check: refactored to add check-mode (info, audio). Info mode compares format structure

symphonia-check/Cargo.toml

@@ -11,7 +11,9 @@ rust-version = "1.77"
 publish = false
 
 [dependencies]
-clap = "3.1.0"
+clap = { version = "4.5.23", features = ["derive"] }
 log = { version = "0.4", features = ["release_max_level_info"] }
 pretty_env_logger = "0.4"
-symphonia = { version = "0.5.4", path = "../symphonia", features = ["all", "opt-simd"] }
+symphonia = { version = "0.5.4", path = "../symphonia", features = ["all", "opt-simd"] }
+serde = { version = "1.0", features = ["derive"] }
+serde_json = "1.0"

symphonia-check/src/audio.rs

@@ -0,0 +1,365 @@
+// Symphonia Check Tool
+// Copyright (c) 2019-2022 The Project Symphonia Developers.
+//
+// This Source Code Form is subject to the terms of the Mozilla Public
+// License, v. 2.0. If a copy of the MPL was not distributed with this
+// file, You can obtain one at https://mozilla.org/MPL/2.0/.
+
+#![warn(rust_2018_idioms)]
+#![forbid(unsafe_code)]
+// Justification: Fields on DecoderOptions and FormatOptions may change at any time, but
+// symphonia-check doesn't want to be updated every time those fields change, therefore always fill
+// in the remaining fields with default values.
+#![allow(clippy::needless_update)]
+
+use std::fs::File;
+use std::path::Path;
+use std::process::{Command, Stdio};
+
+use log::warn;
+use symphonia::core::audio::GenericAudioBufferRef;
+use symphonia::core::codecs::audio::{AudioDecoder, AudioDecoderOptions};
+use symphonia::core::codecs::CodecParameters;
+use symphonia::core::errors::{unsupported_error, Error, Result};
+use symphonia::core::formats::probe::Hint;
+use symphonia::core::formats::{FormatOptions, FormatReader, TrackType};
+use symphonia::core::io::{MediaSourceStream, ReadOnlySource};
+use symphonia::core::meta::MetadataOptions;
+
+use crate::{AudioTestDecoder, AudioTestOptions, RefProcess};
+
+/// The absolute maximum allowable sample delta. Around 2^-17 (-102.4dB).
+const ABS_MAX_ALLOWABLE_SAMPLE_DELTA: f32 = 0.00001;
+
+// The absolute maximum allowable sample delta for a fully compliant MP3 decoder as specified by the
+// ISO. Around 2^-14 (-84.2dB).
+// const ABS_MAX_ALLOWABLE_SAMPLE_DELTA_MP3: f32 = 0.00006104;
+
+#[derive(Default)]
+struct AudioTestResult {
+    n_frames: u64,
+    n_samples: u64,
+    n_failed_samples: u64,
+    n_packets: u64,
+    n_failed_packets: u64,
+    abs_max_delta: f32,
+    tgt_unchecked_samples: u64,
+    ref_unchecked_samples: u64,
+}
+
+fn build_ffmpeg_command(path: &str, gapless: bool) -> Command {
+    let mut cmd = Command::new("ffmpeg");
+
+    // Gapless argument must come before everything else.
+    if !gapless {
+        cmd.arg("-flags2").arg("skip_manual");
+    }
+
+    cmd.arg("-nostats") // Quiet command.
+        .arg("-hide_banner")
+        .arg("-i") // Input path.
+        .arg(path)
+        .arg("-map") // Select the first audio track.
+        .arg("0:a:0")
+        .arg("-c:a") // Encode audio to pcm_s32le.
+        .arg("pcm_f32le")
+        .arg("-f") // Output in WAVE format.
+        .arg("wav")
+        .arg("-") // Pipe output to stdout.
+        .stdout(Stdio::piped())
+        .stderr(Stdio::null()); // Pipe errors to null.
