Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog,
and this project adheres to Semantic Versioning,
and yes, platform and engine support are part of the public API.
Please file a bug if you notice a violation of semantic versioning.

Unreleased

Added

Changed

Deprecated

Removed

Fixed

Security

2.0.1 - 2026-02-19

• TAG: v2.0.1
• COVERAGE: 89.42% – 575/643 lines in 11 files
• BRANCH COVERAGE: 65.23% – 182/279 branches in 11 files
• 97.03% documented

Added

• AGENTS.md

Changed

• appraisal2 v3.0.6
• kettle-test v1.0.10
• stone_checksums v1.0.3
• ast-merge v4.0.6
• tree_haver v5.0.5
  • FFI Backend improvements
  • Error handling improvements
  • Many new features, and more bug fixes
• tree_stump v0.2.0
  • fork no longer required, updates all applied upstream
• Simplified dependency_tags.rb: Removed redundant debug code
  • Removed TOML_MERGE_DEBUG env var handling (use TREE_HAVER_DEBUG instead)
  • tree_haver’s debug output now respects blocked backends via compute_blocked_backends
  • Avoids accidentally loading MRI backend during FFI-only test runs
• Updated documentation on hostile takeover of RubyGems
  • https://dev.to/galtzo/hostile-takeover-of-rubygems-my-thoughts-5hlo

2.0.0 - 2026-01-09

• TAG: v2.0.0
• COVERAGE: 88.38% – 563/637 lines in 11 files
• BRANCH COVERAGE: 64.13% – 177/276 branches in 11 files
• 97.03% documented

Added

• FFI backend isolation for test suite
  • Added bin/rspec-ffi script to run FFI specs in isolation (before MRI backend loads)
  • Added spec/spec_ffi_helper.rb for FFI-specific test configuration
  • Updated Rakefile with ffi_specs and remaining_specs tasks
  • The :test task now runs FFI specs first, then remaining specs
• Emitter autoload - Added Emitter to module autoload list
  • Previously missing, causing NameError: uninitialized constant Emitter in ConflictResolver
  • Now properly autoloaded via autoload :Emitter, "toml/merge/emitter"
• Backends module with constants for backend selection
  • Backends::TREE_SITTER (:tree_sitter_toml) - Native tree-sitter parser
  • Backends::CITRUS (:citrus_toml) - Pure Ruby toml-rb parser
  • Backends::AUTO (:auto) - Auto-detect available backend
  • Backends.validate! and Backends.valid? for validation
• SmartMerger now accepts backend: parameter for explicit backend selection
  • Follows same pattern as markdown-merge
  • Auto-detects backend by default, or use backend: Backends::CITRUS to force pure Ruby
• FileAnalysis now accepts backend: parameter and exposes resolved backend via #backend attr
• NodeWrapper now accepts backend: and document_root: parameters for correct normalization
• Structural normalization for Citrus backend: Tree-sitter and Citrus backends produce different AST structures for tables:
  • Tree-sitter: Table nodes contain pairs as children
  • Citrus: Table nodes only contain header; pairs are siblings at document level
  • NodeWrapper#pairs now finds associated pairs regardless of AST structure
  • NodeWrapper#content now returns full table content on both backends
  • NodeWrapper#effective_end_line calculates correct end line including pairs
  • FileAnalysis passes document_root to all NodeWrappers for sibling lookups
  • This enables the merge logic to work identically across backends
• NodeTypeNormalizer module for backend-agnostic node type handling
  • Maps backend-specific types (e.g., table_array_element) to canonical types (e.g., array_of_tables)
  • Supports both tree_sitter_toml and citrus_toml backends with comprehensive type mappings
  • Provides helper methods: table_type?, value_type?, key_type?, container_type?
  • Extensible via register_backend for custom TOML parsers
  • Follows the same pattern as markdown-merge’s NodeTypeNormalizer
• NodeWrapper#canonical_type method returns the normalized type for a node
• Comprehensive test suite for NodeTypeNormalizer with 26 new specs
• spec/support/dependency_tags.rb for conditional test execution based on backend availability

Changed

• ast-merge v3.1.0
• tree_haver v4.0.3, adds error handling for FFI backend
• Test suite now explicitly tests all available backend modes - Tests previously ran with
  whatever backend was auto-selected. Now specs explicitly test up to five backend configurations:
  • :auto backend - Tests default user experience (backend-agnostic)
  • :mri backend via TreeHaver.with_backend(:mri) - Tests explicit tree-sitter MRI behavior
  • :citrus backend via TreeHaver.with_backend(:citrus) - Tests explicit toml-rb behavior
  • :rust backend via TreeHaver.with_backend(:rust) - Tests explicit tree-sitter Rust behavior
  • :java backend via TreeHaver.with_backend(:java) - Tests explicit tree-sitter Java behavior

  This ensures consistent behavior is verified across all backends, rather than relying
  on auto-selection which may vary by platform. Each shared example group is included
  in all contexts with appropriate dependency tags (:toml_grammar, :toml_rb,
  :toml_parsing, :rust_backend, :java_backend). Tests for unavailable backends
  are automatically skipped.

