core: tracking issue for tracing-core 0.2 breaking changes

[hawkw]

Feature Request

Crates

• tracing-core

Motivation

In order to reduce churn throughout the ecosystem, we intend to synchronize the tokio 0.3 and tracing-core 0.2 releases. This issue will track and discuss tracing-core 0.2 breaking changes.

Note that some of the listed changes involve phrasing like "Consider doing X". In those cases, the benefits and drawbacks of those changes need to be investigated further before we commit to making them in 0.2.

Minor API Tweaks

These are low-hanging fruit API changes that don't require a great deal of work, but can't be changed without breaking existing APIs.

• Make arguments to set_default and set_global_default consistent (Consistency between various *default methods #455)
• Consider changing Event struct to have non-'static metadata (Cow?) (needs issue)
• Rename Subscriber to Collector, such that tracing_subscriber::layer::Layer can be renamed to Subscriber (Rename tracing_subscriber::layer::Layer to something closer to Subscriber. #641)
• Consider changing Subscriber::new_span to return Option<Id> (core: consider changing Subscriber::new_span to return Option<Id> #924)
• Remove APIs deprecated in v0.1.x
• Remove default impls from some Collect methods (core: remove Collect default impls for 0.2 #1262)

Metadata/Callsite improvements

• Investigate removing cyclic Metadata <-> Callsite reference (needs issue)
• Consider using Cow<str, 'static> for Metadata strings rather than &'static str (needs issue)
• Consider introducing statically discoverable metadata (core: introduce statically discoverable metadata #969)

Value System Improvements

See RFC for ValueSet ergonomics improvements #697.

• Rework ValueSets to support indexing as well as visiting (core: Rework ValueSets to support indexing in addition to visiting #926)
• Consider changing Value to dispatch primitives w/ an enum rather than trait objects (core: Consider changing Value to dispatch primitives with an enum rather than trait objects #925)
• Figure out Pass slices into FieldSet::value_set instead of array references #782 while we're messing with ValueSets
• valuable integration (core: tracking issue for valuable integration #1570)

[hawkw]

There are also a handful of proposed changes which don't appear to inherently require breaking API changes, but may not be possible to implement without breaking changes (or could be implemented more easily with breaking changes). It would be worth doing further investigation of those proposed changes while we have the opportunity to break things for 0.2.

These include:

• Avoid two statics for every callsite #921 (probably much easier if we figure out a good way to remove the cyclical callsite/metadata reference)
• core: consider using an atomic linked list for the callsite registry #860

[hawkw]

There are probably other issues and proposed changes I've missed, both on the lists of definite breaking changes, and on the list of proposed changes that should be investigated prior to shipping a breaking release. If anyone can think of anything I've left out, please post them here!

[hawkw]

Opened #1262 for removing some default method impls from Collect that were initially added for backwards compatibility when adding new methods.

[carllerche]

I believe there was some discussion about changing the Subscriber/Collector trait to enable avoiding map lookups when handling events. Is that reflected yet? If not, I am leaving the comment so we don't forget 😄 .

[hawkw]

I believe there was some discussion about changing the Subscriber/Collector trait to enable avoiding map lookups when handling events. Is that reflected yet? If not, I am leaving the comment so we don't forget smile .

There isn't an issue for that yet --- I thought you said you were going to write a proposal for it, but I may have misremembered that. I can try to write one if you'd prefer?

[carllerche]

I can write a proposal, but we should have a placeholder issue until then :)