FailNewMounter = "fail-new-mounter"
)

// CommandScript is used to pre-configure a command that will be executed and
// optionally set it's output (stdout and stderr combined) and return code.
type CommandScript struct {
	// Cmd is the command to execute, e.g. "ls"
	Cmd string
	// Args is a slice of arguments to pass to the command, e.g. "-a"
	Args []string

	// the list), they will already have their goroutineProfiled field set to
	// goroutineProfileSatisfied before their state transitions out of _Gdead.
	//
	// Any goroutine that the scheduler tries to execute concurrently with this
	// call will start by adding itself to the profile (before the act of
	// executing can cause any changes in its stack).
	forEachGRace(func(gp1 *g) {

                                 func::FuncOp loop_body_func,
                                 const int num_replicas) {
  // Find all the ops that are to be included in the 'sc_forward' function which
  // will be executed on the SparseCore. Note, 'forward_pass_ops' is initially
  // seeded with ops from the input MLIR graph that have the
  // _embedding_pipelining="forward" attribute which is set by the TF2 Embedding
  // API.
  //

	// are dependent on MSR C001_1029 and CPU generation.
	// LFENCE on Intel does wait for all previous instructions to have executed.
	// Intel recommends MFENCE;LFENCE in its manuals before RDTSC to have all
	// previous instructions executed and all previous loads and stores to globally visible.
	// Using MFENCE;LFENCE here aligns the serializing properties without
	// runtime detection of CPU manufacturer.

    if (op->hasAttr(kXlaOutsideCompilationAttr)) {
      op->removeAttr(
          StringAttr::get(op->getContext(), kXlaOutsideCompilationAttr));
    }
  });
}

// The results of parallel executes is the combination of return values from
// both host and device.
llvm::SmallVector<Type, 4> GetParallelExecuteResultsTypes(
    ArrayRef<Value> return_value_from_host,
    ArrayRef<Value> return_value_from_device) {

        when:
        def unpackedValue = getter(property, getter, someOtherValue())

        then:
        unpackedValue == someValue()
        1 * sideEffect.execute(someValue())
        0 * _

        where:
        setter  | getter
        "set"   | "get"
        "set"   | "getOrNull"
        "set"   | "getOrElse"
        "value" | "get"

	{"a.RegExp*", ".(|)*+?^$[]", 8},
	{dots + dots + dots, " ", -1},
	{"012abcba210", "\xffb", 6},
	{"012\x80bcb\x80210", "\xffb", 7},
	{"0123456\xcf\x80abc", "\xcfb\x80", 10},
}

// Execute f on each test case.  funcName should be the name of f; it's used
// in failure reports.
func runIndexTests(t *testing.T, f func(s, sep string) int, funcName string, testCases []IndexTest) {
	for _, test := range testCases {

[[sec:scripts]]
== Kotlin DSL scripts

Just like the Groovy-based equivalent, the Kotlin DSL is implemented on top of Gradle's Java API.
Everything you can read in a Kotlin DSL script is Kotlin code compiled and executed by Gradle.
Many of the objects, functions and properties you use in your build scripts come from the Gradle API and the APIs of the applied plugins.

    visibility = ["//visibility:public"],
    win_def_file = ":tensorflow_filtered_def_file",
)

# To avoid duplication, check that the C++ or python library does not depend on
# the stream executor cuda plugins. Targets that want to use cuda APIs should
# instead depend on the dummy plugins in @local_tsl//tsl/platform/default/build_config
# and use header only targets.

	{"a.RegExp*", ".(|)*+?^$[]", 8},
	{dots + dots + dots, " ", -1},
	{"012abcba210", "\xffb", 6},
	{"012\x80bcb\x80210", "\xffb", 7},
	{"0123456\xcf\x80abc", "\xcfb\x80", 10},
}

// Execute f on each test case.  funcName should be the name of f; it's used
// in failure reports.
func runIndexTests(t *testing.T, f func(s, sep []byte) int, funcName string, testCases []BinOpTest) {
	for _, test := range testCases {