Adding `dashem` library as a dependency

Run this command in a terminal, in your project's directory:

cargo add dash-em

To add it manually, edit your project's Cargo.toml file and add to the [dependencies] section:

dash-em = "1.1.2"

The dashem library will be automatically available globally. Read the dashem library documentation.

Back to the crate overview.

Readme

dash-em

Enterprise-Grade Em-Dash Removal Infrastructure — Leveraging Advanced SIMD Vectorization for Optimal Character Stream Processing

Overview

dash-em is an absurdly over-engineered, deliberately meme-grade, production-ready, enterprise-certified string manipulation library designed with singular, unwavering purpose—removing em-dashes (U+2014) from UTF-8 encoded text—with unprecedented obsession.

Building upon decades of accumulated wisdom in systems programming—combined with cutting-edge SIMD acceleration techniques—dash-em delivers truly unnecessary—yet deeply satisfying—performance characteristics in the em-dash elimination category.

🎭 MEME REPOSITORY DISCLOSURE — This project is a deliberately absurd, tongue-in-cheek exploration of over-engineering. The em-dash removal use case is intentionally ridiculous. This is not serious production software, despite being written with genuine engineering rigor. Enjoy the absurdity.

Key Value Propositions

⚡ SIMD-Accelerated Processing — Employing SSE4.2, AVX, AVX2, AVX-512F, and ARM NEON instruction sets for—optimal throughput
🚀 Extraordinary Performance — Up to 538x faster than byte-level iteration—in language bindings
🔒 Memory-Safe Architecture — Engineered with defensive programming—paradigms throughout
📦 Zero External Dependencies — Pure C implementation—no transitive dependency chains
🌍 True Cross-Platform Support — Linux, macOS, Windows—and ARM-based systems—all supported
🎯 Polyglot Language Support — 20+ language bindings—ensuring accessibility across heterogeneous technology stacks
🏢 Enterprise-Ready Infrastructure — Battle-tested, production-hardened, deployable—at scale

Features

What makes dash-em fast?

Instead of checking characters one by one—which is slow—dash-em uses SIMD instructions to process 16–64 bytes in parallel. Modern CPUs can do this crazy fast—we just have to tell them what to do.

Real-world speedups (measured across multiple architectures):

Core C library: 5x-11x faster than scalar implementation—depending on CPU architecture
Python bindings: 211x-538x faster than byte-level iteration
JavaScript bindings: 2x-31x faster than byte-level Buffer manipulation
Best case (no em-dashes—fast path): Up to 15.58 GB/s throughput on modern x86-64

How it works

The library auto-detects your CPU and picks the fastest path:

AVX-512F (if available) — 64 bytes per iteration. Cutting edge. Stupid fast.
AVX2 (fallback) — 32 bytes per iteration. Still very fast. Works on most modern CPUs.
SSE4.2 (older systems) — 16 bytes per iteration. Slower but still beats naive approaches.
ARM NEON (ARM/Apple Silicon) — 16 bytes per iteration. Works on servers and M-series Macs.
Scalar (last resort) — One byte at a time. Works everywhere.

Optimizations under the hood

Fast path for tiny strings — If you're only removing dashes from a few characters, we skip SIMD overhead
Loop unrolling — Process multiple chunks per iteration to keep the CPU pipeline full
Cache prefetching — Tell the CPU to load the next chunk early so it's ready when we need it
Bitmask matching — Use clever bit tricks to find patterns instead of checking bytes individually
Smart memory operations — Bulk copy unchanged regions instead of processing byte-by-byte

It's absurdly optimized. Maybe too optimized. But it works—and it's fast.

Performance

Core Library Performance

Multi-architecture SIMD performance (statistical benchmarks):

Pattern	macos-14-aarch64	windows-2022-msvc	ubuntu-22.04-gcc	ubuntu-22.04-clang
sparse	3.89 GB/s (2.63x)	12.33 GB/s (11.47x)	12.78 GB/s (8.71x)	9.67 GB/s (6.61x)
moderate	2.73 GB/s (1.82x)	7.23 GB/s (6.55x)	7.55 GB/s (5.58x)	6.93 GB/s (5.10x)
dense	1.53 GB/s (0.67x)	1.54 GB/s (1.06x)	2.57 GB/s (1.07x)	2.18 GB/s (1.09x)
alternating	1.53 GB/s (0.67x)	1.58 GB/s (1.09x)	2.57 GB/s (1.07x)	2.18 GB/s (0.94x)
boundary	3.28 GB/s (2.16x)	9.84 GB/s (9.24x)	10.45 GB/s (7.45x)	9.57 GB/s (6.88x)
no	4.27 GB/s (2.90x)	12.02 GB/s (11.10x)	15.58 GB/s (10.51x)	13.66 GB/s (9.21x)

