Dead End: Taligent

Abstract

Taligent began as Pink, a project Apple started in 1988 to write the object-oriented successor to the Macintosh operating system from scratch. Pink was always two years from shipping. In 1991 Apple and IBM agreed to turn it into a joint company, and Taligent opened in Cupertino in March 1992 with about 170 employees, a CEO from IBM, and the task of building an operating system that would run on everyone’s hardware and loosen Microsoft’s hold on the PC. Hewlett-Packard bought in during 1994. By 1995 the investors had spent more than $100 million, the operating system had been cancelled, and the remaining product, a set of C++ frameworks called CommonPoint, ran only on IBM’s AIX Unix. Apple and HP left in December 1995, IBM laid off half the staff, and in January 1998 the company was dissolved into IBM. Its internationalization code survives in Java and in the ICU library on most of the world’s smartphones.

The Taligent logo. Image: Taligent, Inc. (IBM and Apple Computer), public domain, via Wikimedia Commons.

The Gang of Five

By 1987 the Macintosh system software was a few years old and already showed its 1984 origins: no memory protection, no preemptive multitasking, and a design that assumed one program at a time (see The macOS Lineage). The group responsible for it, then called Macintosh System Software, was making small improvements such as support for hard disks larger than 32 MB. Five senior engineers, Erich Ringewald, David Goldsmith, Bayles Holt, Gene Pope and Gerard Schutten, wanted more interesting work and threatened to leave if they did not get it. They became known as the Gang of Five, and their managers gave in.

In March 1988 the Gang, their management and the software manager Mike Potel met at the Sonoma Mission Inn to plan the future of the Mac operating system. (Owen Linzmayer’s Apple Confidential 2.0 dates the start of the project a year earlier.) The ideas went onto index cards in three colours. Blue cards held incremental improvements that could ship soon. Pink cards held the larger changes: preemptive multitasking, componentized applications, a fully object-oriented system. Red cards held the radical ideas, such as speech recognition, said to be “pinker than Pink.” The blue cards became System 7, which shipped in 1991. The pink cards became a project of their own, and gave it its name.

Pink

Pink’s goals were total object orientation, memory protection, preemptive multitasking, built-in internationalization and advanced graphics. Its user interface was to be organised around people, places and things. Ringewald led it, with a target of shipping in about two years, in 1989.

The project moved into a warehouse on Bubb Road in Cupertino, which it shared with the team that later built the Newton. In its first months Pink had a little over 20 engineers while the Blue team had hundreds. Ringewald wanted Pink to remain a version of the Mac OS and to reuse existing code; he warned, “We’re going to have enough trouble just reimplementing the Mac.” David Goldsmith, who wanted a standalone system, resigned. A few months later a senior executive overruled Ringewald, three engineers from Apple’s research group were assigned to write a new microkernel, and Pink became a ground-up rewrite. Three more of the Gang of Five left.

Pink grew past 100 people by April 1989 and about 150 by 1990, when it was renamed the Object Based Systems group. As the 1989 target approached, Pink was an unstable, partly working prototype; by late 1989 it ran as a desktop operating system on Macintosh hardware. Ringewald, who later described the project as a “living, breathing, money-consuming thing” that was “out of control,” left Apple in 1990 for Jean-Louis Gassée’s new company, Be (see BeOS). Ed Birss took over. Inside Apple the project had become a running joke. Asked when Pink would ship, the answer was always the same: in two years.

The group was also secretive. Birss restricted outside access, and the Jaguar RISC workstation team needed an order from chief executive John Sculley to see the code. Other Apple groups, unable to look inside Pink, built their own versions of things Pink was meant to provide; QuickDraw GX, PowerTalk and parts of the Newton operating system duplicated its work. Gassée, then head of product development, urged Sculley to cancel Pink and blame him for it. Sculley refused.

