* fresh permits, thus the RateLimiter becomes /slower/ after a period of underutilization.
   *
   * Last, but not least: consider a RateLimiter with rate of 1 permit per second, currently
   * completely unused, and an expensive acquire(100) request comes. It would be nonsensical to just
   * wait for 100 seconds, and /then/ start the actual task. Why wait without doing anything? A much

                " (${descriptionParts.joinToString(separator = ", ")})"
            } else {
                ""
            }
        description =
            "Builds Gradle with the version of Gradle which is currently under development (twice)$descriptionSuffix"

        requirements {
            requiresNotSharedHost()
        }

    /*
     To avoid unnecessary rerun, what we do here is a bit complicated:

	"fmt"
	"hash/crc32"
	"log"
	"os"
	"strings"

	"github.com/dchest/siphash"
	"github.com/google/uuid"
)

// hashes the key returning an integer based on the input algorithm.
// This function currently supports
// - SIPMOD
func sipHashMod(key string, cardinality int, id [16]byte) int {
	if cardinality <= 0 {
		return -1
	}
	// use the faster version as per siphash docs
	// https://github.com/dchest/siphash#usage

     */
    private static int getAccessibleMethods(Class<?> clazz, MethodInfo[] methodInfos, int upcastCount) {
        int l = methodInfos.length;

        // if this class is public, then check each of the currently
        // 'non-upcasted' methods to see if we have a match
        if (Modifier.isPublic(clazz.getModifiers())) {
            for (int i = 0; i < l && upcastCount < l; ++i) {
                try {

The build state is managed by the `BuildState` class.
An instance is created for each build in the build definition, once per build execution and is discarded at the end of the execution. 

The build state also includes what is currently called the "Gradle state". The Gradle state is being merged into the build state and is mostly empty.

### Project state

   * suppression externally, but it's OK because we don't enforce test-size rules there.)
   *
   * We'd just use PackageSanityTests directly, saving us from needing this separate type, but we're
   * currently skipping MediumTests on Android, and we skip them by not making them present at
   * runtime at all. I could just make _this_ test a MediumTest, but then it wouldn't run on

class TensorSliceReader;

// A wrapper around BundleReader (for V2 checkpoints) and
// checkpoint::TensorSliceReader (for V1), that is more easily SWIG wrapped for
// other languages.
//
// The class currently only interacts with single-slice (i.e., non-partitioned)
// variables.
class CheckpointReader {
 public:
  CheckpointReader(const std::string& filename, TF_Status* status);

      code == 100 -> true

      // Handle Processing (102) & Early Hints (103) and any new codes without failing
      // 100 and 101 are the exceptions with different meanings
      // But Early Hints not currently exposed
      code in (102 until 200) -> true

      else -> false
    }

  /**
   * Returns a request that includes a credential to satisfy an authentication challenge in
   * [response]. Returns null if the challenge cannot be satisfied.
   *
   * The route is best effort, it currently may not always be provided even when logically
   * available. It may also not be provided when an authenticator is re-used manually in an
   * application interceptor, such as when implementing client-specific retries.
   */

      render: java
    validations:
      required: true

  - type: textarea
    attributes:
      label: Current Behavior
      description: What does the feature currently do?
    validations:
      required: true

  - type: textarea
    attributes:
      label: Desired Behavior
      description: What do you want it to do instead?
    validations:
      required: true