msy 0.4.5

use std::fs::File;
use std::io::{self, Read, Write};
use std::path::Path;
use std::str::FromStr;

/// Compression algorithm
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum Compression {
	None,
	/// LZ4: 23 GB/s, lower compression ratio (good for low-CPU scenarios)
	Lz4,
	/// Zstd level 3: 8.7 GB/s, better compression ratio (default)
	Zstd,
}

impl FromStr for Compression {
	type Err = String;

	fn from_str(s: &str) -> Result<Self, Self::Err> {
		match s.to_lowercase().as_str() {
			"none" => Ok(Self::None),
			"lz4" => Ok(Self::Lz4),
			"zstd" => Ok(Self::Zstd),
			_ => Err(format!("Unknown compression type: {}", s)),
		}
	}
}

impl Compression {
	#[allow(dead_code)] // Used in debug logging
	pub fn as_str(&self) -> &'static str {
		match self {
			Self::None => "none",
			Self::Lz4 => "lz4",
			Self::Zstd => "zstd",
		}
	}
}

/// Compress data
pub fn compress(data: &[u8], compression: Compression) -> io::Result<Vec<u8>> {
	match compression {
		Compression::None => Ok(data.to_vec()),
		Compression::Lz4 => compress_lz4(data),
		Compression::Zstd => compress_zstd(data),
	}
}

/// Compress data from a reader to a writer (streaming)
///
/// This avoids loading the entire file into memory by compressing in chunks.
/// Suitable for large files that would otherwise cause OOM.
#[allow(dead_code)] // Reserved for future use if sy-remote protocol is redesigned
pub fn compress_streaming<R: Read, W: Write>(reader: &mut R, writer: &mut W, compression: Compression) -> io::Result<()> {
	match compression {
		Compression::None => {
			// No compression, just copy
			std::io::copy(reader, writer)?;
			Ok(())
		}
		Compression::Lz4 => {
			// LZ4 frame format supports streaming
			let mut encoder = lz4_flex::frame::FrameEncoder::new(writer);
			std::io::copy(reader, &mut encoder)?;
			encoder.finish()?;
			Ok(())
		}
		Compression::Zstd => {
			// Zstd supports streaming natively
			let mut encoder = zstd::Encoder::new(writer, 3)?;
			std::io::copy(reader, &mut encoder)?;
			encoder.finish()?;
			Ok(())
		}
	}
}

/// Decompress data (used by sy-remote binary)
#[allow(dead_code)] // Used by sy-remote binary, not library code
pub fn decompress(data: &[u8], compression: Compression) -> io::Result<Vec<u8>> {
	match compression {
		Compression::None => Ok(data.to_vec()),
		Compression::Lz4 => decompress_lz4(data),
		Compression::Zstd => decompress_zstd(data),
	}
}

fn compress_lz4(data: &[u8]) -> io::Result<Vec<u8>> {
	// LZ4: 23 GB/s throughput (benchmarked), lower CPU usage
	Ok(lz4_flex::compress_prepend_size(data))
}

#[allow(dead_code)] // Called by decompress() which is used by sy-remote
fn decompress_lz4(data: &[u8]) -> io::Result<Vec<u8>> {
	lz4_flex::decompress_size_prepended(data).map_err(|e| io::Error::new(io::ErrorKind::InvalidData, e))
}

fn compress_zstd(data: &[u8]) -> io::Result<Vec<u8>> {
	// Level 3: 8.7 GB/s throughput (benchmarked), optimal balance
	let mut encoder = zstd::Encoder::new(Vec::new(), 3)?;
	encoder.write_all(data)?;
	encoder.finish()
}

#[allow(dead_code)] // Called by decompress() which is used by sy-remote
fn decompress_zstd(data: &[u8]) -> io::Result<Vec<u8>> {
	let mut decoder = zstd::Decoder::new(data)?;
	let mut result = Vec::new();
	decoder.read_to_end(&mut result)?;
	Ok(result)
}