The Alliance

Pink found a buyer in IBM. Apple’s Star Trek project, an experimental port of the Mac OS to Intel processors, had brought Apple into contact with Phil Hester, head of IBM’s RISC processor work, and talks between Sculley, Apple’s chief operating officer Michael Spindler and IBM’s Jack Kuehler followed. On 12 April 1991 Apple demonstrated Pink to IBM, which was impressed. Sculley said that summer that Apple had written 1.5 million lines of code for it.

On 2 October 1991 Apple, IBM and Motorola announced the AIM alliance. Motorola would manufacture a processor derived from IBM’s POWER architecture, the PowerPC, which became the basis of the Power Macintosh. Two joint ventures would supply software: Kaleida Labs, for a cross-platform multimedia scripting language, and Taligent, to bring Pink to market. The name was a blend of “talent” and “intelligent.”

Taligent was formally launched in February and March 1992 in Cupertino, at first in space leased from Apple, with more than 160 employees transferred from the Pink group. Its chairman and chief executive was Joe Guglielmi, an IBM executive; Ed Birss of Apple became chief operating officer. Guglielmi, a 30-year IBM veteran in charge of a staff hired mostly from Apple, was not inclined to defer to Apple tradition: “I’m tired of [Apple] folklore,” he said. “I want some data.”

The Microsoft Killer

Taligent’s mission was larger than replacing the Mac OS. In January 1993 The Wall Street Journal reported that Borland, WordPerfect and Novell backed the venture, and much of the industry expected Taligent’s success to depend on weakening Microsoft’s monopoly. The plan had three layers: TalOS, a native object-oriented operating system; TalAE, an application environment of frameworks that could also run on top of other operating systems; and TalDE, the development environment. Developers who saw early material gave it good reviews. By April 1993 Taligent had about 260 employees.

The design was built around frameworks rather than libraries. A conventional library is a set of functions that an application calls. A framework is a partly written application: the framework calls the programmer’s code at defined points, and the programmer supplies only what differs. Taligent’s frameworks covered text, graphics, documents, networking, international text and much else, all in C++ (see Bjarne Stroustrup and C++ and The OOP Revolution). The idea had worked before, in Apple’s own MacApp and in NeXT’s object frameworks, but nobody had tried to build an entire operating system that way.

IBM’s influence changed the foundation. Taligent dropped Pink’s own microkernel, Opus, for the IBM Microkernel that IBM was building as the base of Workplace OS, a project meant to run OS/2, Unix and other operating system “personalities” side by side on one kernel. Pink was to become one personality on top of it. Taligent’s first internal development environment ran on an IBM RS/6000 workstation with a PowerPC 601. In January 1994 Hewlett-Packard joined as a third partner with a 15 percent stake, for which it reportedly paid about $50 million. Taligent showed its first public demonstrations in March 1994 and began delivering code to its three investors; in June 1994 it shipped a developer preview called PEEK. Late in 1994 the application environment was renamed CommonPoint, and third-party CommonPoint applications were demonstrated at Comdex that November.

CommonPoint

In May 1995 Taligent cancelled TalOS. The company would sell only CommonPoint, a layer of more than 100 frameworks running on other vendors’ operating systems. CommonPoint 1.1 for AIX, IBM’s Unix, shipped on 28 July 1995. A beta for OS/2 followed in December, and versions for Windows NT, HP-UX and the Mac OS were promised.

The shipping product was heavy and expensive. The runtime cost $1,500; with the developer kit it cost $5,900, and the kit also required IBM’s $1,800 C Set++ compiler. CommonPoint added about 18 MB of memory overhead per machine, and Taligent recommended 32 MB of RAM in total. Experienced C++ programmers needed at least three months before they could approach their first application. Analysts called it “too little, too late.” The investors had spent more than $100 million, and the company had about 400 employees.

The management went next. Guglielmi left in September 1995 to join Motorola. His interim successor, Dick Guarino, died of a heart attack in October. On 17 November 1995 Apple and IBM announced the closure of Kaleida Labs, Taligent’s sister venture, and reports said Taligent might follow.

