Previous Table of Contents Next


In May 1983 the first of three Smalltalk-80 books (Goldberg & Robson, 1983) was published, followed shortly after by two others (Krasner, 1984; Goldberg, 1984). The first book introduced the language and its implementation, as reviewed by developers in the four participating companies. The second was a collection of papers written about the porting experience and included the primary documentation on the LOOM object-reclamation system. The third detailed the programming environment tools and user interface. The publisher announced a fourth book about building applications, although no such book was contracted and no further writing energy could be found.

The then head of the Xerox Office Systems Group and inventors and sellers of the Star Workstation belatedly decided that the Smalltalk publication plans were a potential problem. How should they answer a customer request for Smalltalk on the Star Workstation? Xerox had agreed to distribution of the system to universities, but withheld commercial distribution for a year. During that time, we were obligated to port Smalltalk-80 to the Star Workstation. Because the Star office software was written in Mesa (Mitchell, 1979), and because Mesa was the language of choice for the Star, we were also obligated to write the virtual machine using Mesa. We were fortunate to hire two very good developers (Paul McCullough and Frank Zdybel), whose Smalltalk/Mesa we named Molasses in honor of the speed with which it ran. The implementation marks the first time we had created the Smalltalk programming environment as embedded in another programming environment, so we ultimately found the exercise interesting, albeit a waste of skilled manpower. We were relieved that no Star workstation customer ever requested a Smalltalk system.

The Star experience offered us another opportunity to present Smalltalk as a business solution for Xerox. Sales of the Star did not meet expectation, despite a very successful initial Star launch that certainly stunned the computer industry with the sophistication of its design and function. The then head of the PARC Systems Center (the half of a temporarily divided PARC that contained CSL) cooperated with the Dallas-based management team of the Office Systems Group in brainstorming potential cures for the Star sales problem. Four PARC researchers were invited to Dallas, two from CSL and two from SCG (Ingalls and me), to discuss the problem. The CSL recommendation was to design and build a new workstation that was cheaper and faster. SCG suggested that the problem was that the initial buy-in to the Star required too much know-how and expense (the Star initial configuration consisted of three workstations, a printer, and an Ethernet, at a time when people did not understand how to drive a mouse or scan a display screen). As a result, SCG designed Twinkle, a 68010-based personal workstation, with simple text editing and freehand graphics (Smalltalk applications), and submitted the design with intention to build and deliver in 1983. The Twinkle idea was also interesting because it would have given Xerox a pathway to ride the Motorola and Intel technology performance curves.

We never received a response to our proposal (likely changes in management in Dallas left our recommendation in the “unanswered mail” box; also, we were also politically incorrect in offering freehand graphics to the Star designers, who demonstrated deep religious convictions about structured graphics). Given the success in 1984 of the Macintosh launch, we believe Twinkle would indeed have been a constructive solution.

3.5.4. The CIA Analyst and Joshua

Why were we so confident that we could build a Smalltalk-based Twinkle solution in just over a year? Our plan was to leverage work already in progress in the Xerox Dallas Development Center, motivated largely by interest from the CIA.

Smalltalk-80 was implemented on the Xerox 1100 (Dolphin), making the Smalltalk system more generally available within Xerox. In particular, the 1100 implementation gave the Pasadena-based XSIS group its first access to Smalltalk.13 Several early large Smalltalk-80 applications were produced by XSIS, notably a layout system for the New York Times Book Review. XSIS primarily sold into the U.S. intelligence community with a charter to do advanced designs based on research goals directed from within the intelligence community. The CIA was no stranger to PARC. A group of AI researchers from BBN had joined PARC SSL, bringing their government relations with them. As this folk tale is recalled, two very astute technology managers from the CIA saw the Smalltalk demos on a visit to PARC and requested that their new ideas for an Analyst Workstation be built using the language system. By the time we discovered this activity, a first prototype was completed and in use at NPIC, where a professional photographer was using Smalltalk to design new graphical interfaces for managing his desktop information. All this preceded the release of the Smalltalk-80 books. As a result, this photographer figured out how to hack together Smalltalk code by reading our code to see how our tools were implemented.


13XSIS engineers also ported Smalltalk-80 to the follow-on Star Workstation, code named Dandilion.

Fear set in. As far as we could tell, XSIS was building a product—the Analyst—using a research result. No matter how good we were as system developers, we fell short of finishing our work to product level. We clearly had not stopped short enough. The XSIS Analyst was built on an unsupported programming language and development environment. Even though we were just researchers, we knew that this behavior was a formula for disaster.

Then we discovered the group in Dallas. A small group of very good system developers had decided that Smalltalk was worth their attention. Encouraged by interest from the CIA, they designed a 68000 add-on board for the Xerox 800 (an 8086-based personal computer built by Olivetti and sold successfully by Xerox). The add-on board, code named Joshua, had 1MB of RAM and ran Smalltalk adequately. Another design was based on the 68010 with 4MB of RAM, which improved performance. The Dallas team worked with SCG to learn about the PS virtual machine and make use of Deutsch’s 68000 development environment. Our hope was that we could transfer the research results to Dallas, and they would take ownership of product creation and support for the CIA and other future customers. Unfortunately and despite a large order already in place (we were told that the CIA alone wanted 10,000 Smalltalk boards), Xerox’s pricing caused the order to be canceled and then Xerox canceled Joshua.

Instead of our sending research results to Dallas, Dallas came to PARC, and we formed ParcPlace Systems.


Previous Table of Contents Next