/// List of file extensions that are already compressed
/// Compressing these files provides minimal benefit
const COMPRESSED_EXTENSIONS: &[&str] = &[
	// Images
	"jpg", "jpeg", "png", "gif", "webp", "avif", "heic", "heif", // Video
	"mp4", "mkv", "avi", "mov", "webm", "m4v", "flv", "wmv", // Audio
	"mp3", "m4a", "aac", "ogg", "opus", "flac", "wma", // Archives
	"zip", "gz", "bz2", "xz", "7z", "rar", "tar.gz", "tgz", "tar.bz2", // Documents
	"pdf", "docx", "xlsx", "pptx", // Other
	"wasm", "br", "zst",
];

/// Check if file extension indicates already-compressed data
pub fn is_compressed_extension(filename: &str) -> bool {
	if let Some(ext) = filename.rsplit('.').next() {
		COMPRESSED_EXTENSIONS.iter().any(|&e| ext.eq_ignore_ascii_case(e))
	} else {
		false
	}
}

/// Determine if we should compress based on file size, extension, and network conditions
///
/// NOTE: Benchmarks show compression is MUCH faster than originally assumed:
/// - LZ4: 23 GB/s (not 400-500 MB/s as originally thought)
/// - Zstd: 8 GB/s (level 3)
///
/// CPU is NEVER the bottleneck - network always is, even on 100 Gbps!
#[allow(dead_code)] // Public API for future use
pub fn should_compress_adaptive(filename: &str, file_size: u64, is_local: bool, network_speed_mbps: Option<u64>) -> Compression {
	// LOCAL: Never compress (disk I/O is bottleneck, not network/CPU)
	if is_local {
		return Compression::None;
	}

	// HIGH SPEED NETWORK: If network > 1 Gbps (125 MB/s), compression might be CPU bottleneck
	// or simply unnecessary.
	//
	// While Zstd/LZ4 are very fast (GB/s), SSH encryption + Compression overhead can
	// reduce throughput on very fast links.
	// Threshold: 500 Mbps (approx 60 MB/s) - typically internet/WAN speeds are below this.
	// If we are seeing >500 Mbps, we are likely on a fast LAN or datacenter link.
	if let Some(speed) = network_speed_mbps {
		const HIGH_SPEED_THRESHOLD_MBPS: u64 = 500;
		if speed > HIGH_SPEED_THRESHOLD_MBPS {
			return Compression::None;
		}
	}

	// Skip small files (overhead > benefit)
	if file_size < 1024 * 1024 {
		return Compression::None;
	}

	// Skip very large files (would load entire file into RAM)
	// Max 256MB for compression to avoid OOM on large files
	//
	// WHY THIS LIMIT:
	// - sy-remote receive-file protocol requires buffering entire compressed data
	// - Files >256MB use SFTP instead (already efficient, chunks internally)
	// - True streaming compression would require protocol redesign
	// - 256MB covers 99% of compressible files (logs, code, text files)
	//
	// NOTE: compress_streaming() exists for future use if protocol supports it
	const MAX_COMPRESSIBLE_SIZE: u64 = 256 * 1024 * 1024;
	if file_size > MAX_COMPRESSIBLE_SIZE {
		return Compression::None;
	}

	// Skip already-compressed formats (jpg, mp4, zip, etc.)
	if is_compressed_extension(filename) {
		return Compression::None;
	}

	// BENCHMARKED DECISION:
	// Zstd at level 3 compresses at 8 GB/s (64 Gbps equivalent)
	// This is faster than ANY network, so always use it for best compression ratio
	// LZ4 is faster (23 GB/s) but worse ratio, only needed if Zstd bottlenecks
	//
	// Reality: Even 100 Gbps networks (12.5 GB/s) won't bottleneck on Zstd
	// Therefore: Always use Zstd for network transfers
	Compression::Zstd
}

/// Determine if we should compress based on file size and extension
/// (Legacy function for backward compatibility)
#[allow(dead_code)] // Public API for future use
pub fn should_compress(filename: &str, file_size: u64) -> Compression {
	should_compress_adaptive(filename, file_size, false, None)
}

