Re: Ip::Address::IsAnyAddr

On 01/08/11 20:52, Tsantilas Christos wrote:
> On 07/29/2011 06:50 PM, Alex Rousskov wrote:
>> On 07/25/2011 08:59 AM, Tsantilas Christos wrote:
>>> I am sending a version 4 of the patch which is the same with Amos patch
>>> but is a little smaller. IT is easier to see the changes.
>>
>>
>>> bool
>>> Ip::Address::IsIPv4() const
>>> {
>>> - return IsAnyAddr() || IsNoAddr() ||
>>> IN6_IS_ADDR_V4MAPPED(&m_SocketAddr.sin6_addr );
>>> + return IN6_IS_ADDR_V4MAPPED(&m_SocketAddr.sin6_addr );
>>> }
>>>
>>> bool
>>> Ip::Address::IsIPv6() const
>>> {
>>> - return IsAnyAddr() || IsNoAddr() ||
>>> !IN6_IS_ADDR_V4MAPPED(&m_SocketAddr.sin6_addr );
>>> + return !IN6_IS_ADDR_V4MAPPED(&m_SocketAddr.sin6_addr );
>>> }
>>
>> Can we rewrite IsIPv6() as "return !IsIPv4()", to clarify the intent if
>> that is indeed the intent?
>>
>> Also, the documentation for these two methods seems to imply a different
>> relationship, at least in some corner cases:
>>
>>> /** Test whether content can be used as an IPv4 address
>>> \retval true if content was received as an IPv4 address
>>> \retval true if content was received as an IPv4-Mapped address
>>> \retval false if content was received as a non-mapped IPv6 native
>>> address.
>>> */
>>> bool IsIPv4() const;
>>>
>>> /** Test whether content can be used as an IPv6 address.
>>> \retval true if --enable-ipv6 has been compiled.
>>> \retval false if --disable-ipv6 has been compiled.
>>> \retval false if --with-ipv6-split-stack has been compiled AND
>>> content is I
>>> Pv4-mapped.
>>> */
>>> bool IsIPv6() const;
>>
>> As far as I can tell, the above definitions make it possible for both
>> IsIPv4() and IsIPv6() to return true at the same time in some cases, but
>> the implementation does not support that. Thus, the docs or the
>> implementation is wrong.
>
> We should rename the IsIPv6() method to IsIpv6Only(), to avoid
> confusion. And then maybe implement an new IsIPv6 method, which for now
> will return always true (my sense is that everything can be an ipv6
> address).
>

"everything is IPv6 address" is the hybrid model. My work from 2007 was
designed around that. The fine details distinctions between the
variations of TCP stack appeared and bit in a bad way. Each stack places
different limitations on what can be converted and used where. In
split-stack and basic dual-stack the v4-mapped addresses are invalid and
cant be called ether v4 or v6. Just an intermediate state that can be
stored in v6 structures and used/converted to v4-only at need.

The IsIPvN() calls both have the implicit "only" needed for split-stack
use at present. This patch is enforcing and polishing that implict
detail to solid, not changing the semantics.

Amos

-- 
Please be using
   Current Stable Squid 2.7.STABLE9 or 3.1.14
   Beta testers wanted for 3.2.0.10