+
+    cmd
+}
+
+fn build_flac_command(path: &str) -> Command {
+    let mut cmd = Command::new("flac");
+
+    cmd.arg("--stdout").arg("-d").arg(path).stdout(Stdio::piped()).stderr(Stdio::null());
+
+    cmd
+}
+
+fn build_mpg123_command(path: &str, gapless: bool) -> Command {
+    let mut cmd = Command::new("mpg123");
+
+    if !gapless {
+        cmd.arg("--no-gapless");
+    }
+
+    cmd.arg("--wav").arg("-").arg("--float").arg(path).stdout(Stdio::piped()).stderr(Stdio::null());
+
+    cmd
+}
+
+fn build_oggdec_command(path: &str) -> Command {
+    let mut cmd = Command::new("oggdec");
+    cmd.arg(path).arg("-o").arg("-").stdout(Stdio::piped()).stderr(Stdio::null());
+    cmd
+}
+
+#[derive(Default)]
+struct FlushStats {
+    n_packets: u64,
+    n_samples: u64,
+}
+
+struct DecoderInstance {
+    format: Box<dyn FormatReader>,
+    decoder: Box<dyn AudioDecoder>,
+    track_id: u32,
+}
+
+impl DecoderInstance {
+    fn try_open(
+        mss: MediaSourceStream<'static>,
+        fmt_opts: FormatOptions,
+    ) -> Result<DecoderInstance> {
+        // Use the default options for metadata and format readers, and the decoder.
+        let meta_opts: MetadataOptions = Default::default();
+        let dec_opts: AudioDecoderOptions = Default::default();
+
+        let hint = Hint::new();
+
+        let format = symphonia::default::get_probe().probe(&hint, mss, fmt_opts, meta_opts)?;
+
+        let track = format.default_track(TrackType::Audio).unwrap();
+
+        let codec_params = match &track.codec_params {
+            Some(CodecParameters::Audio(params)) => params,
+            _ => return unsupported_error("only audio tracks are supported"),
+        };
+
+        let decoder =
+            symphonia::default::get_codecs().make_audio_decoder(codec_params, &dec_opts)?;
+
+        let track_id = track.id;
+
+        Ok(DecoderInstance { format, decoder, track_id })
+    }
+
+    fn samples_per_frame(&self) -> Option<u64> {
+        self.decoder.codec_params().channels.as_ref().map(|ch| ch.count() as u64)
+    }
+
+    fn next_audio_buf(&mut self, keep_going: bool) -> Result<Option<GenericAudioBufferRef<'_>>> {
+        loop {
+            // Get the next packet.
+            let packet = match self.format.next_packet() {
+                Ok(Some(packet)) => packet,
+                Ok(None) => return Ok(None),
+                Err(Error::IoError(err)) if err.kind() == std::io::ErrorKind::UnexpectedEof => {
+                    // WavReader will always return an UnexpectedEof when it ends because the
+                    // reference decoder is piping the decoded audio and cannot write out the
+                    // actual length of the media. Treat UnexpectedEof as the end of the stream.
+                    return Ok(None);
+                }
+                Err(err) => return Err(err),
+            };
+
+            // Skip packets that do not belong to the track being decoded.
+            if packet.track_id() != self.track_id {
+                continue;
+            }
+
+            // Decode the packet, ignoring decode errors if `keep_going` is true.