Language Bindings Performance

Comparing dash-em bindings against native byte-level implementations:

Language	Test Pattern	Native (μs)	dash-em (μs)	Speedup
javascript	alternating	66.1	19.0	3.48x
javascript	dense	127.4	53.6	2.38x
javascript	moderate	302.7	50.0	6.05x
javascript	no	2932.3	94.1	31.14x
javascript	sparse	2861.1	107.6	26.58x
python	alternating	3624.9	15.5	234.20x
python	dense	7577.0	31.2	243.19x
python	moderate	18888.8	73.3	257.74x
python	no	186188.6	345.9	538.21x
python	sparse	193434.6	915.9	211.21x

Installation

C/C++ Core Library

git clone https://github.com/Gaurav-Gosain/dash-em
cd dash-em
mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release
make
sudo make install

Language-Specific Bindings

JavaScript/TypeScript (Node.js)

npm install dash-em

String API (easy to use):

const dashem = require('dash-em');
const result = dashem.remove('Hello—world');
console.log(result);  // Output: Helloworld

Buffer API (high-performance, zero-copy):

const dashem = require('dash-em');

// Process Buffer directly (10-26x faster than string API)
const input = Buffer.from('Hello—world', 'utf-8');
const output = dashem.removeBuffer(input);
console.log(output.toString('utf-8'));  // Output: Helloworld

// In-place modification (fastest, modifies input)
const buffer = Buffer.from('Hello—world', 'utf-8');
const newLength = dashem.removeBufferInPlace(buffer);
console.log(buffer.slice(0, newLength).toString('utf-8'));  // Output: Helloworld

Python

pip install dash-em

import dashem
result = dashem.remove('Hello—world')
print(result)  # Output: Helloworld

Go

go get github.com/Gaurav-Gosain/dash-em/go

package main

import (
    "fmt"

    dashem "github.com/Gaurav-Gosain/dash-em/go"
)

func main() {
    result, _ := dashem.Remove("Hello—world")
    fmt.Println(result)  // Output: Helloworld
}

Rust

[dependencies]
dash-em = "1.0"

fn main() {
    let result = dash_em::remove("Hello—world").unwrap();
    println!("{}", result);  // Output: Helloworld
}

Java

public class Example {
    public static void main(String[] args) {
        String result = Dashem.remove("Hello—world");
        System.out.println(result);  // Output: Helloworld
    }
}

C# / .NET

string result = Dashem.Remove("Hello—world");
Console.WriteLine(result);  // Output: Helloworld

PHP

<?php
$result = dashem_remove('Hello—world');
echo $result;  // Output: Helloworld
?>

Ruby

require 'dashem'
result = Dashem.remove('Hello—world')
puts result  # Output: Helloworld

Swift

import Dashem
let result = removeEmDashes("Hello—world")
print(result)  // Output: Helloworld

Additional Language Bindings

Comprehensive bindings are provided for—and thoroughly tested against—the following languages:

Kotlin — Native interop with dash-em core
R — Rcpp-based integration layer
Dart — dart:ffi bindings for cross-platform applications
Scala — Native compilation via Scala Native
Perl — XS extension module—providing optimal performance characteristics
Lua — Lightweight C API integration
Haskell — Pure FFI bindings—maintaining functional purity
Elixir — NIF-based native implementation—ensuring BEAM compatibility
Zig — C ABI import with modern language ergonomics
Objective-C — Direct C interoperability layer

WebAssembly

dash-em compiles to—high-performance WebAssembly modules supporting multiple target specifications:

# wasm32 (Emscripten)
cd bindings/wasm && ./build.sh

# WASI (WebAssembly System Interface)
WASI_SDK_PATH=/opt/wasi-sdk ./build.sh wasi

Architecture

The dispatch system checks your CPU once, then uses the best available implementation for all subsequent calls:

graph TD
    A["dashem_remove()"] --> B{Input < 32 bytes?}
    B -->|Yes| C["fast_small() scalar"]
    B -->|No| D["Check CPU capabilities<br/>(cached after first call)"]
    D --> E{AVX-512?}
    E -->|Yes| F["dashem_remove_avx512"]
    E -->|No| G{AVX2?}
    G -->|Yes| H["dashem_remove_avx2_unrolled"]
    G -->|No| I{SSE4.2?}
    I -->|Yes| J["dashem_remove_sse42"]
    I -->|No| K{ARM NEON?}
    K -->|Yes| L["dashem_remove_neon"]
    K -->|No| M["dashem_remove_scalar"]
    C --> N["Output"]
    F --> N
    H --> N
    J --> N
    L --> N
    M --> N

