Changelog

All notable changes to markymark are documented here. Each release links to the full diff on GitHub.

v0.6.0 (2026-02-25)

Completes the extraction pipeline migration started in v0.5.0. All remaining symbol types — code spans, XML tags, tasks, embeds, callouts, block refs, link definitions, query blocks, and properties — are now extracted through the full Zig→FFI→Rust→LSP/MCP stack. The legacy regex-based Rust extractors have been removed; zig-kernels is now mandatory. The RealmIndex has been rebuilt as v2 with string interning and incremental document updates.

Note: Agents like Claude Code use both LSP and MCP simultaneously — they are complementary protocols, not alternative paths. This release completes MCP parity with LSP for all extracted element types.

Features

Complete extraction pipeline (ix3) — 9 new element types extracted by the Zig md4c ExtractionRenderer, serialized in blob v2 format (128-byte header with backward compatibility), and wired through from_blob, from_scan, ScanBackend, and LSP/MCP
RealmIndex v2 (n7wx) — String interning via lasso deduplicates URI and heading allocations; new update_document() computes contribution diffs incrementally; O(1) stem_to_uris index for wiki link resolution; lazy tag_to_docs built on first access
Blob v2 header — 128-byte header with v1/v2 backward compatibility
MCP batch indexing migrated from from_ast to from_scan

Bug Fixes

Preserve frontmatter and mask it before engine/scan parsing
Pick earliest frontmatter close delimiter in mixed-ending files
Property scan past non-property lines and CRLF frontmatter handling
Correct offset recovery for code spans and tasks
Windows: append .exe to binary path and download URL
VSCode: return bare name for PATH fallback on unsupported platforms

Breaking Changes

zig-kernels is now mandatory — regex extractors have been removed
Zig 0.15.2+ required for all builds (enforced by build.rs)

Refactoring

scanner.rs split into 4 submodules
from_blob.rs converted to module directory with header.rs, owned.rs, and 4 test submodules
document.zig split into helpers, free functions, stored types, and FFI types
extraction_renderer_tests.zig split into 4 thematic test files
extract.rs converted to submodule directory with frontmatter, tasks, blocks, links, and tags

Infrastructure

Zig 0.15.2 archive corruption fix — archives repacked with system ar on Linux targets

Full diff: v0.5.1…v0.6.0

v0.5.1 (2026-02-21)

Focused improvements to the Zig md4c parsing pipeline and Rust error handling.

Bug Fixes

O(n) autolink paren trimming — the GFM autolink parenthesis-balancing loop in the Zig md4c port was quadratic; rewritten as a single-pass O(n) scan

Improvements

BlobError now implements Display and Error traits, enabling Box<dyn Error>, ? operator, and anyhow integration with per-variant diagnostic messages
Extract map_md4c_heading / map_md4c_link helpers in Md4cScanBackend

Full diff: v0.5.0…v0.5.1

v0.5.0 (2026-02-20)

Replaces the tree-sitter incremental indexing pipeline with a new md4c-based DocumentEngine. The new pipeline vendors Bun’s md4c Zig parser for single-pass markdown extraction, serializes results to a compact binary blob format, and crosses the FFI boundary into Rust — eliminating double-parse overhead.

Features

New md4c parsing pipeline — Vendored Bun’s md4c CommonMark parser (Zig), streaming ExtractionRenderer extracts headings, links, tags, and block IDs in a single pass, exposed through C ABI with Rust FFI bindings
DocumentEngine with blob serialization — Zig DocumentEngine produces a compact binary blob; Rust-side DocumentIndex::from_blob() deserializes without re-parsing
LSP pipeline overhaul — scan_all replaces tree-sitter incremental indexing, eliminating the previous double-parse (tree-sitter parse + separate extraction pass)
Async debounce — did_change notifications debounced at 75ms with generation counters preventing stale batches after close/reopen cycles

Bug Fixes

Removed undefined behavior in DocumentIndex arena_ref mutex escape
Removed unsound Sync impl on DocumentIndex
Fixed extractFromMarkdown double-free and heading text leak on OOM
Fixed toOwnedSlice cascade leak and append double-frees in parseAll
Fixed normalizeLabel memory leak in vendored md4c parser
Wiki link alias detection: compare text != target
Slug truncation no longer returns empty string

Breaking Changes

Tree-sitter incremental indexing replaced by md4c-based DocumentEngine
New document processing pipeline — plugins relying on tree-sitter internals will need updates

Infrastructure

Release process formalized in RELEASING.md with 7-crate publish order
prepare-release skill added for guided 4-phase release workflow
Golden blob roundtrip test catches unilateral Zig/Rust format drift

Full diff: v0.4.2…v0.5.0

v0.4.2 (2026-02-20)

Features

get-diagnostics MCP tool — new MCP tool with file:// URI validation and structured doc support

Bug Fixes

Include new entries from large insertions in incremental merge
Correct assertion message in realm_stats preview check
Create parent dirs in TempWorkspace::write for nested paths

Performance

Eliminate eager allocation of heading names in SearchSymbols

Refactoring

Split incremental/tests.rs into submodules
Split runtime_engine_tests.rs into 9 submodules
Move realm.rs to realm/ module directory with types.rs, helpers.rs, and tests.rs
Extract shared TempWorkspace test helper and migrate across test suites

Full diff: v0.4.1…v0.4.2

v0.4.1 (2026-02-19)

Features

LSP debounce — did_change notifications debounced with 75ms async cancellation

Bug Fixes

Detect edits in large gaps between extractor entries during incremental indexing
Deduplicate link edges per document in graph analysis
Improve markdown link resolution with path-relative lookup

Full diff: v0.4.0…v0.4.1

v0.4.0 (2026-02-19)

Major release introducing Zig SIMD acceleration kernels, MCP intelligence tools, incremental indexing, and a VSCode extension.

