers failed to Don’t be dogmatic about the Desktop point during keyboard-intensive appreciate the significance of this applica- metaphor and direct manipulation. Di- work; and tion, which may have been more important rect manipulation and the Desktop meta-

September 1989 25 phor aren’t the best way to do everything. Though the reorientation of the industry spective. Finally, we acknowledgethe helpful Remembering and typing is sometimes away from batch and time-shared comput- suggestionsmade by the first and second au- thors’ colleaguesat US West Advanced Tech- better than seeing and pointing. For ex- ing toward personal computing had noth- nologies and several anonymous reviewers to ample, if users want to open a file that is one ing to do with Xerox, PARC, or Star, it was improve the quality of the article. of several hundred in a directory (folder), an important part of the computing phi- the system should let users type its name losophy that led to Star. rather than forcing them to scroll through References the directory trying to spot it so they can select it. tar has had an indisputable influ- 1. D.C. Smith et al., “The Star User Interface: ence on the design of computer sys- An Overview,” Proc. AFIPS Nat’1 Com- puter Conf.. June 1982,pp. 515-528. Many aspects of Star were correct. S tems. For example, the Lisa and Though certain aspects of Star perhaps Macintosh might have been very different 2. D.C. Smith, “Origins of the Desktop Meta- should have been done differently, most of had Apple’s designers not borrowed ideas phor: A Brief History,” presentedin a panel the aspects of Star’s design described at from Star, as the following excerpt of a discussion, “The Desktop Metaphor as an Approach to User Interface Design,” in magazine interview of Lisa’s design- the beginning of this article have with- Byte Proc. ACM Annual Conf., 1985,p. 548. stood the test of time. These include ers shows: 3. L. Tesler, “The Smalltalk Environment,”

l Iconic, direct-manipulation, object- Byte: Do you have a Xerox Star here that you Byte, Vol. 6, No. 8, Aug. 1981,pp. 90-147. oriented user interface. The days of cryp- work with? Tesler: No, we didn’t haveone here..We went 4. L. Winner, “Mythinformation,” Whole tic command languages and scores of to the NCC [National ComputerConference] Earth Review, No. 44, Jan. 1985. commands for users to remember (a la whenthe Star was announcedand looked at it. Unix and MS-DOS) should have passed And in fact it did have an immediateimpact. 5. J. Becker, “Multilingual Word long ago. A few monthsafter looking at it we madesome Processing,” Scientific American, Vol. changesto our user interface basedon ideas 251, No. 1, July 1984, pp. 96-107. (See l Generic commands and consistency that we got from it. For example,the desktop “Further reading” for other articles on in general. Even Macintosh could use managerwe had before was completely dif- Star’s multilingual capability.) some lessons in this regard: the Duplicate ferent; it didn’t use icons at all, and we never command copies files within a disk, but lied it very much.We decidedto changeours 6. E.F. Harslem and L.E. Nelson, “A Retro- users must drag icons to copy them across to the icon base.That was probably the only spective on the Development of Star,” thing we got from Star, I think. Most of our Proc. Sixth Id1 Conf. on Sofiware Engi- disks and must use Copy-Paste to copy Xerox inspiration was Smalltalk rather than neering, Sept. 1982, Tokyo, Japan. Re- anything else. Star.7 printed in Office Systems Technology, l Pointing device. Although cursor OSD-R8203A, Xerox Corp., pp. 261-269. keys have some advantages and certainly Elements of the Desktop metaphor ap- I G. Williams, “The Lisa Computer would have enhanced Star’s market ap- proach also appear in many other systems. System,” Byte, Vol. 8, No. 2, Feb. 1983,pp. peal (as they have Viewpoint’s), Star’s The history presented here has shown, 33-50. designers stand by the system’s primary