Skip to main content

Rascal 0.40.x release notes

In this post we report on the Rascal release 0.40.x

Release 0.40.17 - November 15, 2024

The public release 0.40.x follows release 0.28.x; many improvements have been made in projects that depend on the rascal interpreter and the standard library (the type checker, the VScode extensions, clair, etc.) Some of these improvements depend directly on fixes in the interpreter and additions to the standard library.

The Rascal type-checker which is available in the VScode extension is reaching maturity. The .tpl file format has changed, which requires everybody to throw the old ones away. The new type checker checks .tpl file versions and reports possible conflicts. Use mvn clean or remove your bin or target folders for all your projects and library projects today.

The support for XML, JSON and HTML as exchange formats has been improved or completely rewritten. The main feature that was added was the optional loc src keyword field that provides the exact location of each node as it appears in the XML, HTML or JSON source text.

A new mvn:// scheme was added to be able to address with brief but unique notation the jar files in the local M2 repository. It supports two modes:

  • mvn://groupId!artifactId!optionAlComplexVersionString/pathInsideJar, shorthand for identifying artifacts inside the mvn repository on this machine.
  • mvn:///path/inside/m2/repository/jarFile.jar for exploring the mvn repository from its root. If maven.repo.local is set, then that is the root of the M2 repository. Otherwise if ~/m2/.repository exists (in the users ome directory) then that is used. Otherwise mvn is executed with -Dexpression=settings.localRepository to extract the setting from the current pom.xml file. This is not the official order as mvn resolves these configuration options; but it was chosen for the sake of efficiency.

A great deal of tests were fixed, enhanced or extended as a side-effect of the compiler project. Another visible aspect of the progress of the compiler is that now all Rascal runtime values (from vallang and beyond) now support getFingerprint() methods which help in optimizing pattern matching and dispatch in generated code by the compiler. These methods' return values have become a strict contract for future implementations of Rascal values, including parse trees and reified types and first-class functions.

The following issues were solved:

On the Rascal run-time engine (vallang) these issues were resolved:

Standard library maintenance:

  • Accurate and correct parsers of Windows and Unix file paths were added to the standard library. This includes a new unc:// resolver to accurately represent the semantics of UNC paths, and cwdrive:// which can represent the current working directory on a given drive letter on Windows.
  • HTMLElement(loc src = |unknown:///|) was added to position every tag from start to end via the src attribute.
  • The documentation strings were ported from @doc{ } notation to @synopsis{..}, @description{..}, @examples{..}, @benefits{..}, @pitfalls{..} separate tags.
  • The @deprecated{..} tag is now also used during API documentation generation. It is rendered between the synsopsis and the declaration signature.
  • The Box language for automatic string formatting was revived and its box2text algorithm was optimized. See lang::box
    • Box2text was re-implemented using list comprehensions and list splicing for efficiency and brevity.
    • Box arrays (tables) were fixed and finalized, and utility support for mapping lists to tables was added.
    • Box "groups" were added (ported from ASF+SDF) as a means to easily generate boxes from (separated) lists.
    • The concepts of fonts and highlighting were completely removed from Box, as this is an orthogonal feature implemented elsewhere by highlighter algoriths and mappings to HTML. The new trick is to use Box to format a file, then reparse that file and map it to ANSI or HTML or other markup formalisms.
    • the NULL box was added as a convenience for plugging holes and not loose parity or other counts.
    • Tests for Box2text were added.
    • Tree2Box is a new language-parametric formatter that maps any parse tree to Box using default heuristics. They trigger on the shape of production rules as they are typically found in programming languages. Tree2Box is a re-implementation of the pandora tool of the ASF+SDF Meta-Environment, but written in Rascal instead of C+ApiGen. You can override default behavior by adding rules for your exceptional language constructs.
  • The CSV model now has origin fields for Tables, Records and Fields: loc src=|unknown:///, such that CSV files can be parsed and treated as (DSL) source code.
  • lang::json::IO now has full origin tracking support.
  • util::Monitor progress monitoring is now also supported on textual interfaces using UTF8 and ANSI support for pretty bars. If UTF8 is not supported by the terminal, it used ASCII art. If ANSI is not available, it defaults to normal event logging prins on the console. The progress bar will always default to the latter if in a CI environment of if -Drascal.monitor.batch is set.
  • The progress monitors for module importation in the interpreter and parser generator were rationalized.
  • In lang::rascal::grammar::storage::ModuleParserStorage a new feature for saving generated parsers to disk and loading them again was added. It follows the interface design of ParseTree::parsers. You can load a saved parser and used it as if just generated with parsers.
  • lang::rascal::vis::ImportGraph was added as a port of the ASF+SDF Meta-Environment import graph visual.
  • lang::std::ANSI is an almost complete specification of the ANSI standard for character markup.
  • util::Clipboard was added to give programmatic access to the systems copy/paste feature for textual content. The feature starts up lazily so the first call to copy or paste is slower than subsequent calls. In a headless environment copy always returns the empty string and paste simply has no effect. Paste and copy are "thread friendly" in that they will produce or reproduce a string that at one time was in the system's copy/paste buffer. However, it is not necessarily the last entry and other processes may overwrite the buffer while the Rascal programming is running that has just written to it.
  • util::Eval was refactored and reimplemented for efficiency and robustness, following the TutorCommandExecutor design from the rascal-tutor project. Eval instances are now fully configured via PathConfig.
  • The concept of CodeActions was added to util::IDEServices, they also extend error Message constructors. CodeActions can be used to register quickfixes and other code related actions to editor positions. The positions are either directed by the error message they are attached to, or via syntax-directed pattern matching. This feature plays well with the LanguageServer features in rascal-lsp.
  • added util::Reflective::newRascalProject to set up an empty project template with the right RASCAL.MF file and pom.xml file.
  • ManifestRunner.java was removed because it was unused.
  • The benchmark:/// resolver was removed, so was test-temp:///, test-data:/// and test-modules:///.
  • The memory:// resolver now supports independently garbage collectible filesystems per authority name.

