* Asynchronous interface to evaluation. It is functionally the same as an XPathExpression, but allows for asynchronous
* execution of evaluation of the expression.
*
+ * <p>
* FIXME: Whether or not the resulting XPathResult can perform blocking calls is up for grabs, but implementations are
* definitely allowed to perform things like on-demand data transformation from foreign object and data models.
*
* {@link XPathResult} at some point it the future. If it fails to match anything, it {@link Future#get()} will
* return {@link Optional#absent()}.
*
+ * <p>
* FIXME: The amount of overhead an implementation can incur on the user as data from the resulting
* {@link XPathResult} is being accessed is left UNDEFINED.
- *
* Specifically, the user is expected to store each result returned directly or indirectly in a local
* variable instead of repeated calls to the result's methods, as these may incur CPU processing overhead.
- *
* Furthermore all method invocations can throw {@link LazyXPathExpressionException}, which the users are
* expected to handle gracefully. RESILIENT USERS ARE EXPECTED TO CATCH {@link LazyXPathExpressionException}
* AND RECOVER IN THE SAME MANNER THEY WOULD IF AN {@link XPathExpressionException} WOULD HAVE BEEN THROWN.
- *
* [ FIXME: would it be appropriate to allow implementations to SneakyThrow {@link XPathExpressionException}
* and not introduce a RuntimeExpcetion ? ]
*