A secondary goal was to enable a smoother on-ramp from "Zero" to "Getting some value".
Historically, Istio had to really be consumed all-or-nothing for things to work as expected.
In particular, an easy answer to "I just want to have mTLS everywhere, then I can think about adopting the rest of service mesh" was desired.

Other documents include:

      checkNotNull(function);
    }
    this.functions = functions;
  }

  /**
   * Constructs a {@code HashCode} from the {@code Hasher} objects of the functions. Each of them
   * has consumed the entire input and they are ready to output a {@code HashCode}. The order of the
   * hashers are the same order as the functions given to the constructor.
   */

                         TF_Status* s);

// Creates a new TF_AbstractFunction from the current tracing states in the
// context. The provided `ctx` is consumed by this API call and deleted.
// The returned TF_AbstractFunction must be deleted by the client,
// TODO(aminim): clarify the contract on the state of the context after this
// call.

- Adding new `global` values is discouraged as a general rule. The only exceptions are values that are frequently and consistently consumed across at least 2 charts (things like image tags, common labels, etc), but these would only be accepted on a strict case-by-case basis.

    // Note that this may be be an encoded "end of data" header.
    return DerHeader(tagClass, tag, constructed, length)
  }

  /**
   * Consume a header and execute [block], which should consume the entire value described by the
   * header. It is an error to not consume a full value in [block].
   */
  internal inline fun <T> read(
    name: String?,
    block: (DerHeader) -> T,
  ): T {

// - such an object exists, but the capacity is unset
// - such an object exists, but the capacity is zero
//
// The producer of these objects can decide which approach is more suitable.
//
// They are consumed by the kube-scheduler when a CSI driver opts into
// capacity-aware scheduling with CSIDriverSpec.StorageCapacity. The scheduler
// compares the MaximumVolumeSize against the requested size of pending volumes

		return true
	}
	// Parenthesized register: R(10).
	return p.arch.RegisterPrefix[name] && p.peek() == '('
}

// atRegisterShift reports whether we are at the start of an ARM shifted register.
// We have consumed the register or R prefix.
func (p *Parser) atRegisterShift() bool {
	// ARM only.
	if !p.arch.InFamily(sys.ARM, sys.ARM64) {
		return false
	}
	// R1<<...
	if lex.IsRegisterShift(p.peek()) {
		return true

                    dr.stripPathConsumed(1 + domain.length() + 1 + root.length());

                    if ( dr.getPathConsumed() > ( path != null ? path.length() : 0 ) ) {
                        log.error("Consumed more than we provided");
                    }

                    link = path != null && dr.getPathConsumed() > 0 ? path.substring(0, dr.getPathConsumed()) : "\\";
                    dr.setLink(link);

  val condition: Condition = lock.newCondition()

  // Internal state is guarded by [lock]. No long-running or potentially blocking operations are
  // performed while the lock is held.

  /** The bytes consumed and acknowledged by the stream. */
  val readBytes: WindowCounter = WindowCounter(id)

  /** The total number of bytes produced by the application. */
  var writeBytesTotal = 0L
    internal set

available, we need to hold that thread until either the bytes arrive, the stream is closed, or a timeout elapses. If we get bytes but there's nobody asking for them, we buffer them. We don't consider bytes as delivered for flow control until they're consumed by the application.

Consider an application streaming a video over http/2. Perhaps the user pauses the video and the application stops reading bytes from this stream. The buffer will fill up, and flow control prevents the server from sending...