however, that Star’s designers did not in- reliance on the mouse. This does not imply vent the system from nothingness. Just as 8. Ofice Systems Technology, OSD-R8203A, Xerox Corp. a commitment to the mouse per se, but it has influenced systems that came after it, rather to any pointing device that allows Star was influenced by ideas and systems 9. T. Nate, “The Macintosh Family Tree,” quick pointing and selection. As inter- that came before it. It is difficult to inhibit MacWorld, Nov. 1984, pp. 134-141. faces evolve in the future, high-resolution the spread of good ideas once they are ap- touch screens and other more exotic de- parent to all, especially in this industry. vices may replace mice as the pointing Star was thus just one step in an evolution- Further reading devices of choice. ary process that will continue both at l High-resolution display. Memory is Xerox and elsewhere. That is how it should For interested readers, we provide here a now cheap, so the justification for charac- be.0 guide to further readingon Star and relatedtop- ter displays is gone. of this retrospecive will appearin an upcoming l Good graphic design. Screen graphics Prentice-Hall book series. Also, the September designed by computer programmers will Acknowledgments issueof ZEEE Spectrum includesan article, “Of not satisfy customers. The Star designers Mice and Menus,” on graphical user interfaces recognized their limitations in this regard When Star wasanuounced, several articles on past and present.The authorsof that article con- sulted severalpeople who contributed to or are and hired the right people for the job. As it appearedin trade magazines,journals, and ai conferences.Many of these were reprinted in mentioned in this Star retrospective. color displays gain market presence, the the Xerox publication Office SystemsTechnol- In the following bibliography, readings are participation of graphic designers will ogy.* Sincethen, severalmore articleshave been categorized according to whether they pertain become even more crucial. published that are relevant to this retrospective. to work done prior to the establishmentof Xerox They include an article from MacWorld9 that PARC, to work done at PARC, to the original l 16-bit character set. An eight-bit char- described the historical antecedentsof Apple Star design done at Xerox SD?,8 to enhance- acter set (such as ASCII) cannot accom- Computer’s Lisa and Macintosh computers, mentsto Star and ViewPoint, and to work on re- modate international languages ade- which share much of Star’s history. This retm- lated topics. quately. Star and Viewpoint’s use of a spfzctiveowes a greatdeal to thoseprevious writ- 16-bit character set and of special typing ings. We also acknowledgethe valuable contri- Pre-Xerox and rendering algorithms for foreign lan- butions that Joe Becker, Bill Mallgren, Doug Carothers, Linda Bergsteinsson, and Randy Bush, V., “As We May Think,” Atlantic guages is the correct approach. Polen of Xerox, and Bob Ayers, Ralph Kimball, Monthly, Vol. 176, No. 1, July 1945, pp. lOl- l Distributed, personal computing. Dave Fylstra, and John Shochmade to this retro- 108.

26 COMPUTER Engelbart, D.C., and W.K. English, “A Re- Korean,” Computer, Vol. 18, No. 1, Jan. 1985, search Center for Augmenting Human Intel- pp. 27-34. lect,” Proc. FJCC, Vol. 33, AFIPS, 1968, pp. Becker, J., “Arabic Word Processing,” Comm. 395-410. ACM, Vol. 30, No. 7, July 1987,pp. 600-610. English, W.K., D.C. Engelbart, and M.L. Ber- Bewley, W.L., et al., “Human Factors Testing man, “Display-Selection Techniques for Text in the Design of Xerox’s 8010 Star Office Manipulation,” IEEE Trans. Human Factors in Workstation,” Proc. ACM Conf. on Human Electronics, HFE-8, 1967, pp. 21-31. Factors in Computing Systems, 1983, pp. 72- Kay, A.C., The Reactive Engine, PhD thesis, 77. University of Utah, Salt Lake City, 1969. Bushan,A., and M. Plass,“ The InterpressPage Sutherland, I.E., Sketchpad: A Man-Machine and Document