The Java model has received big maintenance love and attention, including improvements to the generic M3 model:

  • Bumped and upgraded to the JDT version from Eclipse 2020-03, then upgraded mapping support up to and including JLS-14.
  • Up to JLS14 all new Java constructs were added to the AST and the M3 Core model. The previous standard we supported was JLS8.
  • The AST nodes in lang::java::m3::AST now all satisfy the AST contract in analysis::m3::AST. This means that all source code elements are represented in the tree, annotated with src origins and ordered from left-to-right as they were in the original source file.
  • The constructors: enum, enumConstant, compilationUnit, class, interface, method, field, etc., all received extra positional parameters for the concept of modifiers. Before these were modelled as keyword parameters, but that invalidated the earlier mentioned AST contract.
  • An AST and M3 model of the Java 9 module system was added.
  • The isSuper boolean was removed from the AST definition of methodInvocationn and new calls, also to satisfy the AST contract.
  • Java Annotations AST constructors were moved from Expression to Modifier.
  • String based unary and binary operator constructors for Expressions were unfolded to a constructor for each operator, i.e. plus(Expression, Expression) instead of binop(Expression, "+", Expression)
  • lang::java::m3::AST was documented and so was analysis::m3::AST
  • lang::java::m3::Core was documented and so was analysis::m3::Core
  • composeM3 was generalized for any number of keyword fields which are either sets or lists. This makes it language-independent from now on.
  • m3SpecificationTest checks for the internal sanity and completeness of an M3 model. Can be used to test language front-ends.
  • Java2ObjectFlow was upgraded for all the changes in the AST constructors. However it may still need extension for new constrtructors like lambda expressions.
  • Lambda's were added to Expression
  • Method references were added to Expression
  • Intersection types were added to the Type AST class
  • The bounds constructors of Type ASTs were renamed from upperbound and lowerbound to super and extends.
  • Java versions are now specified as constructors of the Language data type, for accurate description of the JLS language level the user needs to reflect.
  • M3 extraction from JVM binary class files was maintained and now also supports language features up to JLS14. In particular the Java 9 module system was added to the mapping.
  • Tracebility with origin tracking was improved for both source code and binary classfile analysis, so if NPE's happen a clear cause can be printed.