/// Detect file compressibility by sampling first 64KB with LZ4
///
/// Returns compression ratio (compressed_size / original_size)
/// - Ratio < 0.9 means compressible (>10% savings)
/// - Ratio >= 0.9 means incompressible (<10% savings)
///
/// Uses LZ4 for fast testing (23 GB/s throughput)
/// Inspired by BorgBackup's auto-compression heuristic
pub fn detect_compressibility(file_path: &Path) -> io::Result<f64> {
	const SAMPLE_SIZE: usize = 64 * 1024; // 64KB sample

	let mut file = File::open(file_path)?;
	let mut buffer = vec![0u8; SAMPLE_SIZE];
	let bytes_read = file.read(&mut buffer)?;

	// Empty file or very small file
	if bytes_read == 0 {
		return Ok(1.0); // No benefit
	}

	let sample = &buffer[..bytes_read];
	let compressed = compress_lz4(sample)?;

	// Calculate compression ratio
	let ratio = compressed.len() as f64 / sample.len() as f64;

	Ok(ratio)
}

/// Compression detection mode
#[derive(Debug, Clone, Copy, Default, PartialEq, Eq, clap::ValueEnum)]
pub enum CompressionDetection {
	/// Content-based detection with sampling (default)
	#[default]
	Auto,

	/// Extension-only detection (legacy behavior)
	Extension,

	/// Always compress (override detection)
	Always,

	/// Never compress (override detection)
	Never,
}

/// Smart compression detection using content sampling
///
/// This function extends should_compress_adaptive() with content-based detection
/// for improved accuracy. It follows BorgBackup's proven approach of sampling
/// file content to determine compressibility.
///
/// # Arguments
/// * `file_path` - Optional path to the file for content sampling
/// * `filename` - Filename for extension-based filtering
/// * `file_size` - Size in bytes
/// * `is_local` - Whether this is a local transfer
/// * `detection_mode` - Detection mode (Auto, Extension, Always, Never)
///
/// # Detection Strategy
/// 1. Fast path: Skip if local transfer, small file, or known compressed extension
/// 2. Content sampling: Read first 64KB, test with LZ4, measure ratio
/// 3. Decision: Ratio <0.9 → compress with Zstd, ≥0.9 → skip compression
pub fn should_compress_smart(file_path: Option<&Path>, filename: &str, file_size: u64, is_local: bool, detection_mode: CompressionDetection) -> Compression {
	// LOCAL: Never compress (disk I/O is bottleneck, not network/CPU)
	if is_local {
		return Compression::None;
	}

	// Handle explicit overrides
	match detection_mode {
		CompressionDetection::Always => return Compression::Zstd,
		CompressionDetection::Never => return Compression::None,
		_ => {} // Continue with detection
	}

	// Skip small files (overhead > benefit)
	if file_size < 1024 * 1024 {
		return Compression::None;
	}

	// Skip very large files (would load entire file into RAM)
	// Max 256MB for compression to avoid OOM on large files
	//
	// WHY THIS LIMIT:
	// - sy-remote receive-file protocol requires buffering entire compressed data
	// - Files >256MB use SFTP instead (already efficient, chunks internally)
	// - True streaming compression would require protocol redesign
	// - 256MB covers 99% of compressible files (logs, code, text files)
	//
	// NOTE: compress_streaming() exists for future use if protocol supports it
	const MAX_COMPRESSIBLE_SIZE: u64 = 256 * 1024 * 1024;
	if file_size > MAX_COMPRESSIBLE_SIZE {
		return Compression::None;
	}

	// Skip known compressed extensions (fast path)
	if is_compressed_extension(filename) {
		return Compression::None;
	}

	// Extension-only mode (legacy behavior)
	if detection_mode == CompressionDetection::Extension {
		return Compression::Zstd;
	}

