heterogeneous extraction, untested

src/datadog/extraction_util.cpp

@@ -1,5 +1,6 @@
 #include "extraction_util.h"
 
+#include <algorithm>
 #include <cstdint>
 #include <sstream>
 #include <string>
@@ -229,5 +230,54 @@ void AuditedReader::visit(
   });
 }
 
+ExtractedData merge(const std::vector<ExtractedData>& contexts) {
+  ExtractedData result;
+
+  const auto found = std::find_if(
+      contexts.begin(), contexts.end(),
+      [](const ExtractedData& data) { return data.trace_id.has_value(); });
+
+  if (found == contexts.end()) {
+    // Nothing extracted a trace ID. Return the first context that includes a
+    // parent ID, if any, or otherwise just return an empty `ExtractedData`.
+    // The purpose of looking for a parent ID is to allow for the error
+    // "extracted a parent ID without a trace ID," if that's what happened.
+    const auto other = std::find_if(
+        contexts.begin(), contexts.end(),
+        [](const ExtractedData& data) { return data.parent_id.has_value(); });
+    if (other != contexts.end()) {
+      result = *other;
+    }
+    return result;
+  }
+
+  // `found` refers to the first extracted context that yielded a trace ID.
+  // This will be our main context.
+  //
+  // If the style of `found` is not W3C, then examine the remaining contexts
+  // for W3C-style tracestate that we might want to include in `result`.
+  result = *found;
+  if (result.style == PropagationStyle::W3C) {
+    return result;
+  }
+
+  const auto other =
+      std::find_if(found, contexts.end(), [&](const ExtractedData& data) {
+        return data.style == PropagationStyle::W3C &&
+               data.trace_id == found->trace_id;
+      });
+
+  if (other != contexts.end()) {
+    result.additional_w3c_tracestate = other->additional_w3c_tracestate;
+    result.additional_datadog_w3c_tracestate =
+        other->additional_datadog_w3c_tracestate;
+    result.headers_examined.insert(result.headers_examined.end(),
+                                   other->headers_examined.begin(),
+                                   other->headers_examined.end());
+  }
+
+  return result;
+}
+
 }  // namespace tracing
 }  // namespace datadog

src/datadog/extraction_util.h

@@ -86,5 +86,12 @@ struct AuditedReader : public DictReader {
                  visitor) const override;
 };
 
+// Combine the specified trace `contexts`, each of which was extracted in a
+// particular propagation style, into one `ExtractedData` that includes fields
+// from compatible elements of `contexts`, and return the resulting
+// `ExtractedData`. The order of the elements of `contexts` must correspond to
+// the order of the configured extraction propagation styles.
+ExtractedData merge(const std::vector<ExtractedData>& contexts);
+
 }  // namespace tracing
 }  // namespace datadog

src/datadog/tracer.cpp

@@ -134,7 +134,7 @@ Expected<Span> Tracer::extract_span(const DictReader& reader,
   AuditedReader audited_reader{reader};
 
   auto span_data = std::make_unique<SpanData>();
-  ExtractedData extracted_data;
+  std::vector<ExtractedData> extracted_contexts;
 
   for (const auto style : extraction_styles_) {
     using Extractor = decltype(&extract_datadog);  // function pointer
@@ -159,19 +159,13 @@ Expected<Span> Tracer::extract_span(const DictReader& reader,
       return error->with_prefix(
           extraction_error_prefix(style, audited_reader.entries_found));
     }
-    extracted_data = *data;
-    extracted_data.headers_examined = audited_reader.entries_found;
-    // If the extractor produced a non-null trace ID, then we consider this
-    // extraction style the one "chosen" for this trace.
-    // Otherwise, we loop around to the next configured extraction style.
-    if (extracted_data.trace_id) {
-      break;
-    }
+    extracted_contexts.push_back(std::move(*data));
+    extracted_contexts.back().headers_examined = audited_reader.entries_found;
   }
 
-  auto& [trace_id, parent_id, origin, trace_tags, sampling_priority,
-         additional_w3c_tracestate, additional_datadog_w3c_tracestate, style,
-         headers_examined] = extracted_data;
+  auto [trace_id, parent_id, origin, trace_tags, sampling_priority,
+        additional_w3c_tracestate, additional_datadog_w3c_tracestate, style,
+        headers_examined] = merge(extracted_contexts);
 
   // Some information might be missing.
   // Here are the combinations considered: