How does an incremental build work? How can you make sure your tasks support running incrementally? Let’s take a look.

[[sec:task_inputs_outputs]]
== Task inputs and outputs

In the most common case, a task takes some inputs and generates some outputs.
We can consider the process of Java compilation as an example of a task.

Both methods return `FileTree` instances that you can then use in the same way as normal file trees.
For example, you can extract some or all of the files of an archive by copying its contents to some directory on the file system.
Or you can merge one archive into another.

Here are some simple examples of creating archive-based file trees:

====

// optimizations run by PartitionedCall can mutate the cluster arbitrarily,
// dropping the kXlaAlreadyClustered attributes from all nodes in the process.
// The correct fix is to use the ConfigProto to pass in some sort of flag into
// the PartitionedCall kernel that tells it to not rerun auto-clustering on the
// cluster.
const char* kXlaAlreadyClustered = "_XlaAlreadyClustered";

class MarkForCompilationPassImpl {
 public:

	const fakeConnectionToken = "X-Fake-Connection-Token"
	const backendResponse = "I am the backend"

	// someConnHeader is some arbitrary header to be declared as a hop-by-hop header
	// in the Request's Connection header.
	const someConnHeader = "X-Some-Conn-Header"

	backend := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
		if c := r.Header.Get("Connection"); c != "" {

//   GTEST_HAS_GLOBAL_STRING  - Define it to 1/0 to indicate that ::string
//                              is/isn't available (some systems define
//                              ::string, which is different to std::string).
//   GTEST_HAS_GLOBAL_WSTRING - Define it to 1/0 to indicate that ::string
//                              is/isn't available (some systems define
//                              ::wstring, which is different to std::wstring).

		// got to the right point in the trace.
		//
		// Note that we also don't advance here if we have a P and we're in a syscall.
		return curCtx, false, nil
	}
	// We can advance this P. Check some invariants.
	//
	// We might have a goroutine if a goroutine is exiting a syscall.
	reqs := event.SchedReqs{Thread: event.MustHave, Proc: event.MustNotHave, Goroutine: event.MayHave}

// has not been met.
//
// The goals are updated after each GC.
//
// Synchronous scavenging happens for one of two reasons: if an allocation would
// exceed the memory limit or whenever the heap grows in size, for some
// definition of heap-growth. The intuition behind this second reason is that the
// application had to grow the heap because existing fragments were not sufficiently

// when not empty.
//
// tls.Config.ServerName (non-empty, set to "example.com") takes
// precedence over "some-other-host.tld" which previously incorrectly
// took precedence. We don't actually connect to (or even resolve)
// "some-other-host.tld", though, because of the Transport.Dial hook.
//
// The httptest.Server has a cert with "example.com" as its name.

	if err != nil {
		return nil, err
	}
	requirements = rs
	return mg, err
}

// expandGraph loads the complete module graph from rs.
//
// If the complete graph reveals that some root of rs is not actually the
// selected version of its path, expandGraph computes a new set of roots that
// are consistent. (With a pruned module graph, this may result in upgrades to

 * `TaskContainer.remove()` now actually removes the given task — some plugins may have accidentally relied on the old behavior.
 * <<#rel4.8:pom_wildcard_exclusions,Gradle now honors implicit wildcards in Maven POM exclusions>>.
 * The Kotlin DSL now respects JSR-305 package annotations.
+
This will lead to some types annotated according to JSR-305 being treated as nullable where they were treated as non-nullable before.