	// Content sampling (auto mode)
	// This is the new smart detection that tests actual compressibility
	if let Some(path) = file_path {
		match detect_compressibility(path) {
			Ok(ratio) if ratio < 0.9 => {
				// Compressible: >10% savings achieved
				Compression::Zstd
			}
			Ok(_ratio) => {
				// Incompressible: <10% savings, not worth CPU overhead
				Compression::None
			}
			Err(_) => {
				// Error reading file, fall back to trying compression
				// Better to compress and waste some CPU than skip and lose bandwidth
				Compression::Zstd
			}
		}
	} else {
		// No file path available, fall back to extension-based heuristic
		// This happens when we only have filename/size but not actual file
		Compression::Zstd
	}
}

#[cfg(test)]
mod tests {
	use super::*;

	#[test]
	fn test_compress_decompress_lz4() {
		let original = b"Hello, world! This is a test of LZ4 compression. ".repeat(100);
		let compressed = compress(&original, Compression::Lz4).unwrap();
		let decompressed = decompress(&compressed, Compression::Lz4).unwrap();

		assert_eq!(original.as_slice(), decompressed.as_slice());
		assert!(compressed.len() < original.len());
	}

	#[test]
	fn test_compress_decompress_zstd() {
		let original = b"Hello, world! This is a test of Zstd compression. ".repeat(100);
		let compressed = compress(&original, Compression::Zstd).unwrap();
		let decompressed = decompress(&compressed, Compression::Zstd).unwrap();

		assert_eq!(original.as_slice(), decompressed.as_slice());
		assert!(compressed.len() < original.len());
	}

	#[test]
	fn test_compress_decompress_none() {
		let original = b"No compression test";
		let compressed = compress(original, Compression::None).unwrap();
		let decompressed = decompress(&compressed, Compression::None).unwrap();

		assert_eq!(original.as_slice(), decompressed.as_slice());
		assert_eq!(compressed.len(), original.len());
	}

	#[test]
	fn test_zstd_compression_ratio() {
		let repetitive = b"AAAA".repeat(1000);
		let compressed = compress(&repetitive, Compression::Zstd).unwrap();

		// Should compress very well (repetitive data)
		let ratio = compressed.len() as f64 / repetitive.len() as f64;
		assert!(ratio < 0.1); // Less than 10% of original
	}

	#[test]
	fn test_is_compressed_extension() {
		// Lowercase
		assert!(is_compressed_extension("file.jpg"));
		assert!(is_compressed_extension("video.mp4"));
		assert!(is_compressed_extension("archive.zip"));
		assert!(is_compressed_extension("document.pdf"));

		// Uppercase (should work case-insensitively)
		assert!(is_compressed_extension("file.JPG"));
		assert!(is_compressed_extension("video.MP4"));
		assert!(is_compressed_extension("archive.ZIP"));

		// Mixed case
		assert!(is_compressed_extension("file.JpG"));
		assert!(is_compressed_extension("video.Mp4"));

		// Not compressed
		assert!(!is_compressed_extension("file.txt"));
		assert!(!is_compressed_extension("code.rs"));
		assert!(!is_compressed_extension("data.csv"));
	}

	#[test]
	fn test_should_compress_small_file() {
		// Small files should not be compressed
		assert_eq!(should_compress("test.txt", 1024), Compression::None);
	}

	#[test]
	fn test_should_compress_already_compressed() {
		// Already compressed files should not be compressed
		assert_eq!(should_compress("image.jpg", 10_000_000), Compression::None);
		assert_eq!(should_compress("video.mp4", 100_000_000), Compression::None);
	}

	#[test]
	fn test_should_compress_large_text() {
		// Large text files should be compressed (now defaults to Zstd)
		assert_eq!(should_compress("data.txt", 10_000_000), Compression::Zstd);
		assert_eq!(should_compress("log.log", 50_000_000), Compression::Zstd);
	}

	#[test]
	fn test_roundtrip_empty_data() {
		let empty: &[u8] = &[];
		for compression in [Compression::None, Compression::Lz4, Compression::Zstd] {
			let compressed = compress(empty, compression).unwrap();
			let decompressed = decompress(&compressed, compression).unwrap();
			assert_eq!(decompressed.as_slice(), empty);
		}
	}

