Consider removing inner Arc from TaskMonitor

[jplatte]

I've started using tokio-metrics at work recently and found it a bit unfortunate that if you put TaskMonitors in statics, you have to wrap them in LazyLock or similar for the initialization. It seems the only reason for this is that TaskMonitor is internally an Arc for easy cloning, but it doesn't seem like there is any need for that.

How about an API like this for construction?

// no Clone impl
pub struct TaskMonitor {
    // all the atomics currently found in the inner state type
}

impl TaskMonitor {
    pub fn new() -> Arc<Self> { /* ... */ }
    pub const fn new_static() -> Self { /* ... */ }
    pub const fn builder() -> TaskMonitorBuilder { /* ... */ }
}

pub struct TaskMonitorBuilder {
    // same as now
}

impl TaskMonitorBuilder {
    // with_* methods just like before
    pub fn build(self) -> Arc<TaskMonitor> { /* ... */ }
    pub const fn build_static(self) -> TaskMonitor { /* ... */ }
}

[rcoh]

Adding a StaticTaskMonitor (or a better name) that doesn't have the clone bound seems like a reasonable change. I think a breaking change in the TaskMonitor API itself would be somewhat painful, but I think we can accomplish the same thing in a way that is backwards compatible. Are you interested in contributing this?

[jplatte]

Makes sense. I don't love the name StaticTaskMonitor since another thing I found this could be useful for is avoiding a double indirection by putting the TaskMonitor state in an Arc alongside something else.

It's probably not much better, but how about TaskMonitorBase?

[jlizen]

@jplatte TaskMonitorBase seems fine to me. Were you interested in opening a PR for this?

[jplatte]

I don't have a huge need for this, plus a lot of other things on my plate, so if you're interested in working on this, don't let me stop you :)

[jlizen]

No worries! No cycles to spare here either, but let's keep this open, and if anybody starts to pick up they can drop a note.