Elevators, toasters, and the birth of Java

James Gosling says he became interested in computing after his father’s friend took him through the University of Calgary’s data center. As a teenager, he taught himself programming using punch cards, paper tape, and a Digital Equipment PDP-8. Physics researchers noticed him, hired him during high school, and had him write software for the ISIS-2 satellite: decoding data streams, correcting errors, and converting data into images. Working with physicists also taught him about the Earth’s magnetosphere and gave him early experience translating mathematical work into practical software.

Gosling rejects the idea that someone asked him to create a programming language. After earning a PhD from Carnegie Mellon, spending about a year and a half at IBM, and joining Sun Microsystems, he worked with consumer-electronics companies making products from elevators to VCRs. Sun’s belief that networking should be central led to discussions about interconnected devices. Gosling and roughly half a dozen colleagues built prototypes, first in C and then C++, and discovered that reliability problems were fundamentally software-engineering and compiler-technology problems.

Before 1995, engineers often traded reliability for performance, but Gosling’s consumer-electronics customers had a stricter rule: “that must not kill your customer.” Manufacturers also disliked proprietary CPUs because unlike interchangeable resistors or capacitors, each processor forced purchasing and engineering trade-offs around cost, power, features, and existing tools. Gosling’s earlier failed thesis proposal suggested an abstract machine that could separate software from physical processors. Combined with the need for security, portability, and dependable prototypes, this led toward a virtual machine and a language with garbage collection, strong typing, and defined arithmetic.

Gosling explains that older language specifications often labeled arithmetic “implementation specific” because IBM, Digital Equipment, HP, and others implemented floating point differently. IEEE 754 gradually standardized behavior, though Intel launched before every detail had settled, contributing to Java’s now-removed strictfp keyword. He recalls Intel’s early trigonometric implementation using a slightly incorrect value of pi for range reduction; numerical-analysis problems could result when calculating sine or cosine for huge angles, such as a motor shaft or airplane propeller. A discussion about portable software thus reached the fine details of numerical computation.

Java’s virtual machines varied widely by deployment. Smart cards, including many credit cards, use machines defined by the smart-card standard, while 1990s phones generally shared the machine but needed reduced libraries because they had little RAM. Data-center Java addressed large teams and projects through interfaces, classes, reusable class files, and binary portability. Gosling valued changing one source file and rebuilding a large system in about ten seconds. He was less enthusiastic about Java applications containing three to six competing HTTP implementations because teams failed to coordinate, even though HTTP itself retained much 1990s-era baggage.

Gosling says Java’s biggest response to C and C++ was eliminating pointer-based memory corruption and leaks rather than relying on debugging tools. He favors exceptions because they expose failures early, contrasting checked file-opening errors with C programs that silently produce empty logs. Modern Java garbage collectors handle large memories and concurrency while allowing millisecond-scale maximum pauses, though deterministic systems such as motor controllers or jet engines cannot tolerate even rare hash-table timing spikes. He sees coding agents as improving IDE-like help, especially for obscure APIs, but says they still generate buggy code and cannot replace a capable human.