Note: Agents like Claude Code use both LSP and MCP simultaneously — they are complementary protocols, not alternative paths. This release adds MCP tools that pair with the existing LSP capabilities.

Features

Zig SIMD kernels — Format scanners (env_scan, ini_scan, toml_scan, yaml_scan, json_keys) with SIMD-accelerated key extraction; link graph engine; batched fuzzy match; Aho-Corasick multi-pattern scanner; slug generation via C ABI
MCP tools — search-workspace (full-text search with frontmatter/property/tag filtering), search-for-pattern (regex search with glob filtering and context lines), graph-analysis (link graph intelligence with orphans, hubs, broken links, clusters)
Incremental indexing (Phase 3) — All 5 extractors carry byte offsets for selective merge; range intersection with neighbor window and tail-boundary guard; benchmarked 1.23x speedup
VSCode extension — Marketplace-ready extension with binary discovery and LSP client

Bug Fixes

Implement ___chkstk_ms for Windows x86_64 to resolve stack frame issues
CRLF offset drift fix in incremental indexing
Insertion-point boundary correction

Breaking Changes

New feature-gated Zig dependency (zig-kernels feature flag)

Refactoring

Split state.rs (1170 lines) into state/{mod,completion,navigation,rename}.rs
Renamed runtime_engine to engine with submodules
Extracted tool handlers into tools/ submodule

Full diff: v0.3.0…v0.4.0

v0.3.0 (2026-02-16)

Introduces the Zig SIMD kernel foundation and incremental tree-sitter parsing.

Features

Zig kernel scaffold — zig/ directory with build.zig, source structure, and markymark-kernels Rust crate with FFI bridge
SIMD kernels — heading_scan, link_scan, tag_scan, block_scan, token_estimate, content_hash implemented in Zig with Rust FFI wrappers
Shared BRZA kernels — Similarity, normalize, entities, quantize, and embedding index kernels ported with Rust FFI wrappers
Incremental tree-sitter parsing — MarkdownTree stored per document, TextDocumentSyncKind::INCREMENTAL enabled, incremental parsing wired end-to-end
Core traits — ScanBackend and EmbeddingProvider traits added to markymark-core

Bug Fixes

Enable PIC in Zig static library for Linux x86_64 linking
Defensive FFI hardening: replace as u32 casts with try_from at FFI boundary
Initialize written parameter in scan FFI functions
Skip invalid incremental edits when old_end < start

Breaking Changes

First Zig dependency introduced (optional via zig-kernels feature flag)

Full diff: v0.2.0…v0.3.0

v0.2.0 (2026-02-16)

Introduces arena allocation, multi-format structured document support, and security hardening.

Features

Arena allocation — bumpalo-based arena infrastructure in markymark-core, migrated through parser and index layers with DocumentArena and ArenaHashMap
Multi-format support — JSON, JSONC, JSON5, JSONL, YAML, TOML, .env, and INI parsers with StructuredDocumentIndex, AnyDocumentIndex, and RealmIndex integration
LSP multi-format — DocumentSymbols, hover, and find-references for structured documents
Tree-sitter migration — Upgraded from tree-sitter 0.19 to 0.26 with tree-sitter-md wrapper
Security — Advisory security workflow with SARIF uploads, lefthook pre-commit hooks, custom semgrep rules with fixture validation
Plugin distribution — marketplace.json for self-hosted plugin distribution

Bug Fixes

Resolve frontmatter_and_properties SIGSEGV
Preserve duplicate block IDs across documents
Ignore XML-like syntax inside fenced code blocks
BlockEntry range propagation for go-to-definition

Performance

Optimize remove_from_cross_doc_indexes to O(doc size)
Real corpus benchmarks added

Refactoring

Split types.rs into submodules below 1000 LOC

Full diff: v0.1.0-alpha.2…v0.2.0

v0.1.0-alpha.2 (2026-02-13)

Second alpha release focused on plugin distribution improvements.

Features

CI per-platform pre-packaging for plugin binary distribution
Download-on-first-run fallback for marketplace installs

Full diff: v0.1.0-alpha.1…v0.1.0-alpha.2

v0.1.0-alpha.1 (2026-02-13)

Initial alpha release of markymark — a high-performance Markdown LSP server built in Rust. Includes core LSP capabilities: document symbols, go-to-definition, find-references, rename, hover, completion, and diagnostics for Markdown files with wiki-link support.

Full changelog: v0.1.0-alpha.1