Hot reload is 4x slower than `cargo run`

[zippoxer]

What were you trying to do?

Hot reload a compilation-speed optimized Bevy app.

Runnable reproduction: https://github.com/zippoxer/bevy-dexterous-developer-test

System: Ubuntu 24.04 under WSL2 on Windows 11 with AMD 5900X and 64 GB RAM.

How did you know this was slow?

cargo run after changing a constant:

Finished `dev` profile [optimized + debuginfo] target(s) in 1.06s

dexterous_developer_cli reload after changing a constant:

Finished `dev` profile [optimized + debuginfo] target(s) in 4.96s

So the build is almost 5x slower without counting the extra delays before and after it (totalling just a few more seconds), although I assume these are just the costs of performing the reload 🤔

Note: the constant I'm changing is BLUE_600 to RED_600 in one of the systems.

[Optional] Why is this slow?

Is there any chance that dexterous_developer_cli uses different build parameters than cargo run does?

I've applied lots of build-time optimizations to the reproducible repo, which made it much faster.

For example: linker modifications in .cargo/config.toml and cranelift usage.

[lee-orr]

Thanks for filing this! The examples/explanation is really helpful in understanding the situation. I'm unfortunately fairly busy right now, so my work has been slow - but I will update you when I get closer to it.