Merge pull request #831 from RustAudio/fix/span-fixes-clean

fix: correct Source trait semantics and span tracking bugs

Author: yara-blue

CHANGELOG.md

@@ -12,21 +12,39 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ### Added
 
 - Added `Skippable::skipped` function to check if the inner source was skipped.
+- All sources now implement `ExactSizeIterator` when their inner source does.
+- All sources now implement `Iterator::size_hint()`.
+- `Chirp` now implements `try_seek`.
 
 ### Changed
 
 - Breaking: `Done` now calls a callback instead of decrementing an `Arc<AtomicUsize>`.
 - Updated `cpal` to v0.18.
+- Clarified `Source::current_span_len()` documentation to specify it returns total span length.
+- Explicitly document the requirement for sources to return complete frames.
+- Ensured decoders to always return complete frames, as well as `TakeDuration` when expired.
+- Breaking: `Zero::new_samples()` now returns `Result<Self, ZeroError>` requiring a frame-aligned number of samples.
+- Improved queue, buffer, mixer and sample rate conversion performance.
 
 ### Fixed
 
 - Fixed `Player::skip_one` not decreasing the player's length immediately.
+- Fixed `Source::current_span_len()` to consistently return total span length.
+- Fixed `Source::size_hint()` to consistently report actual bounds based on current sources.
+- Fixed `Pausable::size_hint()` to correctly account for paused samples.
+- Fixed `MixerSource` and `LinearRamp` to prevent overflow with very long playback.
+- Fixed `PeriodicAccess` to prevent overflow with very long periods.
+- Fixed `BltFilter` to work correctly with stereo and multi-channel audio.
+- Fixed `ChannelVolume` to work correclty with stereo and multi-channel audio.
+- Fixed `Brownian` and `Red` noise generators to reset after seeking.
+- Fixed sources to correctly handle sample rate and channel count changes at span boundaries.
+- Fixed sources to detect parameter updates after mid-span seeks.
 
 ## Version [0.22.2] (2026-02-22)
 
 ### Fixed
 
-- Incorrectly set system default audio buffer size breaks playback. We no longer use the system default (introduced in 0.22 through cpal upgrade) and instead set a safe buffer duration. 
+- Incorrectly set system default audio buffer size breaks playback. We no longer use the system default (introduced in 0.22 through cpal upgrade) and instead set a safe buffer duration.
 - Audio output fallback picked null device leading to no output.
 - Mixer did not actually add sources sometimes.
 
@@ -38,6 +56,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 ## Version [0.22.1] (2026-02-22)
 
 ### Fixed
+
 - docs.rs could not build the documentation.
 
 ## Version [0.22] (2026-02-22)
@@ -65,19 +84,19 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 - Added `SampleRateConverter::inner` to get underlying iterator by ref.
 
 ### Fixed
+
 - docs.rs will now document all features, including those that are optional.
 - `Chirp::next` now returns `None` when the total duration has been reached, and will work
   correctly for a number of samples greater than 2^24.
 - `PeriodicAccess` is slightly more accurate for 44.1 kHz sample rate families.
 - Fixed audio distortion when queueing sources with different sample rates/channel counts or transitioning from empty queue.
 - Fixed `SamplesBuffer` to correctly report exhaustion and remaining samples.
 - Improved precision in `SkipDuration` to avoid off-by-a-few-samples errors.
-- Fixed channel misalignment in queue with non-power-of-2 channel counts (e.g., 6 channels) by ensuring frame-aligned span lengths.
-- Fixed channel misalignment when sources end before their promised span length by padding with silence to complete frames.
 - Fixed `Empty` source to properly report exhaustion.
 - Fixed `Zero::current_span_len` returning remaining samples instead of span length.
 
 ### Changed
+
 - Breaking: _Sink_ terms are replaced with _Player_ and _Stream_ terms replaced
   with _Sink_. This is a simple rename, functionality is identical.
     - `OutputStream` is now `MixerDeviceSink` (in anticipation of future

UPGRADE.md

@@ -10,6 +10,9 @@ changes and new features, see [CHANGELOG.md](CHANGELOG.md).
   - To retain old behavior replace the `Arc<AtomicUsize>` argument in `Done::new` with
     `move |_| { number.fetch_sub(1, std::sync::atomic::Ordering::Relaxed) }`.
   - `Done` has now two generics instead of one: `<I: Source, F: FnMut(&mut I)>`.
+- `Zero::new_samples()` now returns `Result<Zero, ZeroError>` instead of `Zero`.
+  Previously the function accepted any `num_samples` value; passing one that is not a
+  multiple of `channels` now returns an `Err` instead of producing a mis-aligned source.
 