Description Language,” Com- Jeff Johnson is a user-interface researcherat Graphical Communications System, PhD the- puter, Vol. 19, No. 6, June 1986,pp. 72-77. Hewlett-Packard Labs. While working on this sis, MIT., Cambridge,Mass., 1963. Curry, G., and R. Ayers, “Experience with article, he was a memberof the AdvancedUser Interfaces Group of US West Advanced Tech- Traits in Star,” Programming Productivity: Is- nologies. Before that, he worked at Xerox de- Xerox pre-Star sues for the Eighties, IEEE Computer Society Press,Los Alamitos, Calif., 1986, Order #681. signing and implementing enhancementsto the Card, S., W.K. English, and B. Burr, “Evalu- Star/ViewPoint system, and at Cromemco as a Halbert, D., “Programming by Example,” softwareengineer and manager.He receivedBA ation of Mouse, Rate-Controlled Isometric Xerox OSD Tech. Report OSD-T84-02, 1984. Joystick, Step Keys, and Text Keys for Text Se- and PhD degreesin cognitive psychology from lection on a CRT.” Emonomics. Vol. 21, 1978. Lewis, B., and J. Hodges,“ Shared Books: Col- Yale and Stanford universities, respectively. pp. 601-613. - laborative Publication Managementfor an Of- fice Information System,” Proc. ACM Conf. on Ellis, C., and G. Nutt, “Computer Scienceand Office Information Systems, 1988. Office Information Systems,” Xerox PARC Tech. Report SSL-79-6, 1979. Geschke,CM., J.H. Morris, Jr., and E.H. Satter- Miscellaneous thwaite, “Early Experience with Mesa,” Comm. ACM, Vol. 20, No. 8, 1977,pp. 540-553. Bly, S., and J. Rosenberg,“ A Comparisonof Ingalls, D.H., “The Smalltalk Graphics Tiled and Overlapping Windows,” Proc. ACM Kernel,” Byte, Vol. 6, No. 8, Aug. 1981, pp. 168- Conf. on Computer-Human Interaction, 1986, 194. pp. 101-106. Kay, A.C., and A. Goldberg, “Personal Dy- Goldberg, A., Smalltalk-80: The Interactive namic Media,” Computer, Vol. 10,No. 3, March Programming Environment, Addison-Wesley 1977, pp. 31-41. Publishing, Reading, Mass., 1984. Kay, A.C., “Microelectronics and the Personal Goldberg, A., and D. Robson, Smalltalk-80: Computer,” Scientific American, Vol. 237, No. The Language and its Implementation, Ad- 3, Sept. 1977,pp. 230-244. dison-Wesley Publishing, Reading, Mass., Smith, D.C., Pygmalion: A Computer Program 1984. to Model and Simulate Creative Thought, Birk- Halasz, F., and T. Moran, “Analogy Consid- hauserVerlag, Base1and Stuttgart, 1977. ered Harmful,” Proc. ACM Conf. on Human Thacker, C.P., et al., “Alto: A PersonalCom- Factors in Computing Systems, Gaithersburg, Teresa L. Roberts is a user-interfaceresearcher puter,” in Computer Structures: Principles and MD, 1982, pp. 383-386. in the Advanced User Interfaces Group at US Examples, D. Siewioek, C.G. Bell, and A. New- Houston, T., ‘The Allegory of Software: Be- West Advanced Technologies. She spent 10 ell, eds., McGraw Hill, New York, 1982. yond, Behind, and Beneath the Electronic years at Xerox, designing and evaluating parts Desk,” Byte, Dec. 1983,pp. 210-214. of the Star user interface.She receiveda PhD in Star Johnson,J., “Calculator Functions on Bitmap- computer science from . ped Computers,” SIGCHI Bulletin, Vol. 17, Curry, G., et al., “Traits- An Approachto Mul- No. 1, July 1985,pp. 23-28. tiple-Inheritance Subclassing,” Proc. ACM Johnson, J., “How Closely Should the Elec- Conf. on Office Automation Systems (SIGOA), tronic Desktop Simulate the Real One?” 1982. SIGCHI Bulletin, Vol. 19, No. 2, Oct. 1987,pp. Dalal, Y.K., “Use of Multiple Networks in the 21-25. Xerox Network System,” Computer, Vol. 15, Johnson, J., “Modes in Non-Computer De- No. 10, Oct. 1982,pp. 82-92. vices,” in press,Int ’l J. Man-Machine Studies, Johnsson,R.K., and J.D. Wick, “An Overview 1989. of the Mesa ProcessorArchitecture, ” SlGPlan Johnson,J., and R. Beach,“ Styles in Document Notices, Vol. 17, No. 4, 1982. Editing Systems,” Computer, Vol. 21, No. 1, Lipkie, D.E., et al., “Star Graphics: An Object- Jan. 1988,pp. 32-43. Oriented Implementation,” Computer Graph- Malone, T.W., “How Do People Organize