	#[test]
	fn test_roundtrip_large_data() {
		// 1MB of data
		let large: Vec<u8> = (0..1_000_000).map(|i| (i % 256) as u8).collect();

		for compression in [Compression::None, Compression::Lz4, Compression::Zstd] {
			let compressed = compress(&large, compression).unwrap();
			let decompressed = decompress(&compressed, compression).unwrap();
			assert_eq!(decompressed, large);
		}
	}

	#[test]
	fn test_lz4_compression_ratio() {
		let repetitive = b"AAAA".repeat(1000);
		let compressed = compress(&repetitive, Compression::Lz4).unwrap();

		// LZ4 should compress repetitive data well
		let ratio = compressed.len() as f64 / repetitive.len() as f64;
		assert!(ratio < 0.1); // Less than 10% of original
	}

	#[test]
	fn test_adaptive_compression_local() {
		// Local transfers should never compress
		assert_eq!(should_compress_adaptive("test.txt", 10_000_000, true, None), Compression::None);
	}

	#[test]
	fn test_adaptive_compression_any_network() {
		// UPDATED: If network is very fast (>500Mbps), skip compression to save CPU/latency overhead
		// Threshold is 500 Mbps

		// 100 Gbps (100_000 Mbps) -> None (too fast, don't compress)
		assert_eq!(should_compress_adaptive("test.txt", 10_000_000, false, Some(100_000)), Compression::None);

		// 1 Gbps network (1000 Mbps) -> None
		assert_eq!(should_compress_adaptive("test.txt", 10_000_000, false, Some(1000)), Compression::None);

		// 100 Mbps network -> Zstd (still benefits)
		assert_eq!(should_compress_adaptive("test.txt", 10_000_000, false, Some(100)), Compression::Zstd);

		// No network speed info -> Zstd (default for network transfers)
		assert_eq!(should_compress_adaptive("test.txt", 10_000_000, false, None), Compression::Zstd);
	}

	#[test]
	fn test_adaptive_compression_respects_precompressed() {
		// Even on slow network, don't compress already-compressed files
		assert_eq!(should_compress_adaptive("video.mp4", 100_000_000, false, Some(10)), Compression::None);
	}

	#[test]
	fn test_adaptive_compression_small_files() {
		// Small files should not be compressed regardless of network speed
		assert_eq!(should_compress_adaptive("test.txt", 512_000, false, Some(10)), Compression::None);
	}

	#[test]
	fn test_detect_compressibility_text() {
		use std::io::Write;
		use tempfile::NamedTempFile;

		// Create a highly compressible text file
		let mut temp_file = NamedTempFile::new().unwrap();
		let repetitive_text = "Hello world! ".repeat(5000); // ~60KB of repetitive text
		temp_file.write_all(repetitive_text.as_bytes()).unwrap();
		temp_file.flush().unwrap();

		let ratio = detect_compressibility(temp_file.path()).unwrap();

		// Repetitive text should compress very well (ratio should be < 0.5)
		assert!(ratio < 0.5, "Ratio: {}", ratio);
	}

	#[test]
	fn test_detect_compressibility_random() {
		use std::io::Write;
		use tempfile::NamedTempFile;

		// Create a file with high-entropy data (incompressible)
		// Use a better pseudo-random sequence that doesn't compress well
		let mut temp_file = NamedTempFile::new().unwrap();
		let random_data: Vec<u8> = (0u32..65536)
			.map(|i| {
				// Mix bits from multiple positions to create high entropy
				let x = i.wrapping_mul(2654435761); // Knuth's multiplicative hash
				((x ^ (x >> 16)) & 0xFF) as u8
			})
			.collect();
		temp_file.write_all(&random_data).unwrap();
		temp_file.flush().unwrap();

		let ratio = detect_compressibility(temp_file.path()).unwrap();

		// High-entropy data should not compress well (ratio should be > 0.85)
		// Note: Even good pseudo-random may compress slightly, so we use 0.85 threshold
		assert!(ratio > 0.85, "Ratio: {}", ratio);
	}

	#[test]
	fn test_detect_compressibility_empty() {
		use tempfile::NamedTempFile;

