[bug] Command::output() never executes/completes in async commands

[FabianLars]

Describe the bug

If you try to execute a sidecar or Commands in general in an async command it never drives to completion (or probably doesn't even start in the first place).

I tested this only on windows, but according to discord it's a problem on linux or macOS too (can't remember which one they had)

Reproduction

Create a simple async command (I'm using windows for this example):

#[tauri::command]
async fn list_dir() {
    let c = Command::new("powershell.exe").args(["ls"]);
    dbg!(c.output().unwrap());
}

It works if you remove the async keyword. Same if you use (the underlying) std::process::Command::output().
Command::spawn() works fine fwiw.

Expected behavior

No response

Platform and versions

Operating System - Windows, version 10.0.19044 X64
Webview2 - 98.0.1108.55
Visual Studio Build Tools:
   - Visual Studio Build Tools 2022

Node.js environment
  Node.js - 16.14.0
  @tauri-apps/cli - 1.0.0-rc.5
  @tauri-apps/api - 1.0.0-rc.1

Global packages
  npm - 8.3.1
  pnpm - 6.31.0
  yarn - Not installed

Rust environment
  rustup - 1.24.3
  rustc - 1.58.1
  cargo - 1.58.0
  toolchain - stable-x86_64-pc-windows-msvc 

App directory structure
/dist
/node_modules
/src
/src-tauri

App
  tauri - 1.0.0-rc.3 (no lockfile)
  tauri-build - no manifest (no lockfile)
  tao - no manifest (no lockfile)
  wry - no manifest (no lockfile)
  build-type - bundle
  CSP - default-src 'self'
  distDir - ../dist
  devPath - http://localhost:3000/
  framework - React

Stack trace

No response

Additional context

No response

[JonasKruckenberg]

For context the discord thread: https://discord.com/channels/616186924390023171/731495047962558564/944170846707122216

Do we have any ideas why this may be happening? Command::spawn works like a charm so it's the output method that is the problem?

[lucasfernog]

It's probably the safe_block_on implementation.

[lucasfernog]

I think we can refactor the Command APIs a little bit and make them async instead, and offer blocking variants similar to what we did for the dialog APIs.

[lucasfernog]

First I'll try to fix the safe_block_on impl since that one is used on other places too.