  Note: The :java_backend tag now correctly detects whether the Java backend can
  actually load grammars. Standard .so files built for MRI’s tree-sitter C bindings
  are NOT compatible with java-tree-sitter. Tests will be skipped on JRuby unless
  grammar JARs from Maven Central (built for java-tree-sitter’s Foreign Function Memory
  API) are available.

• Backend handling simplified - Let TreeHaver handle all backend selection:
  • Removed backend: parameter from SmartMerger and FileAnalysis
  • Removed Backends module entirely (was unused after removing backend: parameter)
  • Users control backend via TreeHaver directly (TREE_HAVER_BACKEND env var, TreeHaver.backend=, or TreeHaver.with_backend)
  • This ensures compatibility with all TreeHaver backends (mri, rust, ffi, java, citrus)
• Backend naming simplified to align with TreeHaver:
  • NodeTypeNormalizer mappings now keyed by :tree_sitter and :citrus
  • All native TreeHaver backends (mri, rust, ffi, java) produce tree-sitter AST format
• See .github/COPILOT_INSTRUCTIONS.md for comprehensive TreeHaver backend documentation
• NodeWrapper: Now inherits from Ast::Merge::NodeWrapperBase
  • Removes ~80 lines of duplicated code (initialization, line extraction, basic methods)
  • Uses process_additional_options hook for TOML-specific options (backend, document_root)
  • Keeps TOML-specific type predicates using NodeTypeNormalizer
  • Keeps Citrus structural normalization logic for #pairs, #content, #effective_end_line
  • Adds #node_wrapper? method for distinguishing from NodeTyping::Wrapper
• citrus_toml mappings: Updated to match actual Citrus/toml-rb node types
  • table_array → :array_of_tables (Citrus produces :table_array, not :table_array_element)
  • keyvalue → :pair (Citrus produces :keyvalue, not :pair)
  • Added all Citrus-specific integer types: decimal_integer, hexadecimal_integer, octal_integer, binary_integer
  • Added all Citrus-specific string types: basic_string, literal_string, multiline_string, multiline_literal
  • Added all Citrus-specific datetime types: local_date, local_time, local_datetime, offset_datetime
  • Added Citrus-specific boolean types: true, false
  • Added whitespace types: space, line_break, indent, repeat
• FileAnalysis error handling: Now rescues TreeHaver::Error instead of TreeHaver::NotAvailable
  • TreeHaver::Error inherits from Exception, not StandardError
  • TreeHaver::NotAvailable is a subclass of TreeHaver::Error, so it’s also caught
  • Fixes parse error handling on TruffleRuby where Citrus backend raises TreeHaver::Error
• Dependency tags: Refactored to use shared TreeHaver::RSpec::DependencyTags from tree_haver gem
  • All dependency detection is now centralized in tree_haver
  • Use require "tree_haver/rspec" for shared RSpec configuration
  • TomlMergeDependencies is now an alias to TreeHaver::RSpec::DependencyTags
  • Enables TOML_MERGE_DEBUG=1 for dependency summary output
• FileAnalysis: Error handling now follows the standard pattern
  • Parse errors are collected but not re-raised from FileAnalysis
  • valid? returns false when there are errors or no AST
  • SmartMergerBase handles raising the appropriate parse error
  • Consistent with json-merge, jsonc-merge, and bash-merge implementations
• SmartMerger: Added **options for forward compatibility
  • Accepts additional options that may be added to base class in future
  • Passes all options through to SmartMergerBase
  • node_typing parameter for per-node-type merge preferences
    • Enables preference: { default: :destination, special_type: :template } pattern
    • Works with custom merge_types assigned via node_typing lambdas
  • regions and region_placeholder parameters for nested content merging
• ConflictResolver: Added **options for forward compatibility
  • Now passes match_refiner to base class instead of storing locally
• MergeResult: Added **options for forward compatibility
• FileAnalysis: Simplified to use TreeHaver.parser_for API
  • Removed 40+ lines of grammar loading boilerplate
  • Now relies on tree_haver for auto-discovery and Citrus fallback
  • :tree_sitter_toml RSpec tag for tree-sitter-toml grammar tests
  • :toml_rb RSpec tag for toml-rb/Citrus backend tests
  • :toml_backend RSpec tag for tests requiring any TOML backend
• BREAKING: NodeWrapper type predicates now use NodeTypeNormalizer for backend-agnostic type checking
  • array_of_tables? now correctly identifies both table_array_element (tree-sitter) and array_of_tables nodes
  • All predicates (table?, pair?, string?, etc.) use canonical types
  • type? method checks both raw and canonical types
• FileAnalysis#tables now uses NodeTypeNormalizer.table_type? for type detection
• FileAnalysis#root_pairs and #integrate_nodes use canonical type checks
• TableMatchRefiner#table_node? uses NodeTypeNormalizer for backend-agnostic table detection
• compute_signature method uses canonical types for consistent signatures across backends
• Rewrote node_wrapper_spec.rb with proper tests (removed placeholder/pending tests)
• Rewrote table_match_refiner_spec.rb with working tests using :toml_backend tag
• Updated spec_helper.rb load order to ensure TreeHaver is available for dependency detection
• BREAKING: Migrate from direct TreeSitter::Language.load to TreeHaver API
  • Changed require "tree_sitter" to require "tree_haver" in main module file
  • Added automatic grammar registration via TreeHaver::GrammarFinder#register!
  • FileAnalysis#find_parser_path now exclusively uses TreeHaver::GrammarFinder
  • FileAnalysis#parse_toml now uses TreeHaver::Parser and TreeHaver::Language
  • Removed legacy fallback path search (TreeHaver is now a hard requirement)
  • Updated documentation to reference TreeHaver::Node instead of TreeSitter::Node
  • Environment variable TREE_SITTER_TOML_PATH is still supported via TreeHaver
  • This enables support for multiple tree-sitter backends (MRI, Rust, FFI, Java) and Citrus fallback

