Bug #13593: Addrinfo#== behaves oddly

• Status: Rejected
• Assignee: -
• Target version: -
• Backport: 2.2: UNKNOWN, 2.3: UNKNOWN, 2.4: UNKNOWN

[ioquatix (Samuel Williams)]

It appears as if Addrinfo is using object identity. Addrinfo#== doesn't appear to work as one might expect, given the same instantiation:

> irb
2.4.0 :001 > require 'socket'
 => true 
2.4.0 :002 > Addrinfo.tcp('0.0.0.0', 1234) == Addrinfo.tcp('0.0.0.0', 1234)
 => false 
2.4.0 :003 > a = Addrinfo.tcp('0.0.0.0', 1234)
 => #<Addrinfo: 0.0.0.0:1234 TCP> 
2.4.0 :004 > a == a
 => true 
2.4.0 :005 > 

[shyouhei (Shyouhei Urabe)]

Accurate observation. It does not define equality so falls back to Object's definition, which compares identity.

I guess this is not by design; just no one had seriously needed yet.

[valerauko (Balint Erdos)]

• File addrinfo_eql.diff addrinfo_eql.diff added

I wonder if naive comparison like this is advisable?

[akr (Akira Tanaka)]

• Status changed from Open to Feedback

It is difficult compare Addrinfo objects precisely if not impossible.

Using Addrinfo#inspect as addrinfo_eql.diff is fragile.

Addrinfo is basically struct addrinfo which contains struct sockaddr.
Byte-to-byte comparison may not work.
struct sockaddr can have padding which makes byte-to-byte comparison different but
means same address.
Also, ai_family may be AF_UNSPEC which is same meaning to AF_INET or AF_INET6.
I guess there are more concerns.

The critical point is that Ruby can not know all protocols supported by OS.
Since struct sockaddr depends to a protocol, we can not compare all socket address properly.

[akr (Akira Tanaka)]

• Status changed from Feedback to Rejected

Precise definition of comparing Addrinfo is very difficult if not impossible

[ioquatix (Samuel Williams)]

Doing a binary comparison would be acceptable.

[shyouhei (Shyouhei Urabe)]

ioquatix (Samuel Williams) wrote:

Doing a binary comparison would be acceptable.

I don't think so. As Akira already pointed out,

• Addrinfo have "padding" bits which are not stable. Binary comparison might fail at that point.
• Addrinfo can handle arbitrary socket addresses, not only internet things. Which makes it impossible to tell which bit is a padding and which is not.

[ioquatix (Samuel Williams)]

PR here https://github.com/ruby/ruby/pull/2259

[akr (Akira Tanaka)]

What's the use case of this comparison?

[naruse (Yui NARUSE)]

We discussed about this in developer meeting.
This feature has some difficulty in spec for example padding, unknown type (other than AF_INET, AF_INET6, AF_UNIX, and so on), which should it use for the comparison.
Though such difficulty, we can define a spec only for AF_INET and AF_INET6 if there's concrete use case.