practice

Doi: 10.1145/2330667.2330682

Article development led by queue.acm.org

Backward compatibility always
trumps new features.
By DaViD ChiSnaLL
a new
objective-C
Runtime:
From
Research to
Production

ON My Way out of academia, before Cambridge persuaded me to return, the last paper I wrote was a description of a new objective-C runtime for use by the Étoilé project. 1 An objective-C implementation requires two components: a runtime library that implements the dynamic parts of the language and a compiler that emits calls to this library.

When I wrote that paper in 2009, there were two choices for objective-C: Apple’s stack, with a number of new features but an implementation that relied

on some proprietary features and ones that shipped only with Darwin; and the GNU Compiler Collection (GCC) stack maintained by the Free Software Foundation. The GCC implementation had several limitations, not least of which was the license, which effectively prevented the runtime library from being used with any compiler other than GCC (this situation was resolved in later versions).

One of the goals of Étoilé was that no program should contain more than 1,000 nonreusable lines of code. This typically requires a lot of help from do-main-specific languages, which is the research area where I spend most of my time. I had implemented a proof-of-concept Smalltalk compiler on top of the GCC runtime, but it had some

IllustratIon By gaVIn potenZa