Domain-Level Observation and Control for Compiled Executable DSLs

Executable Domain-Specific Languages (DSLs) are commonly defined with either operational semantics (i.e., interpretation) or translational semantics (i.e., compilation). An interpreted DSL relies on domain concepts to specify the possible execution states and steps, which enables the observation and control of executions using the very same domain concepts. In contrast, a compiled DSL relies on a transformation to an arbitrarily different target language. This creates a conceptual gap, where the execution can only be observed and controlled through target domain concepts, to the detriment of experts or tools that only understand the source domain. To address this problem, we propose a language engineering architecture for compiled DSLs that enables the observation and control of executions using source domain concepts. The architecture requires the definition of the source domain execution steps and states, along with a feedback manager that translates steps and states of the target domain back to the source domain. We evaluate the architecture with two different compiled DSLs, and show that it does enable domain-level observation and control while increasing execution time by 2× in the worst observed case.