Player still reads after Pause

[codesoap]

Description

There seems to be a race condition, where the Player can still read from the underlying io.Reader after the Player had been paused. With this code, I could provoke the problem reliably:

package main

import (
	"github.com/ebitengine/oto/v3"
	"io"
	"os"
	"time"
)

type slowReader struct{ r io.Reader }

func (sr slowReader) Read(p []byte) (int, error) {
	time.Sleep(100 * time.Millisecond)
	return sr.r.Read(p)
}

func main() {
	file, err := os.Open("./my-file.raw-s16le")
	if err != nil {
		panic("opening my-file.raw-s16le failed: " + err.Error())
	}

	op := &oto.NewContextOptions{
		SampleRate:   44100,
		ChannelCount: 2,
		Format:       oto.FormatSignedInt16LE,
	}
	otoCtx, readyChan, err := oto.NewContext(op)
	if err != nil {
		panic("oto.NewContext failed: " + err.Error())
	}
	<-readyChan

	reader := slowReader{r: file}
	player := otoCtx.NewPlayer(reader)
	player.Play()
	player.Pause()
	file.Close()
	time.Sleep(200 * time.Millisecond)
	if player.Err() != nil {
		println(player.Err().Error())
	}
}

Expectation

I would expect the Player not to read from its underlying io.Reader once Pause returned. I expected, that Pause even waits until running Read calls have finished before returning.

Background

This is a real problem in my application: I'm writing a music player and when switching songs, I'll stop the old player and close the music file, before creating a new player to play the newly selected song. Sometimes the old player then tries to read from the closed music file, which leads to errors.

Since I'm handling the errors within a custom reader, I'll see "false problems" at a point in my code, where I don't know that the read doesn't have to succeed anymore.

[hajimehoshi]

Even if a player pauses, it can read the stream to its buffer. So this behavior is expected.

Sometimes the old player then tries to read from the closed music file, which leads to errors.

Before closing the file, you should close the player rather than pausing.

[codesoap]

Huh, is this an undocumented behavior of the Close method? The documentation says it does nothing:

Close does nothing and always returns nil.

If Close ensures that no more reads occur, it would work for me. I just didn't see a way to "end" a player when reading the documentation.

Edit: I've tried to use player.Close() with the demo program, but it didn't change anything. Read was still in progress, after Close returned.

[hajimehoshi]

Oops, that's true. Sorry but I was misunderstanding.

So, there is no proper way to close a source before its player dies. I'm a little surprised that there has been no report until now.

Let me have time to think. I might revive Close for this purpose, or might change Pause behavior, or might take another solution.

[hajimehoshi]

Try Reset method before closing the source. Reset is available in the latest commit in the main branch:

go get github.com/ebitengine/oto/v3@a6fe004d7ec62b239a5501f0dfc7f42e3f1adad8

[codesoap]

It works flawlessly for me. Thanks a lot! I especially like that Reset blocks until Read is done.