[Paper Review] Steering Fragments of Instruction Sequences

J.A. Bergstra|arXiv (Cornell University)|Oct 14, 2010

Formal Methods in Verification26 references3 citations

TL;DR

This paper investigates why non-atomic propositional statements appear in steering fragments of instruction sequences despite their potential for transformation into equivalent atomic forms. It identifies implicit constraints—such as marginal side effects, detectable side effects, and real-time dynamics—that justify their use, arguing that compilers should not be expected to eliminate them without increasing instruction count, and that SCL provides a formal basis for preserving meaning during such transformations.

ABSTRACT

A steering fragment of an instruction sequence consists of a sequence of steering instructions. These are decision points involving the check of a propositional statement in sequential logic. The question is addressed why composed propositional statements occur in steering fragments given the fact that a straightforward transformation allows their elimination. A survey is provided of constraints that may be implicitly assumed when composed propositional statements occur in a meaningful instruction sequence.

Motivation & Objective

• To understand why composed (non-atomic) propositional statements appear in steering fragments despite the existence of a straightforward transformation to atomic form.
• To identify implicit constraints that justify the use of non-atomic steering points in meaningful instruction sequences.
• To examine whether such non-atomic statements are necessary in practice, particularly in real-time systems or due to side-effect behavior.
• To argue that compilers should not be expected to eliminate non-atomic steering points without increasing instruction count.
• To propose that SCL (Sequential Conditional Logic) provides a formal framework for meaning-preserving transformations of propositional statements within steering points.

Proposed method

• Uses a formal model of instruction sequences based on thread algebra and polarized process algebra, with atomic steering points represented as +a or -a instructions that evaluate a propositional atom and branch based on its boolean result.
• Applies propositional algebra with short-circuit evaluation, using the conditional composition operator x◃y▹z as a primitive, and derives binary connectives from it.
• Employs axioms from [13] (Table 1) to formalize the behavior of propositional statements, including associativity and distributivity under sequential evaluation.
• Analyzes side effects of atomic actions within steering points, distinguishing between marginal and detectible side effects.
• Considers real-time contexts where external measurements (e.g., sensor data) may cause non-reply-stable evaluations due to dynamic changes during evaluation.
• Proposes classification of instruction sequences into 'applications' and 'system utilities' to model detectibility of side effects across concurrent execution.

Experimental results

Research questions

• RQ1 Why do non-atomic propositional statements persist in instruction sequences despite the existence of a transformation to atomic form?
• RQ2 What implicit constraints justify the use of non-atomic steering points in practical programming?
• RQ3 Under what conditions are side effects of atomic actions in steering points detectible, and how does this affect program structure?
• RQ4 Can real-time dynamics lead to non-reply-stable evaluations of propositional statements in steering fragments?
• RQ5 Is it feasible or necessary to transform non-atomic steering points into atomic ones without increasing instruction count?

Key findings

• Non-atomic steering points are justifiable when side effects are marginal by default, especially in normal execution cases.
• Side effects of steering atoms are not detectible by other atoms within the same instruction sequence or propositional statement, but can be detected in concurrent multithreaded settings governed by strategic interleaving.
• In real-time systems, dynamic changes in external inputs (e.g., sensor data) can lead to non-reply-stable evaluations, justifying non-atomic propositional statements.
• The transformation of non-atomic steering points into atomic ones may lead to a significant increase in instruction count, making it impractical for compilers to enforce such transformations.
• SCL provides a formal specification for meaning-preserving transformations of propositional statements within steering points, supporting both atomic and non-atomic forms.
• Empirical validation of these constraints through survey research on practical programs is needed to confirm their relevance to current imperative programming practice.