Opus tag reading very slow #588
Description
Reproducer
Running this benchmark on a directory of tagged files:
use clap::Parser;
use lofty::probe::Probe;
use lofty::tag::{ItemValue, TagExt};
use lofty::file::TaggedFileExt;
use std::path::PathBuf;
use std::time::Instant;
use walkdir::WalkDir;
use crc32fast::Hasher;
#[derive(Parser)]
#[command(name = "lofty-benchmark")]
struct Cli {
/// Directory to scan for audio files
path: PathBuf,
/// Print each tag as it is read
#[arg(short, long)]
debug: bool,
}
fn main() {
let args = Cli::parse();
let start = Instant::now();
let mut hasher = Hasher::new();
let mut file_count = 0;
// Collect all files
let mut files: Vec<_> = WalkDir::new(&args.path)
.into_iter()
.filter_map(|e| e.ok())
.filter(|e| e.file_type().is_file())
.collect();
// Sort alphabetically to ensure reproducible order
files.sort_by(|a, b| a.path().cmp(b.path()));
for entry in files {
// Attempt to read tags from the file
// Probe::open guesses the format, read() parses it.
// We ignore errors as the directory may contain non-audio files.
if let Ok(results) = Probe::open(entry.path())
.and_then(|p| p.read())
{
file_count += 1;
// Iterate over all tags found in the file (e.g. ID3v2, Vorbis, etc.)
for tag in results.tags() {
if args.debug {
println!("File: {:?}, Tag Type: {:?}, Items: {}", entry.path(), tag.tag_type(), tag.len());
}
// Collect and sort items by key for reproducible order
let mut items: Vec<_> = tag.items().collect();
items.sort_by(|a, b| format!("{:?}", a.key()).cmp(&format!("{:?}", b.key())));
for item in items {
if args.debug {
println!("\t{:?}: {:?}", item.key(), item.value());
}
// Update checksum with key
hasher.update(format!("{:?}", item.key()).as_bytes());
// Update checksum with value
match item.value() {
ItemValue::Text(s) => hasher.update(s.as_bytes()),
ItemValue::Locator(s) => hasher.update(s.as_bytes()),
ItemValue::Binary(b) => hasher.update(b),
}
}
}
}
}
let duration = start.elapsed();
println!("Scanned {} files in {:.3?}", file_count, duration);
println!("Checksum: {:08x}", hasher.finalize());
}Summary
➜ lofty-benchmark git:(master) cargo run ~/Desktop/music/opus/autechre/
Compiling lofty v0.22.4 (/Users/miles/c/lofty-rs/lofty)
Compiling lofty-benchmark v0.1.0 (/Users/miles/c/lofty-benchmark)
Finished `dev` profile [unoptimized + debuginfo] target(s) in 1.65s
Running `target/debug/lofty-benchmark /Users/miles/Desktop/music/opus/autechre/`
Scanned 262 files in 6.362s
Checksum: d8b76029➜ lofty-benchmark git:(master) cargo run ~/Desktop/music/aac/autechre/
Compiling lofty v0.22.4 (/Users/miles/c/lofty-rs/lofty)
Compiling lofty-benchmark v0.1.0 (/Users/miles/c/lofty-benchmark)
Finished `dev` profile [unoptimized + debuginfo] target(s) in 1.53s
Running `target/debug/lofty-benchmark /Users/miles/Desktop/music/aac/autechre/`
Scanned 262 files in 150.477ms
Checksum: 6f172600I also noticed TagLib is massively faster than lofty for this case.
I have an experimental branch that changes this code to read backwards from the end of the file that seems to bring opus speed in line with AAC. If you're interested I could put up a PR or just paste it here.
Lines 43 to 57 in 8f667a2
Expected behavior
Opus tag reading shouldn't be this much slower than m4a.
Assets
No response