Incremental type-checking for free: using scope graphs to derive incremental type-checkers

Abstract

Fast analysis response times in IDEs are essential for a good editor experience. Incremental type-checking can provide that in a scalable fashion. However, existing techniques are not reusable between languages. Moreover, mutual and dynamic dependencies preclude traditional approaches to incrementality. This makes finding automatic approaches to incremental type-checking a challenging but important open question. In this paper, we present a technique that automatically derives incremental type-checkers from type system specifications written in the Statix meta-DSL. We use name resolution queries in scope graphs (a generic model of name binding embedded in Statix) to derive dependencies between compilation units. A novel query confirmation algorithm finds queries for which the answer changed due to an edit in the program. Only units with such queries require reanalysis. The effectiveness of this algorithm is improved by (1) splitting the type-checking task into a context-free and a context-sensitive part, and (2) reusing a generic mechanism to resolve mutual dependencies. This automatically yields incremental type-checkers for any Statix specification. Compared to non-incremental parallel execution, we achieve speedups up to 147x on synthetic benchmarks, and up to 21x on real-world projects, with initial overheads below 10%. This suggests that our framework can provide efficient incremental type-checking to the wide range of languages supported by Statix.

OIL: an industrial case study in language engineering with Spoofax

“…• Static Semantics. Static semantics specification (NaBL2 and its successor Statix [25], both based on a scope graph model [26]) and support for incremental type checking [27].…”

“…There are three types of areas: regions, states and zones (lines [14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33]. A region always refers to an enum variable and contains a number of states.…”

“…Spoofax could be improved by supporting cross-language static analysis by making it possible to merge the scope graphs of two separately defined languages, as also described by Zwaan [61]. In Spoofax 2 this was deemed virtually impossible to implement.…”

“…Thanks to the new architecture of Spoofax 3, it supports the implementation of cross-language static analysis, which should support the OIL and IDL case. Cross-meta-language static analysis was one of the goals of Zwaan [61], but have not yet been materialized.…”