Hacktoberfest 2026：维护者为十月标记出来的 issue，仍然开放、适合新手。 浏览 Hacktoberfest issue

WASAPI: default-device activation fails with RPC_E_CHANGED_MODE after endpoint-notification callbacks put thread-pool threads into an STA

未关闭
  • 哪里不对？
#1,402 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

@dannyward630 已经在做这个了。

开始于 2026年10月24日。

  • #1413 来自 @dannyward630 —— 未关闭

评估

难度
4/5
预计耗时
3-5 天
新手友好度
20/100
Issue 类型
缺陷
描述清晰度
基本清楚
活跃度
活跃
技术栈
rust
领域
audio-video-rtc

调研方向

Start in src/host/wasapi/stream.rs at DefaultDeviceNotificationImpl::OnDeviceRemoved and OnDeviceStateChanged, then follow get_current_default into current_default_endpoint in device.rs to where com_initialized() is called. Run the reproduction in the issue, which polls the default output device after thread-pool callbacks, and confirm RPC_E_CHANGED_MODE on the second open. Done means the default device opens after those callbacks run, without changing the calling thread's apartment.

由索引模型根据 Issue 内容生成。

描述

correctness host - wasapi platform - windows

cpal version: 0.18.2 (WASAPI host), Windows 11

What happens

After some time, opening the default input or output device fails for the rest of the process:

Failed to get audio client: Cannot change thread mode after it is set. (os error -2147417850)

(0x80010106 = RPC_E_CHANGED_MODE). Devices obtained from input_devices() / output_devices() (specific IMMDevices) keep working; only default_input_device() / default_output_device() fail. Only a restart of the process helps.

Cause
  1. DefaultDeviceNotificationImpl::OnDeviceRemoved and OnDeviceStateChanged (src/host/wasapi/stream.rs) call get_current_default → current_default_endpoint (device.rs), which calls com::com_initialized().
  2. These callbacks run on threads owned by Windows (MMDevAPI / thread pool), not on cpal's threads. com_initialized() calls CoInitializeEx(COINIT_APARTMENTTHREADED) there. On a pool thread that has no COM initialization of its own this succeeds, so the pool thread becomes an STA. The thread-local guard is only dropped when the thread exits, which pool threads normally do not, so the thread stays STA.
  3. For the default device, cpal activates via ActivateAudioInterfaceAsync(DEVINTERFACE_AUDIO_*) (activate_audio_interface_sync). When the process has a multithreaded apartment (any thread did CoInitializeEx(COINIT_MULTITHREADED), which many libraries do), the activation runs on the thread pool and fails with RPC_E_CHANGED_MODE on a pool thread that cpal turned into an STA.
Reproduction

Simulates step 1–2 by calling cpal's own path on pool threads (what the notification callbacks do):

use cpal::traits::{DeviceTrait, HostTrait};
use std::sync::atomic::{AtomicUsize, Ordering};
use windows::Win32::System::{Com::*, Threading::*};

static DONE: AtomicUsize = AtomicUsize::new(0);

unsafe extern "system" fn cb(_: PTP_CALLBACK_INSTANCE, _: *mut core::ffi::c_void) {
    let _ = cpal::default_host().default_output_device(); // -> current_default_endpoint -> com_initialized
    std::thread::sleep(std::time::Duration::from_millis(200));
    DONE.fetch_add(1, Ordering::SeqCst);
}

fn open() {
    std::thread::spawn(|| {
        let d = cpal::default_host().default_output_device().unwrap();
        println!("{:?}", d.supported_output_configs().map(|c| c.count()));
    })
    .join()
    .unwrap();
}

fn main() {
    // Some other thread in the MTA, as libraries commonly have.
    std::thread::spawn(|| {
        let _ = unsafe { CoInitializeEx(None, COINIT_MULTITHREADED) };
        loop { std::thread::park(); }
    });
    std::thread::sleep(std::time::Duration::from_millis(100));
    open(); // Ok(..)
    for _ in 0..64 {
        unsafe { TrySubmitThreadpoolCallback(Some(cb), None, None).unwrap() };
    }
    while DONE.load(Ordering::SeqCst) < 64 {
        std::thread::sleep(std::time::Duration::from_millis(10));
    }
    open(); // Err(Failed to get audio client: Cannot change thread mode after it is set. ...)
}

Observed: the first open() succeeds, every later one fails with RPC_E_CHANGED_MODE. Without the MTA thread both succeed. Calling a specific device's supported_output_configs() after the poisoning still succeeds.

Suggested fix

Code that runs on threads cpal does not own (the IMMNotificationClient callbacks, Drop of DefaultDeviceMonitor if it can run elsewhere) should not change that thread's apartment. For example, current_default_endpoint could skip com_initialized() when called from the callbacks, or join the MTA there (CoInitializeEx(COINIT_MULTITHREADED) / CoIncrementMTAUsage) instead of making the thread an STA. (Not verified: on which threads MMDevAPI invokes the callbacks; the reproduction above uses the default thread pool.)

Related: #1302 (enumerator bound to the first thread's apartment).

主要语言
Rust
星标
4k
派生
554
平均合并
2 天 11 小时
30 天内合并 PR
3

环境准备

  • 提供 Dockerfile 或 Docker Compose 文件
  • 没有 Pull Request 模板
  • 阅读贡献指南

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

RustAudio/cpal 的其他 Issue

查看 RustAudio/cpal 的全部 Issue

相似的 Issue

更多 Rust Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。