func TestSmoke(t *testing.T) {
	const src = "if (+foo\t+=..123/***/0.9_0e-0i'a'`raw`\"string\"..f;//$"
	tokens := []token{_If, _Lparen, _Operator, _Name, _AssignOp, _Dot, _Literal, _Literal, _Literal, _Literal, _Literal, _Dot, _Dot, _Name, _Semi, _EOF}

	var got scanner
	got.init(strings.NewReader(src), errh, 0)
	for _, want := range tokens {
		got.next()
		if got.tok != want {

Let's imagine that the consumer requires _instrumented classes_ from the producer, but that this artifact is _not_ the main one.
The producer can expose its instrumented classes by creating a configuration that will "carry" this artifact:

.Declaring an outgoing variant
====

            // then it means we're using the legacy matching, without attributes, and that the metadata
            // we construct should _not_ include the constraints. We keep the constraints in the descriptors
            // because if we actually use attribute matching, we can select the platform variant which
            // does use constraints.

		}
		if k := n.Index[2]; k != nil {
			p.print(_Colon, k)
		}
		p.print(_Rbrack)

	case *AssertExpr:
		p.print(n.X, _Dot, _Lparen, n.Type, _Rparen)

	case *TypeSwitchGuard:
		if n.Lhs != nil {
			p.print(n.Lhs, blank, _Define, blank)
		}
		p.print(n.X, _Dot, _Lparen, _Type, _Rparen)

	case *CallExpr:
		p.print(n.Fun, _Lparen)
		p.printExprList(n.ArgList)
		if n.HasDots {

The key factor is that a dependency should be a "callable".

A "**callable**" in Python is anything that Python can "call" like a function.

So, if you have an object `something` (that might _not_ be a function) and you can "call" it (execute it) like:

```Python
something()
```

or

```Python
something(some_argument, some_keyword_argument="foo")
```

then it is a "callable".

    <class java.clone="deep">
      <name>Plugin</name>
      <version>1.0.0+</version>
      <description>Mapping information for a single plugin within this group.</description>
      <comment>NOTE: plugin version is _NOT_ included here, since it is resolved using a separate algorithm in plugins' artifact.</comment>
      <fields>
        <field>
          <name>name</name>
          <type>String</type>
          <required>true</required>

      context, bundle.saved_object_graph(), &revived_objects));

  // Revive function library functions as concrete functions without captures.
  // This is necessary because object graph functions may refer to functions
  // _not_ in the object graph: A while loop, for example, will create two
  // auxiliary `while_cond` and `while_body` functions that are only present in
  // the graph def function library.
  for (const FunctionDef& function :

 *  We decide that the LHS is a type reference by checking if the left hand
 *  side is a (qualified) name, and, in case it _is_, resolving that name.
 *
 *  If it resolves to a non-class declaration, it does _not_ refer to a type.
 */
private fun doesDoubleColonUseLHS(lhs: PsiElement): Boolean {
    val reference = when (val inner = lhs.unwrapParenthesesLabelsAndAnnotations()) {
        is KtReferenceExpression ->

                    // Starting here we're going to process the component metadata rules
                    // Therefore metadata can be mutated, and will _not_ be stored in the module metadata cache
                    // but will be in the _in memory_ cache
                    ModuleComponentResolveMetadata processedMetadata = metadataProcessor.processMetadata(resolvedMetadata);

These shared caches could be accessed by versions of Gradle with different settings to retain cache artifacts.

This means that:

- If the retention period is _not_ customized, all versions that perform cleanup will have the same retention periods. There will be no effect due to sharing a Gradle User Home with multiple versions.