+            match self.decoder.decode(&packet) {
+                Ok(_) => break,
+                Err(Error::DecodeError(err)) if keep_going => warn!("{}", err),
+                Err(err) => return Err(err),
+            }
+        }
+
+        Ok(Some(self.decoder.last_decoded()))
+    }
+
+    fn flush(&mut self, keep_going: bool) -> Result<FlushStats> {
+        let mut stats: FlushStats = Default::default();
+
+        while let Some(buf) = self.next_audio_buf(keep_going)? {
+            stats.n_packets += 1;
+            stats.n_samples += buf.samples_interleaved() as u64;
+        }
+
+        Ok(stats)
+    }
+}
+
+fn run_check(
+    ref_inst: &mut DecoderInstance,
+    tgt_inst: &mut DecoderInstance,
+    opts: &AudioTestOptions,
+    acct: &mut AudioTestResult,
+) -> Result<()> {
+    // Reference
+    let mut ref_sample_buf: Vec<f32> = Default::default();
+    let mut ref_sample_cnt = 0;
+    let mut ref_sample_pos = 0;
+
+    // Target
+    let mut tgt_sample_buf: Vec<f32> = Default::default();
+    let mut tgt_sample_cnt = 0;
+    let mut tgt_sample_pos = 0;
+
+    let samples_per_frame = tgt_inst.samples_per_frame().unwrap_or(1);
+
+    // Samples/frame must match for both decoders.
+    if samples_per_frame != ref_inst.samples_per_frame().unwrap_or(1) {
+        return unsupported_error("target and reference decoder samples per frame mismatch");
+    }
+
+    let early_fail = 'outer: loop {
+        // Decode the next target audio buffer and copy it to the target sample buffer.
+        match tgt_inst.next_audio_buf(opts.keep_going)? {
+            Some(buf) => buf.copy_to_vec_interleaved(&mut tgt_sample_buf),
+            None => break 'outer false,
+        };
+
+        tgt_sample_cnt = tgt_sample_buf.len();
+        tgt_sample_pos = 0;
+
+        // The number of frames previously read & compared.
+        let frame_num_base = acct.n_frames;
+
+        // The number of failed samples in the target packet.
+        let mut n_failed_pkt_samples = 0;
+
+        while tgt_sample_pos < tgt_sample_cnt {
+            // Need to read a decode a new reference buffer.
+            if ref_sample_pos == ref_sample_cnt {
+                // Get the next reference audio buffer and copy it to the reference sample buffer.
+                match ref_inst.next_audio_buf(true)? {
+                    Some(buf) => buf.copy_to_vec_interleaved(&mut ref_sample_buf),
+                    None => break 'outer false,
+                }
+
+                ref_sample_cnt = ref_sample_buf.len();
+                ref_sample_pos = 0;
+            }
+
+            // Get a slice of the remaining samples in the reference and target sample buffers.
+            let ref_samples = &ref_sample_buf[ref_sample_pos..];
+            let tgt_samples = &tgt_sample_buf[tgt_sample_pos..];
+
+            // The number of samples that can be compared given the current length of the reference
+            // and target sample buffers.
+            let n_test_samples = std::cmp::min(ref_samples.len(), tgt_samples.len());
+
+            // Perform the comparison.
+            for (&t, &r) in tgt_samples[..n_test_samples].iter().zip(&ref_samples[..n_test_samples])
+            {
+                // Clamp the reference and target samples between [-1.0, 1.0] and find the
+                // difference.
+                let delta = t.clamp(-1.0, 1.0) - r.clamp(-1.0, 1.0);
+
+                if delta.abs() > ABS_MAX_ALLOWABLE_SAMPLE_DELTA {
+                    // Print per-sample or per-packet failure nessage based on selected options.
+                    if !opts.is_quiet && (opts.is_per_sample || n_failed_pkt_samples == 0) {
+                        println!(
+                            "[FAIL] packet={:>8}, frame={:>10} ({:>4}), plane={:>3}, dec={:+.8}, ref={:+.8} ({:+.8})",
+                            acct.n_packets,
+                            acct.n_frames,
+                            acct.n_frames - frame_num_base,
+                            acct.n_samples % samples_per_frame,
+                            t,
+                            r,
+                            r - t
+                        );
+                    }
+
+                    n_failed_pkt_samples += 1;
+                }
+
+                acct.abs_max_delta = acct.abs_max_delta.max(delta.abs());
+                acct.n_samples += 1;
+
+                if acct.n_samples % samples_per_frame == 0 {
+                    acct.n_frames += 1;
+                }
+            }
+
+            // Update position in reference and target buffers.