The Breakup

On 19 December 1995 Apple and HP sold their stakes, keeping licences to the technology developed so far, and Taligent became a wholly owned IBM subsidiary. About half of its 375 employees were laid off, including the entire sales and marketing team. What remained was renamed the Taligent Object Technology Center, stayed in Cupertino, and was given the job of folding Taligent technology into IBM products: the VisualAge development tools, IBM’s System Object Model, and OpenDoc, the compound-document architecture that Apple and IBM were promoting at the time. Debbie Coutant, a Taligent founder from Apple, became chief executive, with Mike Potel as chief technology officer.

The trade press was blunt. Computer Business Review reported in January 1996 that IBM had about 150 CommonPoint customers, mostly technical users evaluating it in laboratories, and that there were “no known examples of production code developed from Taligent kits.” Apple said it would develop Taligent technology “only if there is customer demand.” CommonPoint depended on a modern operating system underneath it, and on the Mac that system was Copland, Apple’s own replacement project. Apple cancelled Copland in August 1996 and announced on 20 December 1996 that it would buy NeXT, whose operating system became Mac OS X. IBM’s Workplace OS shipped a final beta in January 1996 and was discontinued without ever carrying CommonPoint.

Taligent’s last products were smaller. CommonPoint technology went into IBM’s VisualAge C++ 4.0 in February 1997, and in June 1997 Taligent released Places for Project Teams, a collaboration product built on Lotus Notes, at $49 per user. On 16 September 1997 IBM announced that Taligent would be dissolved by the end of the year. Most of the remaining engineers moved to IBM’s Santa Teresa laboratory, those with Java skills into a Java porting and tuning centre IBM ran with Sun and Netscape, and in January 1998 Taligent ceased to exist. IBM withdrew CommonPoint for OS/2 from the market on 3 August 1999.

What Survived

The part of Taligent that lasted was its text and internationalization group, led by Mark Davis. Taligent’s frameworks had handled the formatting of dates, numbers and currencies, the sorting of text by the rules of different languages, and the conversion between character sets, all built on Unicode (see Unicode and UTF-8). After IBM took over Taligent, Sun asked for better internationalization in its new Java language, and IBM had the Taligent group write the classes. They became the internationalization APIs of Sun’s JDK 1.1 in 1997 (see James Gosling and Java). IBM then ported the code to C and C++ and in 1999 released the libraries as open source, first as IBM Classes for Unicode and later renamed International Components for Unicode (ICU4C and ICU4J). In 2016 IBM handed ICU to the Unicode Consortium. It formats dates and numbers and sorts names on most of the world’s smartphones. Davis, a co-founder of the Unicode Consortium, later took the same work to IBM and Google.

Two smaller traces remain. In 2008 PC World ranked the Taligent operating system fourth on its list of the 15 greatest vaporware products of all time. And when Google published the first code of its Fuchsia operating system in 2016, the repository opened with a line that combined the colours of two Apple projects: Pink, and Purple, the code name of the iPhone.

Dead End

Pink failed before Taligent existed. A small group with a two-year target became a rewrite of everything, run in secret from the rest of Apple, while the company kept shipping and improving the old system it was meant to replace. Fred Brooks called this the second-system effect: the follow-up to a successful system collects every idea that was left out of the first (see Fred Brooks and the Mythical Man-Month). In Pink’s case the follow-up never had to ship, because the original kept selling.

The joint venture gave the project three owners with different goals. Apple needed a successor for the Mac. IBM wanted a personality for Workplace OS and a weapon against Microsoft, and HP wanted object technology for its Unix. The product that came out of the compromise ran on IBM’s AIX, cost $5,900 with its developer kit, and asked for 32 MB of RAM. By the time CommonPoint shipped in 1995, Windows 95 was a month away and the market Taligent had been formed to contest had been decided. The frameworks were technically well regarded, and there were almost no applications built on them.

Apple’s own escape came from outside: in December 1996 it bought NeXT, whose object-oriented frameworks had shipped on real computers since 1989.