What's Changed

• Implemented collection and object attribute directives. The new collect<C>[e] directive synthesizes a sequence or associative container type C consisting of elements gathered from the inherited or synthesized attributes in expression e. Likewise, there also new synthesize<C,A...>[e], synthesize_shared<C,A...>[e] and synthesize_unique<C,A...>[e] directives for synthesizing objects, shared pointers and unique pointers respectively, constructed from the component attributes in expression e.
• Implemented synthesize_collect directive that combines collect and synthesize directives together for improved code readability and reduced boilerplate when constructing complex data structures from parsed elements. This is particularly useful for building nested collections like arrays of objects or maps with complex value types.
• Added lug::recursive_wrapper template class for handling circular dependencies in abstract syntax trees, particularily those making use of std::variant.
• Updated to Unicode Standard Version 16.0.0 and added support for building tools to CMakeLists.txt.
• Optimized character range and rune set matching for ASCII characters, resulting in significant performance improvements for common text processing operations. Added specialized fast paths for ASCII-only inputs that bypass the more expensive Unicode handling code.
• Implemented test opcodes for head-fail optimization and repeat opcodes for optimized whitespace skipping. These optimizations will be fully enabled in the next release after planned expression tree transformations.
• Enhanced input source handling with improved buffering and error reporting for std::istream sources, and improved the interactive mode support that properly handles line-by-line input for terminal sessions or for line-oriented grammars.
• Moved std::istream support into its own <lug/iostream.hpp> header file, removing the dependency on <iostream> from the main <lug/lug.hpp> header. This reduces compilation times and minimizes header dependencies for projects that don't require stream I/O functionality.
• Refactored parser commit logic by inlining instructions in lug::basic_parser to better align with the stack frame architecture changes introduced in v0.4.0, improving code organization and performance.
• Fixed issue in the BASIC sample where user-defined functions (e.g. FNA(X)) were crashing during evaluation. This was caused by v0.4.0 changes that reset lug::environment during nested parsing operations. Added new lug::environment::should_reset_on_parse member function to provide fine-grained control over this behavior, allowing environments to persist across nested parse operations when needed.
• Added comprehensive test infrastructure for the sample programs, including automated test plans and verification scripts to ensure sample functionality remains correct across changes.
• Restructured include directory hierarchy by moving lug/ to include/lug/, ensuring cleaner include paths and better organization of header files.
• Added support for additional compilers (GCC 9/10/11/12, Clang 14/15/16/17) in the C/C++ CI workflow to ensure broader compatibility.
• Added static analyzer CI workflow for Clang and MSVC to detect potential bugs, memory leaks, and undefined behavior during the build process.
• Added compiler sanitizer CI workflow with Address Sanitizer (ASan), Undefined Behavior Sanitizer (UBSan), and Memory Sanitizer (MSan) to detect memory errors, undefined behavior, and other runtime issues early in the development process.
• Added clang-tidy integration to CI workflow to enforce code quality standards and catch potential issues early in the development process.
• Removed Ubuntu 20.04 from the GitHub workflows as it has reached end-of-life and is no longer supported.

Full Changelog: v0.4.0...v0.5.0

What's Changed

• Added new error handling and recovery with labeled failures, recovery rules and error handlers.
• Added new accept operator that works in conjuction with cut operator. Tested and fixed issues with accept and cut, allowing them to be used for debugging error handling.
• Added repeat(N)[e] and repeat(N,M)[e] control operators that repeat matching of expression e at least N and at most M times.
• Added on_reset and on_drain hooks to the lug::environment class.
• Refactored the encoder expression interfaces into CRTP base classes to allow for common functionality to be shared by all concrete encoder expressions, such as support for labeled failures and recovery rules.
• Implemented new 64-bit instruction encoding scheme that is simpler and more efficient. String data is no longer embedded in the instruction stream, ensuring constant instruction length of 8 bytes.
• Merged lug::program_encoder and lug::rule_encoder into the lug::encoder base class after overhauling how choice/jump offsets are calculated, significantly reducing binary size bloat.
• Extracted common base class of lug::basic_parser into lug::parser_base to reduce template bloat.
• Reduced template bloat for parser directives (improving compiler error messages) and optimized nested directives.
• Merged parser stack frames together into a single stack using std::variant.
• Fixed issues with lug::basic_parser and lug::environment not being fully reset on each invocation of parse().
• Allow for move-only types to be used in the lug::environment attribute stack.
• Removed use of goto in parsing machine main loop.
• Changed grammar program generation to force inline the start rule.
• Implemented tail call optimization during grammar code generation.

