In a very theoretical world, an application programmer could rely on a
DNS lookup to return the best (and only) address of a remote host, and
could then pass that address directly to the network socket interface
without further ado. Unfortunately the real world is not that simple.
Even without considering the version number, there are several types of
IP address, and a DNS lookup may return a variety of addresses. In most
cases, applications will use the function getaddrinfo() ("get address
information") to obtain a list of valid addresses, typically containing
both IPv6 and IPv4 addresses. Which is the best one to use, and should
the program try more than one?
We do not go into this subject in detail, because this book is not aimed
primarily at application programmers. However, operators need to be
aware that the default behavior of most applications is simply to use
the first address returned by getaddrinfo(). Some applications (such
as web browsers) may use a smarter approach known as "happy eyeballs"
(RFC 8305) by means of a
heuristic to detect which address gives the fastest response. However,
operators need to understand the various address types in order to
configure systems optimally, including the getaddrinfo() precedence
table (RFC 6724) in every
host.
When developing IPv6 enabled applications, keep in mind that IPv6
addresses are longer and look different than IPv4 addresses. This may
sound obvious, but the past has shown that these are two of the most
common problems, especially when you store IPv6 addresses in a database
or have an existing input field in your application that is too small.
Also, regular expressions for validating IP addresses are different. As
you will learn later in this book there are different types of IPv6
addresses and several ways to write them. Make sure your application
only accepts the correct type of addresses and is also not too strict by
only accepting one format. Users want to use copy-and-paste or
automation and the input format of an IP address may not always be what
your application expects. Always remember: "Be conservative in what you
do, be liberal in what you accept from others". And it's probably always
a good idea not to reinvent the wheel but use library functions that
your programming language of choice provides, e.g. the ipaddress
module for Python. And please don't hard-code IP addresses of any kind
in your code. Always make them configurable and if possible use FQDNs
(DNS names) instead of IP addresses.
Address types are discussed further in 2. Addresses. Address selection is discussed here. How applications relate to a mixture of IPv4 and IPv6 addresses is also discussed in 3. Dual stack scenarios.