		// Create an empty file
		let temp_file = NamedTempFile::new().unwrap();

		let ratio = detect_compressibility(temp_file.path()).unwrap();

		// Empty file should return 1.0 (no benefit)
		assert_eq!(ratio, 1.0);
	}

	#[test]
	fn test_should_compress_smart_auto_compressible() {
		use std::io::Write;
		use tempfile::NamedTempFile;

		// Create a compressible file
		let mut temp_file = NamedTempFile::new().unwrap();
		let text = "Compressible text data! ".repeat(50000); // ~1.2MB
		temp_file.write_all(text.as_bytes()).unwrap();
		temp_file.flush().unwrap();

		let result = should_compress_smart(Some(temp_file.path()), "test.txt", 1_200_000, false, CompressionDetection::Auto);

		assert_eq!(result, Compression::Zstd);
	}

	#[test]
	fn test_should_compress_smart_auto_incompressible() {
		use std::io::Write;
		use tempfile::NamedTempFile;

		// Create an incompressible file (high entropy)
		let mut temp_file = NamedTempFile::new().unwrap();
		let random_data: Vec<u8> = (0u32..1_200_000)
			.map(|i| {
				// Use Knuth's multiplicative hash for high-entropy data
				let x = i.wrapping_mul(2654435761);
				((x ^ (x >> 16)) & 0xFF) as u8
			})
			.collect();
		temp_file.write_all(&random_data).unwrap();
		temp_file.flush().unwrap();

		let result = should_compress_smart(Some(temp_file.path()), "data.bin", 1_200_000, false, CompressionDetection::Auto);

		// Should skip compression for incompressible data
		assert_eq!(result, Compression::None);
	}

	#[test]
	fn test_should_compress_smart_always() {
		// Always mode should always compress
		let result = should_compress_smart(
			None,
			"test.jpg", // Even for compressed extension
			10_000_000,
			false,
			CompressionDetection::Always,
		);

		assert_eq!(result, Compression::Zstd);
	}

	#[test]
	fn test_should_compress_smart_never() {
		// Never mode should never compress
		let result = should_compress_smart(
			None,
			"test.txt", // Even for text files
			10_000_000,
			false,
			CompressionDetection::Never,
		);

		assert_eq!(result, Compression::None);
	}

	#[test]
	fn test_should_compress_smart_extension_mode() {
		// Extension mode should use legacy behavior (compress unless known extension)
		let result = should_compress_smart(None, "test.txt", 10_000_000, false, CompressionDetection::Extension);

		assert_eq!(result, Compression::Zstd);

		// Should skip known compressed extensions even in extension mode
		let result = should_compress_smart(None, "test.jpg", 10_000_000, false, CompressionDetection::Extension);

		assert_eq!(result, Compression::None);
	}

	#[test]
	fn test_should_compress_smart_local() {
		// Local transfers should never compress regardless of detection mode
		let result = should_compress_smart(
			None,
			"test.txt",
			10_000_000,
			true, // is_local
			CompressionDetection::Auto,
		);

		assert_eq!(result, Compression::None);
	}

	#[test]
	fn test_should_compress_smart_small_file() {
		// Small files should not be compressed
		let result = should_compress_smart(
			None,
			"test.txt",
			512_000, // < 1MB
			false,
			CompressionDetection::Auto,
		);

		assert_eq!(result, Compression::None);
	}

	#[test]
	fn test_should_compress_smart_known_compressed_extension() {
		// Files with known compressed extensions should be skipped (fast path)
		let result = should_compress_smart(None, "video.mp4", 100_000_000, false, CompressionDetection::Auto);

		assert_eq!(result, Compression::None);
	}

	#[test]
	fn test_should_compress_smart_no_path_fallback() {
		// Without file path, should fall back to extension-based heuristic
		let result = should_compress_smart(
			None, // No path
			"data.bin",
			10_000_000,
			false,
			CompressionDetection::Auto,
		);

		// Should default to compressing when path not available
		assert_eq!(result, Compression::Zstd);
	}
}