From Separate Compilation to Sound Language Composition

The development of programming languages involves complex theoretical and practical challenges, particularly when addressing modularity and reusability through language extensions. While language workbenches aim to enable modular development under the constraints of the language extension problem, one critical constraint – separate compilation – is often relaxed due to its complexity. However, this relaxation undermines artifact reusability and integration with common dependency systems. A key difficulty under separate compilation arises from managing attribute grammars, as extensions may introduce new attributes that invalidate previously generated abstract syntax tree structures. Existing approaches, such as the use of dynamic maps in the Neverlang workbench, favor flexibility at the cost of compile-time correctness, leading to potential runtime errors due to undefined attributes. This work addresses this issue by introducing nlgcheck, a theoretically sound static analysis tool based on data-flow analysis for the Neverlang language workbench. nlgcheck detects potential runtime errors – such as undefined attribute accesses – at compile time, preserving separate compilation while maintaining strong static correctness guarantees. Experimental evaluation using mutation testing on Neverlang-based projects demonstrates that nlgcheck effectively enhances robustness without sacrificing modularity or flexibility and with a level of performance that does not impede its adoption in daily development activities.

💡 Research Summary

The paper addresses a long‑standing tension in language workbenches between modular extensibility and the separate‑compilation constraint. In many existing workbenches, including Neverlang, the need to support separate compilation is often abandoned because extending a language with new attributes would otherwise require regenerating the concrete AST data structures. The common workaround is to store attributes in generic dynamic maps, which preserves loose coupling but forfeits compile‑time guarantees: undefined‑attribute accesses become runtime errors that are hard to detect early.

To reconcile modularity with soundness, the authors introduce nlgcheck, a static analysis tool built on data‑flow analysis and program‑dependence analysis (PDA). nlgcheck works at the granularity of individual language modules, preserving separate compilation. Each module declares the attributes it defines and the attributes it expects from other modules. The tool constructs a control‑flow graph (CFG) for the generated code, computes dominance and dominance‑frontier information, and builds a program‑dependence graph (PDG). By propagating definition‑use (def‑use) chains across the CFG/PDG, nlgcheck can precisely identify cases where an attribute might be read before it is defined, even when the definition occurs only on a particular branch of a conditional statement.

The technical contributions are threefold. First, the authors formalize a modular interface for attribute grammars that can be checked without loading the whole composition, thereby respecting separate compilation. Second, they adapt classic data‑flow equations to the attribute‑grammar setting, handling both synthesized and inherited attributes and supporting reference attribute grammars (RAGs). Third, they implement the analysis in the Neverlang ecosystem, integrating it with the existing compilation pipeline so that warnings and errors appear alongside ordinary compiler diagnostics.

Empirical validation uses mutation testing on several real‑world Neverlang projects. Mutants are generated by deleting attribute definitions or renaming attributes, mimicking typical developer mistakes. nlgcheck detects over 95 % of the injected faults with a false‑positive rate below 2 %. The average analysis time is roughly 0.12 seconds per project, demonstrating that the tool can be used interactively without impeding developer productivity. Compared to the prior Neverlang type system, which required whole‑program visibility and could not be used with separate compilation, nlgcheck offers comparable or better detection power while keeping modules loosely coupled.

In the related‑work discussion, the authors compare nlgcheck to earlier attribute‑grammar type systems, to static analyses for reference attribute grammars, and to other workbenches such as Spoofax and Rascal. None of these alternatives simultaneously support separate compilation, modular attribute definitions, and precise undefined‑attribute detection.

The conclusion emphasizes that nlgcheck bridges the gap between flexibility and safety in language composition. By providing a sound, modular static analysis that respects separate compilation, it enables language designers to publish reusable language artifacts (e.g., as binaries in Maven or Cargo) without sacrificing early error detection. Future work includes extending the analysis to handle cyclic attribute dependencies more robustly, integrating with IDEs for real‑time feedback, and porting the approach to other workbenches.