API Reference

C API

/**
 * Remove em-dashes from UTF-8 string
 *
 * @param input       Input UTF-8 string
 * @param input_len   Length of input in bytes
 * @param output      Output buffer
 * @param output_cap  Output buffer capacity
 * @param output_len  Output length (set on return)
 * @return 0 on success, -1 on buffer overflow, -2 on invalid input
 */
int dashem_remove(
    const char *input,
    size_t input_len,
    char *output,
    size_t output_capacity,
    size_t *output_len
);

/**
 * Get library version
 * @return Version string (e.g., "1.0.0")
 */
const char* dashem_version(void);

/**
 * Get active implementation name
 * @return Implementation name (e.g., "AVX2", "SSE4.2", "Scalar")
 */
const char* dashem_implementation_name(void);

/**
 * Detect available CPU features
 * @return Bitmask of DASHEM_CPU_* flags
 */
uint32_t dashem_detect_cpu_features(void);

JavaScript/Node.js API

/**
 * Remove em-dashes from a string
 * @param {string} input - Input string
 * @returns {string} String with em-dashes removed
 */
dashem.remove(input);

/**
 * Remove em-dashes from a Buffer (zero-copy, high-performance)
 * @param {Buffer} buffer - Input Buffer
 * @returns {Buffer} New Buffer with em-dashes removed
 */
dashem.removeBuffer(buffer);

/**
 * Remove em-dashes from a Buffer in-place (ultra-fast, modifies input!)
 * WARNING: This modifies the input Buffer
 * @param {Buffer} buffer - Input Buffer (will be modified)
 * @returns {number} New length of valid data in buffer
 */
dashem.removeBufferInPlace(buffer);

/**
 * Get library version
 * @returns {string} Version string
 */
dashem.version();

/**
 * Get implementation name
 * @returns {string} Implementation name (e.g., "AVX2", "SSE4.2")
 */
dashem.implementationName();

Performance Notes:

remove(): Easy to use but includes UTF-8 conversion overhead
removeBuffer(): 10-26x faster than remove(), zero-copy operation
removeBufferInPlace(): Fastest option, modifies input buffer directly

Running Benchmarks

Performance benchmarks are automatically updated via GitHub Actions. See the Performance section above for the latest results across all architectures and language bindings.

To run benchmarks locally:

C/C++ Core Library:

cd build && ./bench_dashem

Multi-Language Benchmarks:

cd benchmarks
./run_all_benchmarks.sh

Results are generated in JSON format for easy integration with continuous performance monitoring systems.

Testing

Comprehensive test suites—validated across all supported platforms—ensure—correctness and reliability:

# C/C++ tests
cd build && ctest

# Language-specific tests
npm test              # JavaScript
python -m pytest      # Python
cargo test            # Rust
go test ./...         # Go

Continuous Integration

dash-em leverages GitHub Actions—to ensure—consistent quality across:

✓ Linux (x86_64, ARM64)—builds and tests
✓ macOS (Intel, Apple Silicon)—native execution
✓ Windows (MSVC, MinGW)—compatibility verification
✓ WebAssembly (Emscripten, WASI)—cross-compilation
✓ All language bindings—comprehensive integration testing

Contributing

Contributions are welcome! Please ensure:

Code adheres to—professional C/C++ standards—with comprehensive documentation
Commit messages are—descriptive and—reference relevant issues
All tests pass—before submitting—pull requests
Performance characteristics are—benchmarked against—baseline implementations

License

MIT License — See LICENSE file for details.

Citation

If dash-em is utilized in—academic or—commercial contexts, please reference:

@software{gosain2025dashem,
  title={dash-em: Enterprise-Grade Em-Dash Removal Infrastructure},
  author={Gosain, Gaurav},
  year={2025},
  url={https://github.com/Gaurav-Gosain/dash-em}
}

Acknowledgments

This project exists because—em-dashes matter—and they deserve—the most efficient, highly optimized—removal mechanism—available on modern computing platforms.

Building excellence—one em-dash at a time. —

Version: 1.0.1 | Status: Production-Ready | License: MIT | Repository: github.com/Gaurav-Gosain/dash-em

Adding dashem library as a dependency

Adding `dashem` library as a dependency