v1.3.0

New

• SARIF output — mix ex_dna --format sarif generates a report compatible
  with GitHub Code Scanning, VS Code SARIF Viewer, and other standard tools.
• Clone budget for CI — mix ex_dna --max-clones 10 exits with code 1
  only when the count exceeds the budget. Useful for gradual adoption in
  brownfield projects.
• Near-miss detection scales to large codebases — Type-III detection
  reworked with an inverted index on structural sub-hashes. Previously choked
  on 200+ file projects; now handles 500+ files in seconds.
• Sibling window detection — Catches duplicated groups of adjacent
  def/defp that were previously invisible because they didn't share a
  common AST parent. For example, three consecutive functions copied between
  controllers are now detected even if the surrounding module code differs.
• Delegation pattern detection — def fetch(id), do: fetch(id, []) +
  def fetch(id, opts) are grouped as one unit. Duplicated wrapper+body
  pairs across modules are now caught.
• Struct/map field order doesn't matter — %User{name: x, age: y} and
  %User{age: y, name: x} match in Type-II mode.

Fixed

• Crash on %{acc | field: value} syntax in abstract mode.
• Crash on __MODULE__.function() calls during analysis.
• Crash when passing a list of paths — ExDNA.analyze(["lib/", "test/"])
  now works as documented.
• --max-clones output was printing literal text instead of actual numbers.
• Suggestions for clones with 3+ occurrences showed wrong call sites for
  the 3rd+ occurrence. Now each occurrence gets its own anti-unification.
• Cache wasn't invalidated when changing min_mass, literal_mode, or
  other detection options. Stale results were served silently.
• _ and __MODULE__ were renamed during variable normalization, causing
  false positives in Type-II detection.
• HTML report links were dead (href="#"). Now clickable file:// URIs
  that open in your editor.
• Behaviour suggestions fired for private functions. @callback only
  makes sense for def, not defp.
• Same-file clone diagnostics in LSP were missing cross-references to
  other locations within the same file.
• files_analyzed stat counted files that failed to parse. Now only
  counts successfully analyzed files.
• Glob patterns with ?, character classes, and edge cases now work
  correctly.

Improved

• Compiler runs full detection — The incremental compiler now finds
  Type-I, II, and III clones (previously only Type-I).
• Detection timing is accurate — detection_time_ms in stats reflects
  actual detection time, not wall clock including report generation.
• JSON output includes behaviour suggestions — Previously only console
  and HTML reports showed them.
• Config validation — Invalid options like min_mass: -1 or
  literal_mode: :foo now raise immediately with a clear message.
• HTML report uses EEx templates — Easier to customize.
• Zero dialyzer warnings — All previously suppressed errors resolved.

Performance

Benchmarked on real-world open-source projects with full Type-I/II/III
detection (min_mass: 30, literal_mode: :abstract, min_similarity: 0.85):

Project	Files	Clones	Time
Phoenix	74	15	0.3s
Ecto	56	20	0.5s
Oban	64	21	0.1s
Livebook	264	64	2.4s
Plausible	465	83	3.8s
Ash	554	524	9.8s