- Build etcd image v3.5.13 ([#124026](https://github.com/kubernetes/kubernetes/pull/124026), [@liangyuanpeng](https://github.com/liangyuanpeng)) [SIG API Machinery and Etcd]
- Build etcd image v3.5.14 ([#125235](https://github.com/kubernetes/kubernetes/pull/125235), [@humblec](https://github.com/humblec)) [SIG API Machinery]

- Update build images to Debian 11 (Bullseye)
  - debian-base:bullseye-v1.0.0
  - debian-iptables:bullseye-v1.0.0
  - go-runner:v2.3.1-go1.17.1-bullseye.0
  - kube-cross:v1.23.0-go1.17.1-bullseye.1
  - setcap:bullseye-v1.0.0
  - cluster/images/etcd: Build 3.5.0-2 image

  }

  public void testPrecomputedOptimizations() {
    // These are testing behavior that's never promised by the API.
    // Some matchers are so efficient that it is a waste of effort to
    // build a precomputed version.
    CharMatcher m1 = is('x');
    assertSame(m1, m1.precomputed());
    assertEquals(m1.toString(), m1.precomputed().toString());

    CharMatcher m2 = anyOf("Az");

///

### <a href="https://www.starlette.io/" class="external-link" target="_blank">Starlette</a>

Starlette — це легкий фреймворк/набір інструментів <abbr title="The new standard for build asynchronous Python web">ASGI</abbr>, який ідеально підходить для створення високопродуктивних asyncio сервісів.

		r.fileList = make([]fileListEntry, 0, len(r.File))
		// files and knownDirs map from a file/directory name
		// to an index into the r.fileList entry that we are
		// building. They are used to mark duplicate entries.
		files := make(map[string]int)
		knownDirs := make(map[string]int)

		// dirs[name] is true if name is known to be a directory,
		// because it appears as a prefix in a path.

	"cmd/asm/internal/lex"
	"cmd/internal/obj"
	"cmd/internal/obj/ppc64"
	"cmd/internal/obj/riscv"
	"cmd/internal/obj/x86"
	"cmd/internal/sys"
)

// TODO: configure the architecture

var testOut *strings.Builder // Gathers output when testing.

// append adds the Prog to the end of the program-thus-far.
// If doLabel is set, it also defines the labels collect for this Prog.

func (b *Reader) ReadString(delim byte) (string, error) {
	full, frag, n, err := b.collectFragments(delim)
	// Allocate new buffer to hold the full pieces and the fragment.
	var buf strings.Builder
	buf.Grow(n)
	// Copy full pieces and fragment in.
	for _, fb := range full {
		buf.Write(fb)
	}
	buf.Write(frag)
	return buf.String(), err
}

// WriteTo implements io.WriterTo.

## Changelog since v1.18.0

## Changes by Kind

### Feature

- deps: Update to Golang 1.13.9
  - build: Remove kube-cross image building ([#89398](https://github.com/kubernetes/kubernetes/pull/89398), [@justaugustus](https://github.com/justaugustus)) [SIG Release and Testing]

### Other (Bug, Cleanup or Flake)

* **Deep Learning**: this is a sub-field of Machine Learning, so, the same applies. It's just that there is not a single spreadsheet of numbers to multiply, but a huge set of them, and in many cases, you use a special processor to build and / or use those models.

### Concurrency + Parallelism: Web + Machine Learning { #concurrency-parallelism-web-machine-learning }

   * could perform a cast to LegacyConverter, but we can't because it's an internal-only class.
   *
   * TODO(cpovirk): So make it part of the open-source build, albeit package-private there?
   *
   * So we use uncheckedCastNullableTToT here. This is a weird usage of that method: The method is