Rename otel/label -> otel/attribute

[punya]

Fixes #1069.

[XSAM]

Should we use attribute as the package name instead of attr?

attr is short and elegant, but there is no naming in Specification called attr; I'm worried about this inconsistency might confuse end-users.

[Aneurysm9]

I appreciate the desire to minimize the changes to review in this diff, but it seems that using import renaming to avoid changing all references from label to attribute results in go doc continuing to reference label in exported function and type signatures. I worry that this could lead to confusion when users look to import a label package but don't find one.

[codecov]

Codecov Report

Merging #1541 (4ee7baf) into main (1b5b662) will increase coverage by 0.1%.
The diff coverage is 85.7%.

@@           Coverage Diff           @@
##            main   #1541     +/-   ##
=======================================
+ Coverage   77.9%   78.0%   +0.1%     
=======================================
  Files        127     127             
  Lines       6582    6582             
=======================================
+ Hits        5128    5136      +8     
+ Misses      1209    1201      -8     
  Partials     245     245             

Impacted Files	Coverage Δ
attribute/encoder.go	85.1% <ø> (ø)
attribute/iterator.go	82.9% <ø> (ø)
attribute/key.go	100.0% <ø> (ø)
attribute/kv.go	82.1% <ø> (ø)
attribute/set.go	57.1% <ø> (ø)
attribute/type_string.go	10.0% <ø> (ø)
attribute/value.go	90.9% <ø> (ø)
bridge/opentracing/internal/mock.go	74.3% <0.0%> (ø)
exporters/stdout/config.go	70.0% <ø> (ø)
exporters/trace/zipkin/model.go	98.6% <ø> (ø)
... and 46 more

[punya]

Note that after this change, we still have some public identifiers that mention "label". Should any of these be changed too?

10 results - 4 files

exporters/otlp/internal/opentelemetry-proto-gen/metrics/v1/metrics.pb.go:
   850  	// The set of labels that uniquely identify this timeseries.
   851: 	Labels []*v11.StringKeyValue `protobuf:"bytes,1,rep,name=labels,proto3" json:"labels,omitempty"`
   852  	// start_time_unix_nano is the last time when the aggregation value was reset

   951  	// The set of labels that uniquely identify this timeseries.
   952: 	Labels []*v11.StringKeyValue `protobuf:"bytes,1,rep,name=labels,proto3" json:"labels,omitempty"`
   953  	// start_time_unix_nano is the last time when the aggregation value was reset

  1054  	// The set of labels that uniquely identify this timeseries.
  1055: 	Labels []*v11.StringKeyValue `protobuf:"bytes,1,rep,name=labels,proto3" json:"labels,omitempty"`
  1056  	// start_time_unix_nano is the last time when the aggregation value was reset

  1207  	// The set of labels that uniquely identify this timeseries.
  1208: 	Labels []*v11.StringKeyValue `protobuf:"bytes,1,rep,name=labels,proto3" json:"labels,omitempty"`
  1209  	// start_time_unix_nano is the last time when the aggregation value was reset

exporters/stdout/config.go:
   46  	// LabelEncoder encodes the labels.
   47: 	LabelEncoder attribute.Encoder
   48  

  119  type labelEncoderOption struct {
  120: 	LabelEncoder attribute.Encoder
  121  }

oteltest/meter.go:
   31  		Instrument *Sync
   32: 		Labels     []attribute.KeyValue
   33  	}

   38  		Ctx          context.Context
   39: 		Labels       []attribute.KeyValue
   40  		LibraryName  string

  201  	InstrumentationVersion string
  202: 	Labels                 map[attribute.Key]attribute.Value
  203  	Number                 number.Number

sdk/metric/processor/reducer/reducer.go:
  32  	// specific Filter to an instrument.
  33: 	LabelFilterSelector interface {
  34  		LabelFilterFor(descriptor *metric.Descriptor) attribute.Filter

[Aneurysm9]

Note that after this change, we still have some public identifiers that mention "label". Should any of these be changed too?

Other than the ones in the protobuf, which should probably be addressed via OTLP changes, yeah I would think those should change to Attribute as well. Thoughts @MrAlias?

[punya]

@Aneurysm9 @XSAM can we leave those for future PRs? Some of them are related to metrics (not trace) and the change should probably cover names, comments, docs, etc. The package rename should suffice to unblock the trace release.

I can file an issue to track getting rid of the concept of "labels" more broadly.

[Aneurysm9]

Going to merge this as-is to avoid keeping a large PR hanging out there for too long then I'll create a new issue for 1.0.0 to address the public API pieces still referencing Label.

[Aneurysm9]

Suggested change

	// IndexedLabel returns current index and attribute. Must be called only
	// IndexedAttribute returns current index and attribute. Must be called only

Also the Label() method should be changed to Attribute(). Since these are public API changes I don't think we can defer them.