Changelog

0.7.0

37× faster joins: LATERAL join for line-range containment replaces nested loop over all fragments
8× faster structural search: (kind, name, arity) btree index bypasses GIN when pattern specifies a function name
2× faster text search: file-first search with LATERAL fragment lookup instead of joining all fragments per file
ParadeDB BM25 enabled by default (was opt-in)
Fixed BM25 index creation: removed bigint columns incompatible with pdb.literal in ParadeDB 0.21+
Fixed deadlock on concurrent term inserts: advisory lock per transaction instead of retry loop
Recommended Postgres tuning settings documented (max_parallel_workers_per_gather, shared_buffers, pg_prewarm)
Complete documentation rewrite: all guides updated for v0.7, hex indexing, performance tuning, architecture

mix exograph.index.hex — download and index Hex.pm packages in a streaming pipeline
Streaming: download tarball → extract to tmpdir → index → cleanup (disk = concurrency × 1 package)
Resume by default: skips already-indexed packages by name+version
Non-Elixir packages detected and skipped before disk write
--mode latest|top|all, --limit, --concurrency, --mirror, --cache-tarballs
CLI progress: per-package status icons, progress bar, rate, ETA
LiveView progress dashboard at /progress via PubSub (use --web flag)
Source viewer: click code icon to see full file with syntax highlighting and line highlight
pg_trgm GIN indexes on files.source and files.comments_text for fast ILIKE
Removed all full-table-scan fallbacks (ILIKE/regex in Postgres instead)
Deadlock retry for concurrent term inserts
Fixed non-semver version sorting in registry
New deps: hex_core ~> 0.15 (optional), req ~> 0.5 (optional)

Text and regex search modes in web UI and API (mode: "text" / mode: "regex")
pg_trgm GIN indexes on files.source and files.comments_text for fast ILIKE
Replaced all full-table-scan fallbacks with Postgres ILIKE/regex queries
Hex.pm source links on file paths (package version extracted from file paths)
Source viewer modal — click the code icon to see full file with highlighted match line
New guides: web-ui.md, api.md
Updated README with Web UI, JSON API sections; Installation points to Hex
Updated querying.md with text search, mix-tasks.md with mix exograph.web

Monaco editor with built-in Elixir syntax highlighting, auto-closing brackets, indentation
Elixir intellisense via Code.Fragment.cursor_context — completes DSL macros, modules, fields
IDE-style error diagnostics with red squiggly underlines and hover messages
Format button via Code.format_string!
Example query cards with descriptions
Collapsible package groups, loading spinner, sticky editor
URL query params (?q=) for shareable links
Hex.pm links on package names
Load more pagination
Thin scrollbar via Tailwind v4.3 scrollbar-thin
phoenix_iconify for icons

Cursor-based pagination (cursor param, next_cursor in response)
Rate limiting via Hammer (x-ratelimit-limit, x-ratelimit-remaining headers, 429 on excess)

DSL limit: clause parsed and respected
Kind filtering — def patterns only match def fragments, not modules/expressions
Keyset pagination replaces OFFSET in internal streaming (O(1) per page)
Preview shows correct absolute line numbers

Web UI with Monaco editor, syntax-highlighted search results, and autocompletion (mix exograph.web)
JSON API: POST /api/search, POST /api/query, GET /api/packages, GET /api/stats
Consolidated Exograph.Postgres.* namespace (FragmentStore, InvertedIndex, TreeStore)
Renamed Query → StructuralQuery to disambiguate from DSL.Query
Split DSL executor into focused modules (Predicates, Scope)
Removed dead code: legacy Planner subtree, unused Indexer delegator
Removed single-implementation behaviour modules (Extractor, FragmentStore, InvertedIndex, TreeStore)
Simplified Index struct from 6 fields to 3

Module renames — update any direct references:

Schema changes require a fresh index. Drop existing tables and re-migrate.

Initial public release.

Postgres-backed indexing for Elixir source code with integer primary keys and normalized terms
ExAST-verified structural search with GIN term index acceleration
Streaming batched fragment verification with early termination
Normalized Ecto schemas for files, fragments, comments, definitions, references, packages, versions
Optional Reach call graph extraction (graph nodes and call edges)
Ecto-shaped DSL with fragment/definition/reference/call-edge sources
Fragment joins with definitions, references, and call edges
Multi-join fragment queries (up to 3 joins)
Containing-function join semantics via precomputed end_line
Fact-first join execution for selective predicates
DSL plan validation (bindings, associations, structural predicates)
Package/version-scoped indexing and queries
Optional ParadeDB BM25 text/code-fact retrieval
ExDNA structural similarity search
Mix tasks for indexing and searching
Tested against 2000 Hex packages (1.8M fragments, 5.5M references, 1M call edges)