body.awaitSuccess()
  }

  /**
   * OkHttp currently doesn't implement failing the request body stream independently of failing the
   * corresponding response body stream. This is necessary if we want servers to be able to stop
   * inbound data and send an early 400 before the request body completes.
   *
   * This test sends a slow request that is canceled by the server. It expects the response to still

   * UnsupportedOperationException}; this should only apply to {@code getAnnotatedBounds()}. This
   * does mean that users on Java 8 who obtain an instance of {@code TypeVariable} from {@link
   * TypeResolver#resolveType} will not be able to call {@code getAnnotatedBounds()} on it, but that
   * should hopefully be rare.
   *
   * <p>TODO(b/147144588): We are currently also missing the methods inherited from {@link

// existing ETag entry.
//
// Due to legacy S3 clients, that make incorrect assumptions
// about HTTP headers, Set should be used instead of
// http.Header.Set(...). Otherwise, some S3 clients will not
// able to extract the ETag.
func Set(etag ETag, h http.Header) {
	// Some (broken) S3 clients expect the ETag header to
	// literally "ETag" - not "Etag". Further, some clients

Then I found that there was a standard to document APIs, using JSON (or YAML, an extension of JSON) called Swagger.

And there was a web user interface for Swagger APIs already created. So, being able to generate Swagger documentation for an API would allow using this web user interface automatically.

At some point, Swagger was given to the Linux Foundation, to be renamed OpenAPI.

     * an object type argument, since in reflection
     * primitive type arguments are represented by
     * their object counterparts, so for an argument of
     * type (say) java.lang.Integer, it will not be able
     * to decide between a method that takes int and a
     * method that takes java.lang.Integer as a parameter.
     * <p>
     * This turns out to be a relatively rare case

    if (address.url.host == this.route().address.url.host) {
      return true // This connection is a perfect match.
    }

    // At this point we don't have a hostname match. But we still be able to carry the request if
    // our connection coalescing requirements are met. See also:
    // https://hpbn.co/optimizing-application-delivery/#eliminate-domain-sharding

So, the frontend (that runs in the browser) would try to reach `/openapi.json` and wouldn't be able to get the OpenAPI schema.

Because we have a proxy with a path prefix of `/api/v1` for our app, the frontend needs to fetch the OpenAPI schema at `/api/v1/openapi.json`.

```mermaid
graph LR

browser("Browser")

      tensorflow::unwrap(h);
  // TODO(b/175427838): It would be nice to be able to use tensorflow::isa here.
  if (tensorflow::CustomDeviceTensorHandle::classof(unwrapped_handle)) {
    return tensorflow::down_cast<tensorflow::CustomDeviceTensorHandle*>(
               unwrapped_handle)
        ->DevicePointer();
  }
  // TODO(b/175427838): It would be nice to be able to use tensorflow::isa here.

	// what decoding mechanisms must be applied to obtain the object referenced
	// by the Content-Type header field.
	ContentEncoding string

	// Date and time at which the object is no longer able to be cached
	Expires time.Time

	// Cache-Control - Specifies caching behavior along the request/reply chain
	CacheControl string

	// Specify object storage class
	StorageClass string

## Before you start

Before starting to work on a feature or a bug fix, please open an issue to discuss the use case or bug with us. This can save everyone a lot of time and frustration.

For any non-trivial change, we need to be able to answer these questions:

* Why is this change done? What's the use case?
* For user facing features, what will the API look like?
* What test cases should it have? What could go wrong?