Timer appears to be broken in headless setups

[lyonbeckers]

Bevy version

v0.16.0

What you did

Using a headless setup with manual updates, when upgrading from v0.15 to v0.16, Time.delta_secs() is reporting durations which are ~2x the actual elapsed time between calls to update. This was originally observed in a Godot project using bevy via gdext, but it can be reproduced with a minimal project:

Cargo.toml:

[package]
name = "slow_timer"
version = "0.1.0"
edition = "2024"

[dependencies]
bevy = { version = "0.16.0", default-features = false }

use std::time::Duration;

use bevy::prelude::*;

fn main() {
    let mut app = App::new();
    app.add_plugins(MinimalPlugins).finish();

    loop {
        app.update();
        let time = app.world().resource::<Time>();
        dbg!(time.delta_secs());
        std::thread::sleep(Duration::from_secs_f32(0.01));
    }
}

Which logs:

...
[src/main.rs:12:9] time.delta_secs() = 0.025558336
[src/main.rs:12:9] time.delta_secs() = 0.02548195
[src/main.rs:12:9] time.delta_secs() = 0.025628762
[src/main.rs:12:9] time.delta_secs() = 0.025477076
[src/main.rs:12:9] time.delta_secs() = 0.025474293
[src/main.rs:12:9] time.delta_secs() = 0.0254678
[src/main.rs:12:9] time.delta_secs() = 0.025397876
[src/main.rs:12:9] time.delta_secs() = 0.025550036

Note that if I downgrade this minimal example to v0.15.3, I get the correct expected output:

...
[src/main.rs:12:9] time.delta_secs() = 0.010308759
[src/main.rs:12:9] time.delta_secs() = 0.010201677
[src/main.rs:12:9] time.delta_secs() = 0.010415471
[src/main.rs:12:9] time.delta_secs() = 0.010514128
[src/main.rs:12:9] time.delta_secs() = 0.010201179
[src/main.rs:12:9] time.delta_secs() = 0.010178561
[src/main.rs:12:9] time.delta_secs() = 0.010178767

What went wrong

• Expecting: time.delta_secs() to represent the actual time elapsed between calls to update.
• Actual results: values that are roughly 2x the actual elapsed time.

Additional information

I was trying to look through the changes to spot what could have possibly changed since v0.15, and the only thing I can spot is that there seems to have been some changes to support no_std, but it's not clear to me what within that could be causing it, if anything.