ix

Sub-millisecond code search via sparse trigram indexing.

ix pre-computes a byte-level trigram index to narrow search candidates to a fraction of the total file set, then verifies matches with a memory-constant streaming architecture. This eliminates the linear-scan bottleneck that slows grep and ripgrep on large codebases.

Install

cargo install moeix

This installs two binaries:

ix — CLI search tool
ixd — background daemon for continuous indexing (requires notify feature, enabled by default)

Quick Start

# Index a directory (defaults to current directory if no path given)
ix --build
ix --build /path/to/repo

# Search
ix "fn validate"

# Search with regex
ix --regex "fn\s+\w+_handler"

# Search with context lines
ix --context 3 "TODO"

# Negation filter (exclude matches)
ix "error" --negate "test"

# Show query statistics
ix --stats "struct Config"

How It Works

Index — ix --build walks the directory, extracts byte-level trigrams from every file, and writes a compressed index to .ix/shard.ix.
Plan — On search, the query is decomposed into trigrams. The index is consulted to find candidate files that contain all required trigrams.
Verify — Candidates are streamed through a regex matcher with constant memory usage, producing precise line-level results.

Index Format (v1.3)

All integers are little-endian, all offsets absolute from file start, 8-byte aligned sections.

Section	Description
Header	256 bytes: magic `IX01`, version, flags, section offsets
File table	Per-file metadata: path hash, content hash, size, posting offset
Trigram table (CDX)	Delta-encoded + varint + ZSTD compressed in 1024-entry blocks
Block index	`(u32 first_key, u64 block_offset)` × N + sentinel
Posting lists	Per-trigram file IDs, delta-encoded + varint + ZSTD
String pool	Interned file paths

CDX compression is always-on since v1.3. The reader does a two-level search: block index → decompress block → linear scan.

Not backward compatible with v1.1 or v1.2. Rebuild indexes after upgrading:

rm -rf .ix/
ix --build .

Daemon

ixd watches a directory for file changes and incrementally updates the index:

ixd /path/to/repo

The daemon exposes a Unix domain socket for external consumers (editors, tooling):

$XDG_RUNTIME_DIR/ixd/{hash}.sock

Protocol is NDJSON — one JSON object per newline-terminated line. Push notifications for file changes and status updates; query/response for history and status queries.

ix CLI does not use the socket — it reads the index file directly.

Performance

Metric	Value
Index ratio	~4× source size (ZSTD level 3)
Selective query (10% match)	40ms — scans 10× fewer files than ripgrep
Small dataset (all match)	305ms — ripgrep wins on small/all-match workloads
Cold start	<3s
Hot path p99	<50ms

ix wins when the trigram index can eliminate most files from scanning. On small repos or queries where every file matches, linear-scan tools like ripgrep are faster.

Feature Flags

Flag	Default	Description
`notify`	yes	File watcher + daemon (`ixd`)
`decompress`	no	gz/zst/bz2/xz decompression
`archive`	no	zip/tar archive support
`full`	no	All optional features

Library

ix is also a library (moeix on crates.io, ix as the crate name):

[dependencies]
moeix = "0.5"

use ix::{Reader, Executor};

let reader = Reader::open(".ix/shard.ix")?;
let mut executor = Executor::new(&reader);
let matches = executor.execute(/* query */);

See docs.rs/moeix for full API documentation.

Building

cargo build --all-features
cargo test --all-features
cargo clippy --all-features -- -D warnings

Requires Rust 1.85+.

License

MIT

Clean-Before-Build

The daemon uses a "clean-before-build" pattern to prevent stale file descriptor bugs:

Old temp files are cleaned at the start of each build (not at the end)
Fresh writers are initialized for each build
No temp file accumulation across consecutive builds
Prevents inode exhaustion on Linux

This fixes the critical bug where incremental rebuilds failed with "I/O: No such file or directory (os error 2)" after the first successful build.

moeix 0.6.2

ix