Recording for Floats

[jracollins]

Motivation

Currently floats cannot be logged/recorded other than in Debug or Display format which ends up stringifying them on output.

Simple Example:

#[macro_use]
extern crate tracing;

use std::error::Error;

fn main() -> Result<(), Box<dyn Error>> {
    let builder = tracing_subscriber::fmt().json();
    builder.init();

    let integer: i64 = 25;
    let float: f64 = 25.567;

    event!(tracing::Level::INFO, integer = integer, "Test Integer");
    event!(tracing::Level::INFO, float = %float, "Test Display Float");
    event!(tracing::Level::INFO, float = ?float, "Test Debug Float");

    Ok(())
}

results in:

{"timestamp":"Apr 25 14:15:35.469","level":"INFO","fields":{"message":"Test Integer","integer":25},"target":"tracing_test"}
{"timestamp":"Apr 25 14:15:35.469","level":"INFO","fields":{"message":"Test Display Float","float":"25.567"},"target":"tracing_test"}
{"timestamp":"Apr 25 14:15:35.469","level":"INFO","fields":{"message":"Test Debug Float","float":"25.567"},"target":"tracing_test"}

And won't compile if passed the float directly:

error[E0277]: the trait bound `f64: Value` is not satisfied
  --> src/main.rs:15:5
   |
15 |     event!(tracing::Level::INFO, float = float, "Test Float");
   |     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ the trait `Value` is not implemented for `f64`
   |
   = note: required for the cast to the object type `dyn Value`
   = note: this error originates in a macro (in Nightly builds, run with -Z macro-backtrace for more info)

error: aborting due to previous error

Log ingesting/processing tools (eg Datadog/ELK etc) cannot parse these float strings without custom rules for each field to try and coerce the outputted string into the appropriate type, which is required in order to generate metrics etc.

This resolves issue mentioned in the Tokio Discord a few times and #726

Solution

In order to implement I have added the relevant record_64 functions.

However, in order to get it working I had to comment out the impl_one_value!(nonzero, ...) macro.

I don't currently understand the benefits of the implementation (why it is needed), or how to avoid it causing conflict with the float implementation.

Before I write/fix the tests for the implementation/remove current NonZero failing test, if a reviewer could explain why it is needed and offer pointers as to how to work around it, or green light its removal it would be much appreciated 🙏

[hawkw]

Obsoleted by #1507.

[jracollins]

Hi @hawkw

I just updated my app to use version 1.29 instead of this PR's forked version, and whilst it does now accept floats as input values, the actual output of the floats are stringified? (Which makes it basically the same as before using output = %float_value) and is one of the problems described that I was trying to fix?

i.e. output from

fn main() -> Result<(), Box<dyn Error>> {
    tracing_subscriber::fmt().json().init();
    event!(tracing::Level::INFO, float = 0.24, "Float");
    Ok(())
}

using 1.29 is:

{"timestamp":"Oct 22 13:29:02.154","level":"INFO","fields":{"message":"Float","float":"0.24"},"target":"trace_test"}

Where float is "0.24" rather than 0.24

[w1th0utnam3]

I just updated my app to use version 1.29 instead of this PR's forked version, and whilst it does now accept floats as input values, the actual output of the floats are stringified?

This still seems to be an open problem as of now. Is there an update on this?