stream(tags).of(stream -> stream.sorted().reduce((l, r) -> l + r).ifPresent(v -> buf.append(v)));
        buf.append(CACHE_KEY_SPLITTER);
        stream(roles).of(stream -> stream.sorted().reduce((l, r) -> l + r).ifPresent(v -> buf.append(v)));
        buf.append(CACHE_KEY_SPLITTER);
        stream(fields).of(stream -> stream.sorted().reduce((l, r) -> l + r).ifPresent(v -> buf.append(v)));

Doing this provides the following specific benefits:
- **Reduce the API surface** - We no longer need to maintain two methods.
- **Consistency** - All languages have consistent access to the same APIs and ergonomics in the DSL.
- **Reduce the size of the Gradle distribution** - We no longer need to carry multiple standard libraries for different languages.

## Decision

        .sort(s => -(s.additions + s.deletions));

    let totalAdditions = Object.values(preppedStatistics).reduce((acc, summary) => acc + summary.additions, 0);
    let totalDeletions = Object.values(preppedStatistics).reduce((acc, summary) => acc + summary.deletions, 0);
    let newRatio = totalAdditions / (totalAdditions + totalDeletions) * 100;

Doing this provides the following specific benefits:
- **Reduce the API surface** - We no longer need to maintain two methods.
- **Consistency** - All languages have consistent access to the same APIs and ergonomics in the DSL.
- **Reduce the size of the Gradle distribution** - We no longer need to carry multiple standard libraries for different languages.

## Decision

`Make[T]` function, which produces a reference to a canonical copy of
the value in the form of a `Handle[T]`.
Two `Handle[T]` are equal if and only if the values used to produce the
handles are equal, allowing programs to deduplicate values and reduce
their memory footprint.
Comparing two `Handle[T]` values is efficient, reducing down to a simple

      return DummyProxy.this;
    }

    @Override
    public String toString() {
      return "Dummy proxy for " + interfaceType;
    }

    // Since type variables aren't serializable, reduce the type down to raw type before
    // serialization.
    private Object writeReplace() {
      return new DummyHandler(TypeToken.of(interfaceType.getRawType()));
    }
  }

      return DummyProxy.this;
    }

    @Override
    public String toString() {
      return "Dummy proxy for " + interfaceType;
    }

    // Since type variables aren't serializable, reduce the type down to raw type before
    // serialization.
    private Object writeReplace() {
      return new DummyHandler(TypeToken.of(interfaceType.getRawType()));
    }
  }

 *
 * <p>This interface is now a legacy type. Use {@code java.util.function.Function} (or the
 * appropriate primitive specialization such as {@code ToIntFunction}) instead whenever possible.
 * Otherwise, at least reduce <i>explicit</i> dependencies on this type by using lambda expressions
 * or method references instead of classes, leaving your code easier to migrate in the future.
 *

  TF_DeleteTensor(model_out_tensor);

  // If the output is a scalar, then return the scalar output
  if (num_dims_out == 0) {
    outputs[0] = model_out.release();
    return absl::OkStatus();
  }

  // Else, reduce sum the output to get a scalar

  // Will sum all dimensions, so get a Tensor containing [0,...,num_dims_out-1].
  AbstractTensorHandlePtr sum_dims;
  {
    vector<int32_t> vals(num_dims_out);

 *
 * <p>This interface is now a legacy type. Use {@code java.util.function.Supplier} (or the
 * appropriate primitive specialization such as {@code IntSupplier}) instead whenever possible.
 * Otherwise, at least reduce <i>explicit</i> dependencies on this type by using lambda expressions
 * or method references instead of classes, leaving your code easier to migrate in the future.
 *