 # rodio 0.21.1 to 0.22
 - _Sink_ terms are replaced with _Player_ and _Stream_ terms replaced

src/decoder/flac.rs

@@ -2,7 +2,7 @@ use std::io::{Read, Seek, SeekFrom};
 use std::mem;
 use std::time::Duration;
 
-use crate::source::SeekError;
+use crate::source::{padding_samples_needed, SeekError};
 use crate::Source;
 
 use crate::common::{ChannelCount, Sample, SampleRate};
@@ -24,6 +24,8 @@ where
     sample_rate: SampleRate,
     channels: ChannelCount,
     total_duration: Option<Duration>,
+    samples_in_current_frame: usize,
+    silence_samples_remaining: usize,
 }
 
 impl<R> FlacDecoder<R>
@@ -69,6 +71,8 @@ where
             )
             .expect("flac should never have zero channels"),
             total_duration,
+            samples_in_current_frame: 0,
+            silence_samples_remaining: 0,
         })
     }
 
@@ -119,6 +123,12 @@ where
     #[inline]
     fn next(&mut self) -> Option<Self::Item> {
         loop {
+            // If padding to complete a frame, return silence
+            if self.silence_samples_remaining > 0 {
+                self.silence_samples_remaining -= 1;
+                return Some(Sample::EQUILIBRIUM);
+            }
+
             if self.current_block_off < self.current_block.len() {
                 // Read from current block.
                 let real_offset = (self.current_block_off % self.channels.get() as usize)
@@ -142,6 +152,8 @@ where
                         (raw_val << (32 - bits)).to_sample()
                     }
                 };
+                self.samples_in_current_frame =
+                    (self.samples_in_current_frame + 1) % self.channels.get() as usize;
                 return Some(real_val);
             }
 
@@ -153,7 +165,16 @@ where
                     self.current_block_channel_len = (block.len() / block.channels()) as usize;
                     self.current_block = block.into_buffer();
                 }
-                _ => return None,
+                _ => {
+                    // Input exhausted - check if mid-frame
+                    self.silence_samples_remaining =
+                        padding_samples_needed(self.samples_in_current_frame, self.channels);
+                    if self.silence_samples_remaining > 0 {
+                        self.samples_in_current_frame = 0;
+                        continue; // Loop will inject silence
+                    }
+                    return None;
+                }
             }
         }
     }

src/decoder/mp3.rs

@@ -3,7 +3,7 @@ use std::num::NonZero;
 use std::time::Duration;
 
 use crate::common::{ChannelCount, Sample, SampleRate};
-use crate::source::SeekError;
+use crate::source::{padding_samples_needed, SeekError};
 use crate::Source;
 
 use dasp_sample::Sample as _;
@@ -21,6 +21,8 @@ where
     // what minimp3 calls frames rodio calls spans
     current_span: Frame,
     current_span_offset: usize,
+    samples_in_current_frame: usize,
+    silence_samples_remaining: usize,
 }
 
 impl<R> Mp3Decoder<R>
@@ -43,6 +45,8 @@ where
             decoder,
             current_span,
             current_span_offset: 0,
+            samples_in_current_frame: 0,
+            silence_samples_remaining: 0,
         })
     }
 
@@ -98,21 +102,40 @@ where
     type Item = Sample;
 
     fn next(&mut self) -> Option<Self::Item> {
-        let current_span_len = self.current_span_len()?;
-        if self.current_span_offset == current_span_len {
-            if let Ok(span) = self.decoder.next_frame() {
-                // if let Ok(span) = self.decoder.decode_frame() {
-                self.current_span = span;
-                self.current_span_offset = 0;
-            } else {
-                return None;
+        loop {
+            // If padding to complete a frame, return silence
+            if self.silence_samples_remaining > 0 {
+                self.silence_samples_remaining -= 1;
+                return Some(Sample::EQUILIBRIUM);
+            }
+
+            let current_span_len = self.current_span_len()?;
+            if self.current_span_offset == current_span_len {
+                if let Ok(span) = self.decoder.next_frame() {
+                    self.current_span = span;
+                    self.current_span_offset = 0;
+                } else {
+                    // Input exhausted - check if mid-frame
+                    let channels = self.channels();
+                    self.silence_samples_remaining =
+                        padding_samples_needed(self.samples_in_current_frame, channels);
+                    if self.silence_samples_remaining > 0 {
+                        self.samples_in_current_frame = 0;
+                        continue; // Loop will inject silence
+                    }
+                    return None;
+                }
             }
-        }
 
-        let v = self.current_span.data[self.current_span_offset];
-        self.current_span_offset += 1;
+            let v = self.current_span.data[self.current_span_offset];
+            self.current_span_offset += 1;
 
-        Some(v.to_sample())
+            let channels = self.channels();
+            self.samples_in_current_frame =
+                (self.samples_in_current_frame + 1) % channels.get() as usize;
+
+            return Some(v.to_sample());
+        }
     }
 }
 

src/decoder/symphonia.rs

@@ -20,8 +20,10 @@ use symphonia::{
 use super::{DecoderError, Settings};
 use crate::{
     common::{assert_error_traits, ChannelCount, Sample, SampleRate},
-    source, Source,
+    source::{self, padding_samples_needed},
+    Source,
 };
+use dasp_sample::Sample as _;
 
