Fix inconsistencies for record types across NET 6.x.x SDK

[sungam3r]

fixes #245

[sungam3r]

@Shane32 3.1 runtime fails all tests, 6.0.400 success on both Windows and Linux!

[sungam3r]

It seems that there are some inconsistencies between CompilerGenerated attributes accross SDKs. Nevermind, I just ignored EqualityContract member to unify behavior. Also NET5 (and even NET 6.0.100) has completely different output from NET6:

 public class User : System.IEquatable<PublicApiGeneratorTests.Examples.User>
    {
        protected User(PublicApiGeneratorTests.Examples.User original) { }
        public User(string login, string password) { }
        public string login { get; set; }
        public string password { get; set; }
        public virtual PublicApiGeneratorTests.Examples.User <Clone>$() { }
        public void Deconstruct(out string login, out string password) { }
        public virtual bool Equals(PublicApiGeneratorTests.Examples.User? other) { }
        public override bool Equals(object? obj) { }
        public override int GetHashCode() { }
        protected virtual bool PrintMembers(System.Text.StringBuilder builder) { }
        public override string ToString() { }
        public static bool operator !=(PublicApiGeneratorTests.Examples.User? left, PublicApiGeneratorTests.Examples.User? right) { }
        public static bool operator ==(PublicApiGeneratorTests.Examples.User? left, PublicApiGeneratorTests.Examples.User? right) { }

vs

public class User : System.IEquatable<PublicApiGeneratorTests.Examples.User>
    {
        public User(string login, string password) { }
        public string login { get; set; }
        public string password { get; set; }
    }

[Shane32]

I think I remember a .NET compiler bug saying that the generated members were not marked as compiler generated or something. I don't have a reference, but chances are the bug was resolved after 6.0.100 and hence the difference.

[sungam3r]

@stakx I will be grateful to you for review/publish the fix.

[stakx]

@sungam3r, I'm about to go on vacation for a few weeks' time. If time permits, I'll try to review this before I'm off.

[Shane32]

@stakx It would be greatly appreciated

[sungam3r]

Also affects EasyNetQ/EasyNetQ#1441

[stakx]

Looks good to me. Just a few comments on some minor points.

[stakx]

Also NET5 (and even NET 6.0.100) has completely different output from NET6

Like @Shane32 I also suspect this may be related to [CompilerGenerated] attribute now emitted. .NET 6 emits record types differently than .NET 5 did (e.g. it emits clone methods using covariant returns), that may be one of those differences. There's likely nothing that can be done about the "superfluous" members appearing for .NET 5.

[sungam3r]

There's likely nothing that can be done about the "superfluous" members appearing for .NET 5.

Yes. Let's consider them as a "normal" parts of public api set.

[sungam3r]

Sorry for delay.

[sungam3r]

@stakx OK to merge?

[sungam3r]

@stakx ?

[danielmarbach]

@sungam3r I could merge this since I still have contributor rights but I can't release it. Let me know

[sungam3r]

As you wish. We have already upgraded our CI to 6.0.402 and 7.0.101.

[stakx]

@sungam3r, my apologies for letting this PR sit around for such a long time. I've lately been unexpectedly busy at work and needed some time off from programming-related activities in my spare time. I'm now finally going through all the PRs that have accumulated since.

Sorry for that again!

[sungam3r]

busy at work and needed some time off from programming-related activities in my spare time

The same thing.

[sungam3r]

@stakx To keep history clean I suggest to use squash option

Current history with merge-commits:

And also this

[stakx]

@sungam3r, I agree that squashing these commits would have resulted in a cleaner commit history, and I was almost going to squash them... but then decided against it at the last minute because I didn't want to risk erasing the commit (co-) authors.

[sungam3r]

I didn't want to risk erasing the commit (co-) authors

This info is preserved. Before merge you can edit final commit message. Usually I just CTRL+A and DELETE all that stuff. As a rule only PR title matters and flows into final commit message... but again - you can change it each time as you want.

[stakx]

I knew that I could edit the final commit message, but I wasn't sure about the co-authorships, which cannot be edited through GitHub's webpage. Good to know it's preserved... thanks for letting me know! 👍

[sungam3r]

Nevertheless, it's better to check twice :).