Can't block on a runtime from within a block_in_place call

[bdonlan]

Version

Using tokio 0.2.21 on the playground: https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=3f63cb60db402855462f41a6980e719b

When reproducing locally, here are the tokio crate versions:

│   │   │   ├── tokio v0.2.21
│   │   │   │   └── tokio-macros v0.2.5
│   │   │   └── tokio-util v0.3.1
│   │   │       └── tokio v0.2.21 (*)
│   │   ├── tokio v0.2.21 (*)
│   ├── tokio v0.2.21 (*)
└── tokio v0.2.21 (*)

Platform
64-bit linux, 5.4.30

Description

Often synchronous library APIs can be implemented in terms of a private tokio runtime. While it would be preferable to structure calls into these libraries as async calls, this isn't always possible, and so it should be possible to enter a blocking context using block_in_place and then block_on a different runtime. However, currently this fails:

use tokio; // 0.2.21

#[test]
fn tokio_nesting() {
    let mut rt = tokio::runtime::Builder::new()
        .enable_all()
        .threaded_scheduler()
        .build()
        .unwrap();

    rt.block_on(async {
        tokio::task::block_in_place(|| {
            let mut rt = tokio::runtime::Builder::new()
                .enable_all()
                .basic_scheduler()
                .build()
                .unwrap();
            rt.block_on(tokio::time::delay_for(std::time::Duration::from_millis(50)));
        })
    });
}

fails with:

running 1 test
test tokio_nesting ... FAILED

failures:

---- tokio_nesting stdout ----
thread 'tokio_nesting' panicked at 'Cannot start a runtime from within a runtime. This happens because a function (like `block_on`) attempted to block the current thread while the thread is being used to drive asynchronous tasks.', /playground/.cargo/registry/src/github.com-1ecc6299db9ec823/tokio-0.2.21/src/runtime/enter.rs:38:5
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    tokio_nesting

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

[Darksonn]

To be specific, this fails because you are not doing it on a worker thread managed by the threaded runtime. If you add a spawn around the block_in_place, it works.

[carllerche]

This is still a bug though. We probably missed something when adding support for block_in_place off threaded RT threads.