ics, Vol. 16, No. 3, July 1982,pp. 115-124. Their Desks: Implications for the Design of Shoch, J.F., et al., “Evolution of the Ethernet Office Information Systems,” ACM Trans. on Local ComputerNetwork, ” Computer, Vol. 15, Office Information Systems, Vol. 1, No. 1, 1983, No. 9, Aug. 1982,pp. 10-27. pp. 99-l 12. Smith, D.C., et al., “Designing the Star User Rosenberg, J.K., and T.P. Moran, “Generic Interface,” Byte, Vol. 7, No. 4, April 1982, pp. Commands,” Proc. First Int’l Conf. on Human- William Verplank is an interaction&sign con- 242-282. Computer Interaction (Interact-84), 1984, pp. sultant with IDTwo and a lecturer in humanfac- Sweet,R.E., and J.G. Sandman,Jr., “Empirical 1,360-l ,364. tors and design at Stanford University. He Analysis of the Mesa Instruction Set,” SlGPlan Teitelman, W., “A Tour Through Cedar,” worked at Xerox from 1978to 1986,doing user Notices, Vol. 17,No. 4, 1982. IEEE Software, April 1984. testing and user-interface design for Star and ViewPoint. He receiveda BS in mechanicalen- Star/Viewpoint enhancements Whiteside, J., et al., “User Performancewith gineering from Stanford University and MS and Command, Menu, and Iconic Interfaces,” PhD degrees in man-machine systems from Becker, J., “Typing Chinese, Japanese,and Proc. ACM SIGCHI ‘85, 1985,pp. 185-191. MIT.

28 COMPUTER At GTE’s Computer and Intelligent Systems Laboratory, we are creating opportunities to apply new ideas to research and development projects for improved information and telecom- munication systems. By expanding our scope and responsibility, we can better support GTE’s telecommunications businesses. And our ac- tivities create challenges for individuals at the MSiPhD level in Computer Science. Join us as we continue to make history in teIecommunica- tions technology. Our present areas of interest include: Distributed Operating Systems We are currently growing a group that is conduct- ing research in distributed operating systems and distributed transaction systems. These systems will unify a network of cooperating autonomous, heterogeneous processors and replicated data- bases. Issues concern not only the tradeoffs between network and distributed operating systems, but the interoperability of software components imple- mented on a mix of platforms and languages; included are such topics as transport, access, apphcation, distributed control and control migration. Research is conducted using synthesis and prototyping with emphasis on architectural and applicability issues. We are looking for individuals at all levels of ex- perience, however, we require a PhD in Computer Science along with a familiarity with the various current approaches to the design of distributed systems. Significant experience with at least one such system is highly desirable. Intelli ent Database f?ystems Our Distributed Object Management (DOM) pro- ject is conducting research into interconnectivity and intelligent interoperability among heterogeneous computer systems. We require PhD-level re- searchers at all levels with a minimum of 2 years’ experience in databases, operating systems, distributed systems or artificial intelligence. Software Reusability Topics of interest in the area of software reusability include domain analysis, object-oriented development, reuse centered software pro- cesses, tools support for reusability and software classification. This area requires an MSiPhD in Computer Science and a minimum of 2 years’ experience in at least one of the above areas. GTE Laboratories offers attractive facili- ties located in a quiet, wooded setting just outside of Boston, as well as a highly competitive salary and benefits package. We invite you to send a resume to Vanessa Stern, GTE Laboratories, Inc., Box CACMS, 40 Sylvan Road, Waltham, MA 02254. An equal opportunity emnlover. M/F/H/V. Laboratories Keader~ ma! contact Jett Johnxm at He\* tett-P;~cLard Lab\. t SOt Page m .Mlll Rd.. I ALI, Palo Alto. CA 03303. THE POWER IS ON September 1989