🥮 mooncakes.io
    Sign in

    rg

    A port of ripgrep (rg) to MoonBit, for native and WebAssembly

    grep
    ripgrep
    search
    Download zip
    Version
    0.2.0
    License
    MIT
    Repository
    Last updated
    57 minutes ago
    Downloads
    2

    #rg.mbt — ripgrep ported to MoonBit

    A port of ripgrep (version 15.2.0) to MoonBit. The goal is behavioral fidelity: the same flags, output format, error messages and exit codes as rg.

    It builds for two backends:

    • native (the default): uses a small C stub for OS access.
    • wasm (linear-memory WebAssembly): uses a subset of WASI preview 1 that is supported by both wasmtime and MoonBit's own moonrun.

    #Installing

    The module is published on mooncakes.io as moonbit-community/rg. To build the rg binary from source:

    git clone https://github.com/moonbit-community/rg.mbt
cd rg.mbt
moon build --release cmd/rg

    #Building and running

    # native
moon build --release cmd/rg
_build/native/release/build/cmd/rg/rg.exe --help

# wasm, via moon's runner (paths are relative to the current directory)
moon run --target wasm cmd/rg -- PATTERN some/dir

# wasm, via wasmtime (preopen the directories you want to search)
moon build --release --target wasm cmd/rg
wasmtime run --dir . _build/wasm/release/build/cmd/rg/rg.wasm PATTERN .

    #Layout

    Each ripgrep crate maps to a MoonBit package:

    ripgrep cratepackagenotes
    regex-syntaxregex/syntax, regex/unicodeparser, HIR, Unicode tables, literal extraction
    regex-automataregex/automata, regexThompson NFA, PikeVM, lazy DFA, prefilters, meta regex
    grep-matchermatcherthe Matcher trait
    grep-regexgrep_regexRegexMatcher
    grep-searchersearcherline-oriented searching, binary detection, transcoding
    grep-printerprinter, termcolorstandard, summary and JSON output
    globsetglobset
    ignoreignoregitignore, file types, overrides, directory walking
    grep-clicli
    ripgrep (core)core, cmd/rgflags, help/man/completions, search driver
    (std)sysOS abstraction for native and WASI

    The regex engine is written from scratch in MoonBit following the design of Rust's regex-syntax/regex-automata (leftmost-first semantics, byte oriented, Unicode aware), and is tested against the regex crate's TOML test corpus (regex/internal/testdata).

    #Testing

    moon test                 # native
moon test --target wasm   # wasm

    ripgrep's own integration suite (tests/*.rs in the ripgrep repository) can be run against this port. It needs a ripgrep checkout in .repos/ripgrep (CI uses revision 3fce3b5bb0) and a Rust toolchain that can build it:

    scripts/run_rg_integration_tests.sh native
scripts/run_rg_integration_tests.sh wasm     # needs wasmtime

    platformtestsnativewasm (wasmtime)
    Linux332all pass306 pass, 26 skipped
    macOS323all pass302 pass, 21 skipped

    The tests skipped on wasm can't pass under WASI, and they would fail for ripgrep built for WASI too. They need to spawn processes (-z/--search-zip, --pre), detect stdin as readable when no path is given, follow symlinks with absolute targets (refused by wasmtime's sandbox), use memory maps, or use file names that aren't valid UTF-8. CI runs both suites on Linux.

    #Performance

    The hot paths (memchr, substring search, and Teddy for multiple literals) are vectorized in pure MoonBit with moonbitlang/core/v128, which lowers to NEON/SSE on native and to wasm SIMD on wasm. Like Rust's memchr, they test 32–64 bytes per step with a cheap "any match" check, and the search loops avoid allocating per line or per match.

    Time relative to ripgrep 15.2.0 (hyperfine, 170 MB text file, Apple M2 Ultra; ripgrep's -j1 --no-mmap is the same speed on a single file):

    searchnativewasm (wasmtime)wasm (moonrun / V8)
    rare literal / no match1.0–1.1x2.0–2.1x2.9–3.1x
    case-insensitive literal (-i)1.1x2.2x2.3x
    alternation of literals1.0x2.0x2.4x
    regex with a literal suffix ([A-Z][a-z]+Error)1.1x2.7x3.2x
    regex dominated by the DFA (\w+_\w+\()0.9x2.5x1.9x
    Unicode class (\p{Greek})1.0x4.0x2.7x
    many matches (-c self, -n self, -w, -o)1.2–1.3x3.0–3.7x3.0–3.7x
    -v (every other line through the sink)1.3x3.2x2.9x
    -C2 (context lines)1.6x5.2x4.2x

    V8 is faster on compute-heavy searches; wasmtime starts about 5 ms faster, which dominates short searches.

    Searching a directory tree is at parity with rg -j1; ripgrep's default parallel search is faster on trees. On wasm, MoonBit currently emits bounds checks and reference counting as out-of-line runtime calls, which accounts for most of the remaining gap to native.

    #Differences from ripgrep

    • No PCRE2 (--pcre2 reports that PCRE2 is unavailable, like a build of ripgrep without the pcre2 feature).
    • Searching is single threaded (-j/--threads is accepted; output is the same as rg -j1, including the single-threaded walker's error messages, e.g. IO error for operation on <dir>: ... for unreadable directories).
    • No memory maps. Where ripgrep would use one (on Linux, when searching a few files given explicitly, or with --mmap), the whole file is read and searched as a slice instead, which gives the same output.
    • Where ripgrep 15.2.0 panics, the port does something sensible instead: --help/--version in a config file are ignored, and JSON submatches of a multi-line match truncated by -m are clipped to the reported lines.
    • --regex-size-limit uses regex-automata's accounting of NFA size, so the smallest accepted limit is the same as ripgrep's for most patterns; for patterns with large Unicode classes it can be slightly lower (by a few percent).
    • On wasm:
      • No subprocesses: --pre reports that the preprocessor could not start, and -z/--search-zip searches compressed files as-is (as ripgrep does when the decompression tool isn't installed).
      • No file timestamps: --sort modified etc. report that sorting isn't supported.
      • Stdin is only searched when - is given explicitly (ripgrep built for WASI behaves the same way, since the type of stdin can't be queried).
      • Only preopened directories are reachable. wasmtime run --dir . --dir / gives full access (except to symlinks with absolute targets); moonrun only exposes the current directory, so ignore files in parent directories and absolute paths outside it are not visible.

    Powered by MoonBit

    Site sourceReport issuePackagesBuild queueSkillsStatistics

    © 2026 mooncakes.io