2 Main Goals

S1 and SK had some very different goals. Many of the goals were the same, for example to be as efficient as C as well as cleaner and safer than Eiffel. However, each group had other goals not shared by both.

Some goals of S1 were:

• Support for separate compilation and multiple developers.

  Issue:

  Composability of code requires that it be possible to statically check classes over all possible parameterizations.

  Choice:

  This lead to S1 parameter type bounds and the restriction that overloading not occur between two unrelated abstract arguments (see section 6).

• Trivial porting to a wide range of platforms.

  Issue:

  Sather should be available on all platforms. This includes embedded control, the bulk of installed CPU use today. This can be achieved by a custom back end for each platform, if the manpower is available.

  Choice:

  Because ICSI has limited interest in compiler back-ends, S1 chose to use C as a portable assembler. Compiling through C is slow. This intensified the demand for separate compilation, and may have affected the choice for S1 arrays, see section 7 below.

• Extendability for threaded and distributed parallel programming.

  Issue:

  What about parallel programming?

  Choice:

  ICSI has been actively engaged in research on parallel and distributed programming since the inception of Sather. ICSI's vision of parallel programming is multithreaded and explicitly distributed. Multithreaded code imposes very different constraints than data parallel code, which in turn reflects on decisions about the semantics of serial code. (Eg. What is the semantics of a stream shared by more than one thread?)

• Rapid compilation.

  Issue:

  Karlsruhe has a substantial investment in compiler research, including quality compiler tools. Development of SK naturally used these.

  Choice:

  Because the resulting compiler is so much faster than compiling through C, SK has not been concerned with separate compilation.

• Pedogogy.

  Issue:

  SK was envisioned as a language with which one would teach undergraduates. One wants to be able to expose language constructs in a clean order, without forward references in the language to concepts students are not prepared for. The language is used in undergraduate as well as graduate courses. It is used for teaching both, imperative and object-oriented programming in Gerhard Goos: Vorlesungen über Informatik II, Springer Verlag, 1996

  Choice:

  Some SK constructs differ from S1. For example, in S1 is in SK, which avoids explaining about iterator notation when loops are introduced. Similarly, in SK local variables must be declared at the beginning of the block in which they appear and there is no type inference for declarations.

• Research vehicle for understanding library design.

  Some observations in constructing reliable libraries influenced SK. One goal was to avoid errors (either compile-time or run-time) which point into library code. For the motivation of structural conformance, see section 6.