Remove a bit more unsafe

This commit is contained in:
Jonathan Kelley committed 2023-07-19 12:26:32 -07:00
1 parent 2871c8bb4d
commit 163fe68f45
1 file changed
+8 -5
+8 -5
View File
@@ -1,9 +1,9 @@
//! Verify that tasks get polled by the virtualdom properly, and that we escape wait_for_work safely
use dioxus::prelude::*;
use std::time::Duration;
use std::{sync::atomic::AtomicUsize, time::Duration};
static mut POLL_COUNT: usize = 0;
static POLL_COUNT: AtomicUsize = AtomicUsize::new(0);
#[tokio::test]
async fn it_works() {
@@ -18,7 +18,10 @@ async fn it_works() {
// By the time the tasks are finished, we should've accumulated ticks from two tasks
// Be warned that by setting the delay to too short, tokio might not schedule in the tasks
assert_eq!(unsafe { POLL_COUNT }, 135);
assert_eq!(
POLL_COUNT.fetch_add(0, std::sync::atomic::Ordering::Relaxed),
135
);
}
fn app(cx: Scope) -> Element {
@@ -26,14 +29,14 @@ fn app(cx: Scope) -> Element {
cx.spawn(async {
for x in 0..10 {
tokio::time::sleep(Duration::from_micros(50)).await;
unsafe { POLL_COUNT += x }
POLL_COUNT.fetch_add(x, std::sync::atomic::Ordering::Relaxed);
}
});
cx.spawn(async {
for x in 0..10 {
tokio::time::sleep(Duration::from_micros(25)).await;
unsafe { POLL_COUNT += x * 2 }
POLL_COUNT.fetch_add(x * 2, std::sync::atomic::Ordering::Relaxed);
}
});
});