subscriber: added explicit with_timestamp method for any impls that use without_time

[mkatychev]

Motivation

• Using .without_time() is inconsistent and lacks the ergonimics of current Builder::with_flag methods.

Example of how timestamps are currently toggled:

impl Opts {
    pub fn init(&self) -> Result<(), SetGlobalDefaultError> {
        let filter = EnvFilter::builder()
            .with_env_var("RUST_LOG")
            .with_default_directive(LevelFilter::DEBUG.into())
            .from_env_lossy();
        // Initialize tracing;
        let collector = fmt::Subscriber::builder()
            .with_env_filter(filter)
            .with_line_number(self.line_number)
            .with_file(self.file)
            .with_thread_ids(self.thread_ids)
            .with_target(self.target)
            .with_ansi(self.ansi);
        // life becomes difficult from here on
        if !self.timestamp {
            return tracing::subscriber::set_global_default(collector.without_time().finish());
        };
        tracing::subscriber::set_global_default(collector.finish())
    }
}

Solution

• Added tracing_subscriber::fmt::Layer::with_timestamp(self, display_timestamp: bool)
• Changed timer type to an Option:

pub struct Format<F = Full, T = SystemTime> {
    // ..
-   pub(crate) timer: <T>,
+   pub(crate) timer: Option<T>,
    // ..
}

• Retained all without_time() methods to maintain backwards compatibility

[mkatychev]

should without_time be marked with a deprecated flag?

[mkatychev]

Do we want to maintain impl FormatTime for () or should it have a deprecation flag now that it is replaced by Option::None as <T: FormatTime>?

tracing/tracing-subscriber/src/fmt/time/mod.rs

Lines 81 to 85 in 55fdc75

	impl FormatTime for () {
	fn format_time(&self, _: &mut Writer<'_>) -> fmt::Result {
	Ok(())
	}
	}

[hawkw]

Do we want to maintain impl FormatTime for () or should it have a deprecation flag now that it is replaced by Option::None as <T: FormatTime>?

tracing/tracing-subscriber/src/fmt/time/mod.rs

Lines 81 to 85 in 55fdc75

	impl FormatTime for () {
	fn format_time(&self, _: &mut Writer<'_>) -> fmt::Result {
	Ok(())
	}
	}

The implementation for () has less runtime overhead than the implementation for Option, as it doesn't have to check at runtime whether the Option is Some or None; it just always does nothing. Therefore, I don't think it should be removed, as it's preferable in cases where the timestamp formatter will never be enabled.

[boardwalk]

I ran into this recently (no with_timestamp method, and without_time changing the type making things a little more ugly).

The other, related thing, I ran into is not being able to toggle display_timestamp off separately from time as a whole, which, from my shallow understanding of the code, would stop the timestamp from being displayed (which is what I want, I have an outside process that timestamps logs) but not the duration timing (which I want to keep). It seems to be an all or nothing thing.

What if there was a with_timestamp that worked like the other with_* functions that just change a display_* bool? Calling something that disables durations as well as the timestamps with_timestamp as in this PR seems wrong.