fcoreutils

High-performance GNU coreutils replacement in Rust — 100+ tools, SIMD-accelerated, drop-in compatible, cross-platform.

Independent Benchmarks (v0.9.2)

Source: AiBrush/coreutils-rs-independent-test — Linux x86_64, GitHub Actions, 100MB file, hyperfine

Summary: 107 tools tracked · 1059/1209 tests passed (87.6%) · fastest: wc at 31.5x vs GNU

Sizes are raw binary sizes. Compat is GNU test pass rate. Speedup is peak across all benchmark scenarios. - = no data collected yet for this tool/metric.

All Tools

Text Processing

Encoding/Decoding

Checksums

File Operations

Permissions

Text/Data Generation

Path Utilities

System Information

Process/Environment

Shell Utilities

Tools with Known Issues

Not Yet Tested

Installation

cargo install fcoreutils

Or build from source:

git clone https://github.com/AiBrush/fcoreutils.git
cd fcoreutils
cargo build --release

Binaries are in target/release/.

Usage

Each tool is prefixed with f to avoid conflicts with system utilities:

# Word count (drop-in replacement for wc)
fwc file.txt
fwc -l file.txt          # Line count only
fwc -w file.txt          # Word count only
fwc -c file.txt          # Byte count only (uses stat, instant)
fwc -m file.txt          # Character count (UTF-8 aware)
fwc -L file.txt          # Max line display width
cat file.txt | fwc       # Stdin support
fwc file1.txt file2.txt  # Multiple files with total

# Cut (drop-in replacement for cut)
fcut -d: -f2 file.csv    # Extract field 2 with : delimiter
fcut -d, -f1,3-5 data.csv  # Multiple fields
fcut -b1-20 file.txt     # Byte range selection

# Hash tools (drop-in replacements)
fsha256sum file.txt       # SHA-256 checksum
fmd5sum file.txt          # MD5 checksum
fb2sum file.txt           # BLAKE2b checksum
fsha256sum -c sums.txt    # Verify checksums

# Base64 encode/decode
fbase64 file.txt          # Encode to base64
fbase64 -d encoded.txt    # Decode from base64
fbase64 -w 0 file.txt     # No line wrapping

# Sort, translate, deduplicate, reverse
fsort file.txt            # Sort lines alphabetically
fsort -n file.txt         # Numeric sort
ftr 'a-z' 'A-Z' < file   # Translate lowercase to uppercase
ftr -d '[:space:]' < file # Delete whitespace
funiq file.txt            # Remove adjacent duplicates
funiq -c file.txt         # Count occurrences
ftac file.txt             # Print lines in reverse order

# File viewing and transformation
fhead -n 20 file.txt      # First 20 lines
ftail -n 20 file.txt      # Last 20 lines
ftail -f logfile.txt      # Follow file for new lines
fcat file1.txt file2.txt  # Concatenate files
fcat -n file.txt          # With line numbers
frev file.txt             # Reverse each line

# Text formatting
fexpand file.txt          # Convert tabs to spaces
funexpand file.txt        # Convert spaces to tabs
ffold -w 80 file.txt      # Wrap lines at 80 columns
fnl file.txt              # Number lines
fpaste file1 file2        # Merge files line by line
fpaste -s file.txt        # Serial mode (join all lines)

# Set operations on sorted files
fcomm file1 file2         # Compare two sorted files
fcomm -12 file1 file2     # Only lines common to both
fjoin file1 file2         # Join on common field
fjoin -t, -1 2 -2 1 a b  # Join CSV files on specific fields

Key Optimizations

• Zero-copy mmap: Large files are memory-mapped directly, avoiding copies
• SIMD scanning: memchr crate auto-detects AVX2/SSE2/NEON for byte searches
• stat-only byte counting: wc -c uses stat() without reading file content
• Hardware-accelerated hashing: sha2 detects SHA-NI, blake2 uses optimized implementations
• SIMD base64: Vectorized encode/decode with 4MB chunked streaming
• Parallel processing: Multi-file hashing and wc use thread pools
• SIMD range translate/delete: tr detects contiguous byte ranges and uses AVX2/SSE2 SIMD
• Chunk-based reverse scan: tac processes backward in 512KB chunks with forward SIMD within each chunk
• Optimized release profile: Fat LTO, single codegen unit, abort on panic, stripped binaries

GNU Compatibility

Output is byte-identical to GNU coreutils. All flags are supported including --files0-from, --total, --complement, --check, and correct column alignment.

Assembly Optimization Path

We are pursuing a second optimization track alongside Rust: hand-crafted x86_64 assembly for platforms where maximum throughput matters. We started with yes — it is simple enough to implement completely and serves as a proof-of-concept for the approach.

Our assembly yes achieves ~2.6 GB/s (1.89x faster than GNU yes, 1.25x faster than our Rust implementation) while compiling to under 1,300 bytes with no runtime dependencies.

Benchmarked on Linux x86_64. At pipe-limited throughput all three write at ~2.1 GB/s. The assembly wins on binary size (25x smaller), memory (70x less RSS), and startup latency (3x faster).

On Linux x86_64 and Linux ARM64, releases ship the assembly binary. All other platforms (macOS, Windows) use the Rust implementation. The assembly binary is a static ELF with only two syscalls (write and exit/exit_group), no dynamic linker, and a non-executable stack.

Our priority remains 100% GNU compatibility in Rust first. We will pursue assembly implementations for additional commands over time, as the tooling and verification process matures. The goal is not to rush assembly ports but to do them right — with full security review and byte-for-byte compatibility testing.

See assembly/yes/ for the source and tests/assembly/ for the test suite.

Roadmap

We are actively working toward 100% compatibility with GNU coreutils — byte-identical output, same exit codes, and matching error messages for all 90+ tools. Once we achieve full compatibility, we will focus on performance optimization targeting 10-30x speedup over GNU coreutils across all tools.

Contributing

We welcome contributions! Please see CONTRIBUTING.md for guidelines.

This project follows the Contributor Covenant Code of Conduct.

Architecture

See ARCHITECTURE.md for design decisions and PROGRESS.md for development status.

Security

To report a vulnerability, please see our Security Policy.

License

MIT