and test another isolated pipeline from any branch. 

We'll explain everything via an example. Let's say you make some changes on your branch `myTestBranch`
(we highly recommend to name this branch without prefix and hyphen (`-`) because it's used to generate build type ID) and want to
test these changes without affecting `master`/`release` pipeline. Here are the instructions.

all-container-registry: $(addprefix sub-container-registry-,$(ALL_OS_ARCH))

# split words on hyphen, access by 1-index
word-hyphen = $(word $2,$(subst -, ,$1))
sub-container-%:
	$(MAKE) OUTPUT_TYPE=$(call word-hyphen,$*,1) OS=$(call word-hyphen,$*,2) ARCH=$(call word-hyphen,$*,3) OSVERSION=$(call word-hyphen,$*,4) container

build: $(foreach binary, ${BIN}, bin/${binary}-${OS}-${ARCH})

push: build

# split words on hyphen, access by 1-index
word-hyphen = $(word $2,$(subst -, ,$1))

sub-build-%:
	$(MAKE) OUTPUT_TYPE=docker OS=$(call word-hyphen,$*,1) ARCH=$(call word-hyphen,$*,2) build

all-build: $(addprefix sub-build-,$(ALL_OS_ARCH))

sub-push-image-%:

                // Note: throw the failure rather than returning a result object containing the failure, as console failure logging based on the _result_ happens down in the root build scope
                // whereas console failure logging based on the _thrown exception_ happens up outside session scope. It would be better to refactor so that a result can be returned from here
                throw UncheckedException.throwAsUncheckedException(t);

                // Note: throw the failure rather than returning a result object containing the failure, as console failure logging based on the _result_ happens down in the root build scope
                // whereas console failure logging based on the _thrown exception_ happens up outside session scope. It would be better to refactor so that a result can be returned from here
                throw UncheckedException.throwAsUncheckedException(t);

	if method, ok := f.Interface().(func([]byte) (Stt, error)); ok {
		s, e := method(nil)
		// The bug in issue27695 happens here, during the return
		// from the above call (at the end of reflect.callMethod
		// when preparing to return). The result value that
		// is assigned to s was not being scanned if GC happens
		// to occur there.
		i := interface{}(s)
		return i, e
	}
	return nil, nil

//
// If `initial` is non-nil then that value is Set at creation time.
//
// If a `Get` is waiting soon after the channel associated with the
// `doneCtx` becomes selectable (which never happens for the nil
// channel) then `Set(doneVal)` effectively happens at that time.
func NewWriteOnce(initial interface{}, doneCtx context.Context, doneVal interface{}) WriteOnce {
	p := &promise{
		doneCtx: doneCtx,
		doneVal: doneVal,

  // CHECK-NEXT: "tf.AssignVariableOp"(%[[HASH_TABLE]], %[[CST]])
}

// -----


// Nothing happens when there are no `tf_saved_model.session_initializer`.

// CHECK-LABEL: module
module attributes {tf_saved_model.semantics} {
}

// -----

// Nothing happens when there are no initializer functions.

// CHECK-LABEL: module
module attributes {tf_saved_model.semantics} {

        } catch (Throwable t) {
            // Note: throw the failure rather than returning a result object containing the failure, as console failure logging based on the _result_ happens down in the root build scope
            // whereas console failure logging based on the _thrown exception_ happens up outside session scope. It would be better to refactor so that a result can be returned from here
            notifyEnterprisePluginManager(Result.failed(t));

     * {@link org.gradle.api.java.archives.Manifest} or a file path as interpreted by {@link org.gradle.api.Project#file(Object)}.
     *
     * The merge is not happening instantaneously. It happens either before writing or when {@link #getEffectiveManifest()}
     * is called.
     *
     * @return this
     */
    Manifest from(Object... mergePath);

    /**