* Remove instrument selector predicates, add instrument name wildcards
* Cache name pattern to regex translation
* Use alternative glob match algo, get rid of cache
* Switch View and ViewBuilder to classes
* Make AttributesProcessor an empty interface, move content to AbstractAttributesProcessor
* Drop io.opentelemetry.sdk.metrics.view package
* Simplify MetricReader
* Tighten up PR
* Restore some methods / classes and deprecate
* Revert "Restore some methods / classes and deprecate"
This reverts commit fee90e5da7.
* fix:add an ending period to form notes
* feat:add the shortcut of removeif in ResourceBuilder
* fix:fix the errors of check
* feat:run two gradlew task to achieve check
* fix:fix format violations
* feat:remove extra function and modify the unit test
* fix: fix format violations
* feat:format code and modify unit test
* feat:make unit test specification
* feat:cleanup for normalize
* feat: Add enabled resource providers configuration
* style: Code style changes by running spotlessApply
* test: Add unit test for specific resource providers in both configurations
* test: Add unit test for only specific resource providers in enabled configuration
* Add experimental view config module
* Rename view-config to metric-incubator
* Switch naming from camelCase to snake_case
* Extend with attribute key filter
* Wire up to autoconfiguration
* Use snakeyaml instead of jackson
* PR feedback
* PR feedback
* Use noop meter provider if no readers registered
* Extend AutoConfigurationCustomizer with metrics customizers
* Switch from inline to instance level mocks
* Deprecate SdkMeterProviderConfigurer
* Make SPI factories instantiate implementations lazily
Previously, this would load all SPI factories and then eagerly instantiate each implementation. This would potentially result in implementations that would get instantiated even if they didn't end up getting used. An example was if the AWS contrib jar was included, this would result in an AwsXrayRemoteSampler being instantiated eagerly despite the configuration not using it. This could then result in errors or unused resources depending on the implementation i.e. in the AwsXrayRemoteSampler example it spawns a thread and makes a network request.
This change transitions the factories so that they only construct the implementation objects if the configuration specifically asks for it. Note the SPI factories themselves must still must be eagerly loaded to determine which factory names are available. This change applies to all auto-configuration areas following this pattern: Samplers, Metric Exporters, Text Map Propagators and Span Exporters.
* Now that instantiating implementations is on-demand (only if used), allow any exceptions to propagate
* format fix
* Change NamedSpiManager from interface to class
Co-authored-by: jack-berg <jberg@newrelic.com>
* refactor: adjust tracer and meter provider customizer execution order
* refactor: Modify tracer provider customizer flow
* style: Format code using spotlessApply
jack-berg
Anuraag Agrawal
Mehdi Riahi
Trask Stalnaker
Brian Yan
jkamon
wallezhang
John Watson
Ryan Rupp