Full Changelog: v0.3.0...v0.4.0

What's Changed

• Added list repetition operator e1 >> e2 to the DSL that is shorthand for e1 > *(e2 > e1).
• Added support for parsing characters and character literals where applicable without explicitly needing to wrap them with chr() or _cx.
• Symbols now respects caseless mode, allowing for case-insensitive matching against symbol definitions.
• Allow for use of variables of all types in attribute bindings and removed the lug::variable template class that was used previously. Variable state is automatically saved and restored across rule boundaries.
• Allow for capturing text to a lug::syntax object or any string-like object that is convertible from std::string_view, and renamed syntax::capture to syntax::str in order to match std::sub_match::str.
• Added lug::source_options::interactive flag that ignores eoi tokens for TTY input sources.
• Rewrote the expression function objects/lambdas as expression template classes. Allows for multiple passes over the expression tree as well as top-down and bottom-up traversal, which was needed when implementing attribute state tracking. This will also allow for additional optimizations to be implemented in the future.
• Renamed syntactic_capture to semantic_capture_action to reflect that it is executed during the semantic action evaluation phase.
• Make all variations of callables that return a non-void value that can be type-erased by semantic_action and semantic_capture_action push their result onto the attribute result stack.
• Removed semantic_response from the public API as it was only used internally inside of the parser.
• Attempting to bind a variable to a nonexistent value from the attribute result stack now throws an attribute_stack_error.
• implicit_space_rule no longer causes a compiler warning with Clang, uses RAII to push/pop the thread-local white space rule for grammars.
• Moved call_depth(), prune_depth() and escape() functions into the lug::environment class since they are used exclusively during semantic action phase.
• Moved line/column tracking and current match/subject string views to lug::environment class, fully removing the environment's dependency on lug::parser.
• Turned lug::parser into an alias of a new lug::basic_parser template class parameterized with an input source strategy. This allows for parsing and capturing of text without making a copy of the input.
• Placed all DSL operator overloads inside of an inline namespace operators within lug::language. This allows only the operators to be imported into the current scope if desired.
• Enabled -Wconversion and -Wshadow warnings for Clang and GCC and fixed warnings.
• Full clang-tidy pass on all of the library headers and fixed all warnings.
• Added CMake build support and removed old MSVS solution and vcxproj files.
• Handle situation where compilation with RTTI is disabled.

Full Changelog: v0.2.0...v0.3.0

What's Changed

• Implemented new support for context-sensitive grammars with symbol tables and parsing conditions, based on the PEG extensions described in the paper "A Declarative Extension of Parsing Expression Grammars for Recognizing Most Programming Languages" by Tetsuro Matsumura and Kimio Kuramitsu (2015).
• Added an XML Standard 1.0 matcher sample program demonstrating use of symbol tables.
• Finished the BASIC language interpreter sample program, which is now feature complete, using parsing conditions.
• Updated Unicode support to version 15.1.0 and automated Unicode table generation via Makefile build.
• Various compilation error fixes for modern C++17 compilers (Clang, GCC, MSVC++), as well as fixing compilation errors when compiling in C++20 and C++23 modes. Includes fixes merged from @trombonehero in #1.
• Code cleanup pass, with numerous improvements applying const, constexpr, [[nodiscard]], noexcept and fixing new compilation warnings enabled with -Wextra, -Wextra-semi, -Wsign-conversion and -Wsuggest-override on Clang and GCC.
• Initial clang-tidy support with .clang-tidy file and lint Makefile target.
• Migrated from Travis CI to Github Actions CI.

New Contributors

• @trombonehero made their first contribution in #1

Full Changelog: v0.1.0...v0.2.0