+            ref_sample_pos += n_test_samples;
+            tgt_sample_pos += n_test_samples;
+        }
+
+        acct.n_failed_samples += n_failed_pkt_samples;
+        acct.n_failed_packets += u64::from(n_failed_pkt_samples > 0);
+        acct.n_packets += 1;
+
+        if opts.stop_after_fail && acct.n_failed_packets > 0 {
+            break true;
+        }
+    };
+
+    // Count how many samples were remaining for both the target and references if the loop did not
+    // break out early due to a failed sample.
+    if !early_fail {
+        let tgt_stats = tgt_inst.flush(true)?;
+        let ref_stats = ref_inst.flush(true)?;
+
+        acct.n_packets += tgt_stats.n_packets;
+        acct.tgt_unchecked_samples = (tgt_sample_cnt - tgt_sample_pos) as u64 + tgt_stats.n_samples;
+        acct.ref_unchecked_samples = (ref_sample_cnt - ref_sample_pos) as u64 + ref_stats.n_samples;
+    }
+
+    Ok(())
+}
+
+fn run_test(path: &str, opts: &AudioTestOptions, result: &mut AudioTestResult) -> Result<()> {
+    let command = match opts.ref_decoder {
+        AudioTestDecoder::Ffmpeg => build_ffmpeg_command(path, opts.gapless),
+        AudioTestDecoder::Flac => build_flac_command(path),
+        AudioTestDecoder::Mpg123 => build_mpg123_command(path, opts.gapless),
+        AudioTestDecoder::Oggdec => build_oggdec_command(path),
+    };
+
+    // 1. Start the reference decoder process.
+    let mut ref_process = RefProcess::try_spawn(command)?;
+
+    // 2. Instantiate a Symphonia decoder for the reference process output.
+    let ref_ms = Box::new(ReadOnlySource::new(ref_process.child.stdout.take().unwrap()));
+    let ref_mss = MediaSourceStream::new(ref_ms, Default::default());
+
+    let mut ref_inst = DecoderInstance::try_open(ref_mss, Default::default())?;
+
+    // 3. Instantiate a Symphonia decoder for the test target.
+    let tgt_ms = Box::new(File::open(Path::new(path))?);
+    let tgt_mss = MediaSourceStream::new(tgt_ms, Default::default());
+
+    let tgt_fmt_opts = FormatOptions { enable_gapless: opts.gapless, ..Default::default() };
+
+    let mut tgt_inst = DecoderInstance::try_open(tgt_mss, tgt_fmt_opts)?;
+
+    // 4. Begin check.
+    run_check(&mut ref_inst, &mut tgt_inst, opts, result)
+}
+
+pub fn run_audio(opts: AudioTestOptions) -> Result<()> {
+    let mut res: AudioTestResult = Default::default();
+
+    run_test(&opts.input, &opts, &mut res)?;
+
+    if !opts.is_quiet {
+        println!();
+    }
+
+    println!("Test Results");
+    println!("=================================================");
+    println!();
+    println!("  Failed/Total Packets: {:>12}/{:>12}", res.n_failed_packets, res.n_packets);
+    println!("  Failed/Total Samples: {:>12}/{:>12}", res.n_failed_samples, res.n_samples);
+    println!();
+    println!("  Remaining Target Samples:          {:>12}", res.tgt_unchecked_samples);
+    println!("  Remaining Reference Samples:       {:>12}", res.ref_unchecked_samples);
+    println!();
+    println!("  Absolute Maximum Sample Delta:       {:.8}", res.abs_max_delta);
+    println!();
+
+    if res.n_failed_samples == 0 {
+        Ok(())
+    }
+    else {
+        unsupported_error("Some samples didn't pass validation")
+    }
+}