v1.1.0-beta.14

Changed

• Requires dartastic_opentelemetry_api ^1.0.0-rc.2. Picks up the
  semantic conventions at registry v1.44.0 — including the full
  browser.web_vital.* set — and three spec-compliance fixes.

  One of those changes behaviour visible from this package:
  Context.withSpanContext now returns a derived Context when the incoming
  span context belongs to a different trace, instead of throwing
  ArgumentError. Per the Context specification a set-value operation always
  returns a derived Context, and per the Propagators API extract must never
  throw — receiving a valid span context for another trace during extraction
  is ordinary, not an error. Four tests that asserted the throw now assert the
  derived Context.

  SDKSpan.end() no longer forwards the deprecated spanStatus argument to
  its delegate; setStatus() had already applied it to the same delegate, so
  the second pass was redundant.

• BREAKING (spec compliance): sampler decisions are now honored end to end
  (#120, #121, #122, #123, #129). A Drop decision produces a non-recording
  span that reaches no processor; RecordOnly records without setting the
  W3C Sampled flag; built-in processors deliver only recording spans to
  onStart/onEnd and only sampled spans to exporters, per the spec's
  IsRecording/Sampled reaction table; the forbidden Sampled+non-recording
  combination is unrepresentable; createSpan routes through the sampler and
  processors instead of bypassing them. Code that relied on unsampled or
  dropped spans being exported must adjust its sampler configuration.

• BREAKING (spec compliance): the default sampler is now
  ParentBased(root: AlwaysOn) instead of bare AlwaysOn (#126). Root
  spans still sample by default, but child spans of unsampled remote or local
  parents now respect the parent's decision. Pass
  sampler: const AlwaysOnSampler() to restore the old behavior.

• Child spans inherit the parent TraceState (#124), and
  SamplingResult gains a traceState field (#125) so samplers can
  modify or replace it: null keeps the inherited parent TraceState, an
  explicitly empty TraceState clears it. Existing custom samplers keep
  working unchanged.

• BREAKING: OTelEnv configuration functions (getOtlpConfig, getBspConfig, getServiceConfig, getLogRecordLimits, etc.) now return strongly-typed Dart Records instead of Map<String, dynamic>.
  Migration hint: Update map accesses to record property accesses (e.g., config['endpoint'] as String? → config.endpoint).

Added

• TracerProvider.hasSpanProcessors — allocation-free check for registered
  span processors.

• OTLP exporters send an identifying User-Agent header. Per the OTLP
  spec, OTLP requests SHOULD identify the exporter, language, and version.
  Every OTLP request now carries OTel-OTLP-Exporter-Dart/<version> (HTTP
  headers on the HTTP exporters; ChannelOptions.userAgent on the gRPC
  exporters). A user-supplied user-agent header is prepended to the default
  rather than replacing it (#228).

• OTel.defaultGrpcEndpoint (http://localhost:4317), the OTLP/gRPC
  default endpoint. The endpoint default is now picked per signal after the
  protocol is resolved instead of defaulting everything to the HTTP port
  4318 (#220).

Fixed

• browser.* resource attributes and user_agent.original are now
  populated on web (#190). Invalid @JS bindings made the web resource
  detector throw on first use; the error was swallowed and the attributes
  were silently missing. A detector failure now omits attributes instead of
  emitting blanks.

• browser.mobile is now a boolean, which is how the registry types it.
  It was emitted as the string 'true'/'false', so a backend filtering
  browser.mobile = true matched nothing.

• browser.languages is now a string array, and its key comes from the
  API. It was a comma-joined string under a key this package declared
  privately. The naming rules say an attribute that can represent multiple
  entities "SHOULD be pluralized and the value type SHOULD be an array",
  which is also how the registry shapes the neighbouring browser.brands.
  The key is now BrowserCandidate.browserLanguages, staged in the API as an
  upstream candidate, so the name and the argument for it live in one place.
  It is set only when the browser reports languages: an empty array is a real
  value on the wire and would claim the browser accepts none.

• browser.vendor is no longer emitted. navigator.vendor is a frozen
  legacy API that returns a hardcoded vendor string rather than the real
  vendor, and the registry's browser.brands is the structured answer to
  the same question.

• browser.mobile is now correct on iPad. Since iPadOS 13 an iPad
  requests desktop sites by default and reports a Macintosh user agent,
  so a user-agent test alone reported every iPad as a desktop. The
  detector now also consults navigator.maxTouchPoints, which
  distinguishes a touch device from a Mac. A touchscreen laptop is still
  not mobile — both signals have to agree.

• OtlpHttpMetricExporter.forceFlush() and shutdown() now await
  in-flight exports (#262). Both returned immediately, and shutdown()
  closed the HTTP client under the live request — failing an export that
  was about to succeed. Now matches the span and log HTTP exporters.

• Tracer.enabled now returns false when TracerProvider has no span
  processor(s) registered, per the Trace SDK spec, sparing span-creation cost
  when nothing is listening. Thanks to @abidiahmedcom (#138, #175).

• OTLP/gRPC exporters now default to port 4317, not 4318. The OTLP spec
  defaults the endpoint to http://localhost:4317 for OTLP/gRPC and
  http://localhost:4318 for the two HTTP protocols. Previously a single
  4318 default was applied before the protocol was known, so gRPC-only
  deployments silently exported to the wrong port (#220).

• An empty environment variable value is treated as unset. Per the spec,
  an empty value of an environment variable MUST be interpreted the same way
  as when the variable is unset. EnvironmentService.getValue now normalizes
  empty strings to null, so every consumer (endpoint, protocol, service
  name, log level, …) reads an empty value as unset (#213).

  Thanks to @abidiahmedcom; reported by @yuzurihaaa (#213, #220, #228).

• service.name in OTEL_RESOURCE_ATTRIBUTES no longer overrides an
  explicit serviceName: argument or OTEL_SERVICE_NAME (#103).
  Precedence is now, highest first: explicit argument, OTEL_SERVICE_NAME,
  OTEL_RESOURCE_ATTRIBUTES, default. service.version follows the same
  order.

• OTEL_LOG_LEVEL now takes effect at the start of OTel.initialize, so
  debug logging covers the environment parsing itself.

• Baggage values containing = (e.g. base64 padding) are no longer
  dropped on extract, and an unparsable baggage header leaves existing
  baggage untouched instead of clearing it. Thanks to @abidiahmedcom
  (#199, #200, #261).