Runtime: IPAddress.TryParse() incorrectly parses "::2:3:4:5:6:7:8"

[vyrp]

The method System.Net.IPAddress.TryParse() incorrectly parses "::2:3:4:5:6:7:8".

According to RFC 4291, "The use of "::" indicates _one_ or more groups of 16 bits of zeros" (emphasis mine). Therefore, I would expect "::2:3:4:5:6:7:8" to be valid and the leading "::" to represent a zero.

For example, for the following code:

Console.WriteLine(IPAddress.TryParse("::2:3:4:5:6:7:8", out var addr) ? addr.ToString() : "error");

Actual output:

::

Expected output:

0:2:3:4:5:6:7:8

I find the actual output very strange. In .NET Framework, the result was false, which I don't think is correct, but at least wass less surprising.

Environment:

.NET Core SDK (reflecting any global.json):
 Version:   2.2.107
 Commit:    2212cac826

Runtime Environment:
 OS Name:     Windows
 OS Version:  10.0.18362
 OS Platform: Windows
 RID:         win10-x64
 Base Path:   C:\Program Files\dotnet\sdk\2.2.107\

Host (useful for support):
  Version: 2.2.5
  Commit:  0a3c9209c0

[EgorBo]

Probably fixed by https://github.com/dotnet/corefx/pull/37734

[MarcoRossignoli]

I did some test seem that bug is present on master yet, investigating.
EDIT: found an issue with leading compressor handling...I'll PR a fix.

[davidsh]

I've done some research on whether or not IPv6 addresses such as "::2:3:4:5:6:7" are actually valid IPv6 addresses.

According to WinRT Windows.Networking.Hostname API parsing, it is not valid.

The following code throws an exception "The parameter is incorrect".

```c#
using Windows.Networking;

var host = new HostName("[::2:3:4:5:6:7:8]");
```

Note: It's also interesting that this WinRT API will parse alleged IPv6 addresses without the brackets. But in the case of the string "::2:3:4:5:6:7:8", it will parse into an object but the .Type property says "Bluetooth" and not "IPv6". The WinRT API parses a variety of network addresses including IP and Bluetooth.

I think the premise of the issue raised by dotnet/corefx#38634 is not correct. I think the RFC might have a mis-print in it. I think the intention was that "::" always represented "2 or more" groups of zeros.

I will follow-up after I've discussed with our team's IETF experts for the IPv6 RFC.

[davidsh]

And more curious:

c# var host = new HostName("[1::3:4:5:6:7:8]");

parses successfully as an IPv6 address. So, at least for inner “::” it can be “1 or more” groups of zeroes.

I wonder whether "[::2:3:4:5:6:7:8]" is actually valid (i.e. should result as "[0:2:3:4:5:6:7:8]") but we have a separate bug in the WinRT Windows.Networking.HostName API as well. I'm waiting to hear back from our IETF representative.

[MarcoRossignoli]

"2 or more" groups of zeros.

In that case there is an incorrect handling also for trailing compressor.
Does WinRT throws for 1 group trailing compressor? [1:2:3:4:5:6:7::]

[davidsh]

After discussions with the IETF rep and the Windows team, the RFC is correct. So, the fixes for IPAddress class are correct. It turns out there is a separate bug for the WinRT Windows.Networking.HostName class that will be filed separately.