Removed

• Load-time grammar registration - TreeHaver’s parser_for now handles grammar discovery
  and registration automatically. Removed manual GrammarFinder calls and warnings from
  lib/toml/merge.rb.

Fixed

• Citrus backend normalization improvements for TruffleRuby compatibility:
  • NodeWrapper#key_name now strips whitespace from key text (Citrus includes trailing spaces)
  • NodeWrapper#table_name now strips whitespace from table header text
  • NodeWrapper#extract_inline_table_keys now recursively handles Citrus’s deeply nested
    structure (inline_table -> optional -> keyvalue -> keyvalue -> stripped_key -> key -> bare_key)
  • NodeWrapper#elements now recursively handles Citrus’s array structure where elements
    are nested in array_elements -> repeat -> indent -> decimal_integer chains
  • Both methods now correctly extract all values instead of just the first one
  • NodeWrapper#value_node now skips Citrus internal nodes (whitespace, unknown, space)
  • All NodeTypeNormalizer.canonical_type() calls now pass @backend parameter for correct type mapping
  • FileAnalysis#root_pairs now correctly filters pairs to only include those BEFORE the first table
    (Citrus has flat AST structure where all pairs are document siblings)
  • MergeResult#add_node now uses effective_end_line to include table pairs on Citrus backend
  • TableMatchRefiner#table_node? now uses node’s backend for correct type checking
  • Test helper parse_toml now uses FileAnalysis for proper backend detection
• NodeTypeNormalizer.canonical_type now defaults to :tree_sitter_toml backend when no backend is specified
  • Added DEFAULT_BACKEND constant and overrode canonical_type and wrap methods
  • Fixes issue where calling canonical_type(:table_array_element) without a backend argument would passthrough instead of mapping to :array_of_tables
  • Value type predicates (string?, integer?, float?, boolean?, array?, inline_table?, datetime?) now work correctly
• Consolidated duplicate describe blocks in spec files (file_analysis_spec.rb, merge_result_spec.rb, node_wrapper_spec.rb)
• Fixed lint violations: added missing expectations to tests, used safe navigation where appropriate
• No longer warns about missing TOML grammar when the grammar file exists but tree-sitter runtime is unavailable
  • This is expected behavior when using non-tree-sitter backends (Citrus, Prism, etc.)
  • Warning now only appears when the grammar file is actually missing

1.0.0 - 2025-12-19

• TAG: v1.0.0
• COVERAGE: 94.05% – 506/538 lines in 9 files
• BRANCH COVERAGE: 76.64% – 164/214 branches in 9 files
• 96.55% documented

Added

• Initial release
• Added support for pure Ruby TOML parsing via tree_haver v3 Citrus backend with toml-rb
  • Automatically registers both tree-sitter-toml (native, fast) and toml-rb (pure Ruby) grammars
  • TreeHaver auto-selects best available backend (tree-sitter preferred, Citrus fallback)
  • Enables TOML parsing on platforms without native library support
  • Can force Citrus backend via TREE_HAVER_BACKEND=citrus environment variable
• Added graceful error handling when neither tree-sitter-toml nor toml-rb are available