 #[derive(Clone)]
 pub(crate) struct Registry(Arc<RwLock<CodecRegistry>>);
@@ -55,6 +57,8 @@ pub(crate) struct SymphoniaDecoder {
     spec: SignalSpec,
     seek_mode: SeekMode,
     selected_track_id: u32,
+    samples_in_current_frame: usize,
+    silence_samples_remaining: usize,
 }
 
 impl SymphoniaDecoder {
@@ -176,6 +180,8 @@ impl SymphoniaDecoder {
             spec,
             seek_mode,
             selected_track_id: track_id,
+            samples_in_current_frame: 0,
+            silence_samples_remaining: 0,
         }))
     }
 
@@ -328,44 +334,87 @@ impl Iterator for SymphoniaDecoder {
     type Item = Sample;
 
     fn next(&mut self) -> Option<Self::Item> {
-        if self.current_span_offset >= self.buffer.len() {
-            let decoded = loop {
-                let packet = self.format.next_packet().ok()?;
-
-                // If the packet does not belong to the selected track, skip over it
-                if packet.track_id() != self.selected_track_id {
-                    continue;
-                }
+        loop {
+            // If padding to complete a frame, return silence
+            if self.silence_samples_remaining > 0 {
+                self.silence_samples_remaining -= 1;
+                return Some(Sample::EQUILIBRIUM);
+            }
 
-                let decoded = match self.decoder.decode(&packet) {
-                    Ok(decoded) => decoded,
-                    Err(Error::DecodeError(_)) => {
-                        // Skip over packets that cannot be decoded. This ensures the iterator
-                        // continues processing subsequent packets instead of terminating due to
-                        // non-critical decode errors.
-                        continue;
+            if self.current_span_offset >= self.buffer.len() {
+                let decoded = loop {
+                    let packet = match self.format.next_packet() {
+                        Ok(packet) => {
+                            if packet.track_id() == self.selected_track_id {
+                                packet
+                            } else {
+                                continue;
+                            }
+                        }
+                        Err(_) => {
+                            // Input exhausted - check if mid-frame
+                            let channels = self.channels();
+                            self.silence_samples_remaining =
+                                padding_samples_needed(self.samples_in_current_frame, channels);
+                            if self.silence_samples_remaining > 0 {
+                                self.samples_in_current_frame = 0;
+                                break None;
+                            }
+                            return None;
+                        }
+                    };
+                    let decoded = match self.decoder.decode(&packet) {
+                        Ok(decoded) => decoded,
+                        Err(Error::DecodeError(_)) => {
+                            // Skip over packets that cannot be decoded. This ensures the iterator
+                            // continues processing subsequent packets instead of terminating due to
+                            // non-critical decode errors.
+                            continue;
+                        }
+                        Err(_) => {
+                            // Input exhausted - check if mid-frame
+                            let channels = self.channels();
+                            self.silence_samples_remaining =
+                                padding_samples_needed(self.samples_in_current_frame, channels);
+                            if self.silence_samples_remaining > 0 {
+                                self.samples_in_current_frame = 0;
+                                break None;
+                            }
+                            return None;
+                        }
+                    };
+
+                    // Loop until we get a packet with audio frames. This is necessary because some
+                    // formats can have packets with only metadata, particularly when rewinding, in
+                    // which case the iterator would otherwise end with `None`.
+                    // Note: checking `decoded.frames()` is more reliable than `packet.dur()`, which
+                    // can resturn non-zero durations for packets without audio frames.
+                    if decoded.frames() > 0 {
+                        break Some(decoded);
                     }
-                    Err(_) => return None,
                 };
 
-                // Loop until we get a packet with audio frames. This is necessary because some
-                // formats can have packets with only metadata, particularly when rewinding, in
-                // which case the iterator would otherwise end with `None`.
-                // Note: checking `decoded.frames()` is more reliable than `packet.dur()`, which
-                // can resturn non-zero durations for packets without audio frames.
-                if decoded.frames() > 0 {
-                    break decoded;
+                match decoded {
+                    Some(decoded) => {
+                        decoded.spec().clone_into(&mut self.spec);
+                        self.buffer = SymphoniaDecoder::get_buffer(decoded, &self.spec);
+                        self.current_span_offset = 0;
+                    }
+                    None => {
+                        // Break out happened due to exhaustion, continue to emit padding
+                        continue;
+                    }
                 }
-            };
+            }
 
-            decoded.spec().clone_into(&mut self.spec);
-            self.buffer = SymphoniaDecoder::get_buffer(decoded, &self.spec);
-            self.current_span_offset = 0;
-        }
+            let sample = *self.buffer.samples().get(self.current_span_offset)?;
+            self.current_span_offset += 1;
 
-        let sample = *self.buffer.samples().get(self.current_span_offset)?;
-        self.current_span_offset += 1;
+            let channels = self.channels();
+            self.samples_in_current_frame =
+                (self.samples_in_current_frame + 1) % channels.get() as usize;
 
-        Some(sample)
+            return Some(sample);
+        }
     }
 }