Add snake_case support for System.Text.Json

[hez2010]

Contents below were taken from @khellang, thanks!

API Proposal

Add additional casing support for System.Text.Json, based on this comment.

namespace System.Text.Json;

public partial class JsonNamingPolicy
{
    public static JsonNamingPolicy CamelCase { get; }
+    public static JsonNamingPolicy SnakeLowerCase { get; }
+    public static JsonNamingPolicy SnakeUpperCase { get; }
+    public static JsonNamingPolicy KebabLowerCase { get; }
+    public static JsonNamingPolicy KebabUpperCase { get; }
}

public enum JsonKnownNamingPolicy
{
    Unspecified = 0,
    CamelCase = 1,
+    SnakeLowerCase = 2,
+    SnakeUpperCase = 3,
+    KebabLowerCase = 4,
+    KebabUpperCase = 5,
}

Behavior Proposal

I propose the same behavior as Newtonsoft.Json, just like the existing camel case behavior. The implementation is here and the tests are here.

Other Comments

I think snake case naming is pretty common, especially in the Ruby world, and should be supported out of the box. GitHub's API is probably the most popular example using this naming convention off the top of my head.

Implements

See: dotnet/corefx#40003

[Gnbrkm41]

If you have any designs in mind, could you update the top post with what the proposed APIs would look like / how commonly used snake case is / etc? Here's what I think is a good example: https://github.com/dotnet/corefx/issues/28933#issue-312415055. This should help people better understand the problem/importance of this issue, and may come up with improvements.

[khellang]

API Proposal

namespace System.Text.Json
{
    public abstract class JsonNamingPolicy
    {
        protected JsonNamingPolicy() { }
        public static JsonNamingPolicy CamelCase { get { throw null; } }
+       public static JsonNamingPolicy SnakeCase { get { throw null; } }
        internal static JsonNamingPolicy Default { get { throw null; } }
        public abstract string ConvertName(string name);
    }

+   internal sealed class JsonSnakeCaseNamingPolicy : JsonNamingPolicy
+   {
+       public override string ConvertName(string name) { throw null; }
+   }
}

Behavior Proposal

I propose the same behavior as Newtonsoft.Json, just like the existing camel case behavior. The implementation is here and the tests are here.

Other Comments

I think snake case naming is pretty common, especially in the Ruby world, and should be supported out of the box. GitHub's API is probably the most popular example using this naming convention off the top of my head.

UPDATE: Looks like dotnet/corefx#40003 implemented more or less this exact proposal.

[hez2010]

@khellang I've pasted your proposal contents to the top post, thanks!

[xsoheilalizadeh]

I've prepared a benchmark that compares other implementation with dotnet/corefx#40003 PR.

BenchmarkDotNet=v0.11.5, OS=Windows 10.0.18362
Intel Core i7-7700HQ CPU 2.80GHz (Kaby Lake), 1 CPU, 8 logical and 4 physical cores
.NET Core SDK=3.0.100-preview8-013656
  [Host]     : .NET Core 3.0.0-preview8-28405-07 (CoreCLR 4.700.19.37902, CoreFX 4.700.19.40503), 64bit RyuJIT
  DefaultJob : .NET Core 3.0.0-preview8-28405-07 (CoreCLR 4.700.19.37902, CoreFX 4.700.19.40503), 64bit RyuJIT

|                          Method |        Mean |      Error |     StdDev | Rank |  Gen 0 | Gen 1 | Gen 2 | Allocated |
|-------------------------------- |------------:|-----------:|-----------:|-----:|-------:|------:|------:|----------:|
|  ToSnakeCaseStringBuilderBySpan |    85.97 ns |  1.0500 ns |  0.9821 ns |    1 | 0.0637 |     - |     - |     200 B |
| ToSnakeCaseNewtonsoftJsonBySpan |    86.35 ns |  1.7816 ns |  1.4877 ns |    1 | 0.0484 |     - |     - |     152 B |
|       ToSnakeCaseNewtonsoftJson |    88.89 ns |  0.8417 ns |  0.7461 ns |    2 | 0.0484 |     - |     - |     152 B |
|               ToSnakeCaseBySpan |   122.01 ns |  1.9066 ns |  1.6902 ns |    3 | 0.0253 |     - |     - |      80 B |
|                 ToSnakeCaseLinq |   365.75 ns |  4.4839 ns |  3.5008 ns |    4 | 0.1450 |     - |     - |     456 B |
|                ToSnakeCaseRegex | 2,108.71 ns | 40.6829 ns | 31.7625 ns |    5 | 

[xsoheilalizadeh]

I investigated a couple of weeks about this case and tried other approaches + dotnet/corefx#40003 that have better performance to having snake case naming policy. Finally, I got the best result and it have less allocation and fast execution. My last implementation is ToSnakeCaseBySpan . Shoud I open a new PR for this?

|                          Method |        Mean |      Error |     StdDev | Rank |  Gen 0 | Gen 1 | Gen 2 | Allocated |
|-------------------------------- |------------:|-----------:|-----------:|-----:|-------:|------:|------:|----------:|
|               ToSnakeCaseBySpan |    27.47 ns |  0.4629 ns |  0.4330 ns |    1 | 0.0153 |     - |     - |      48 B |
|  ToSnakeCaseStringBuilderBySpan |    85.23 ns |  1.6495 ns |  1.3774 ns |    2 | 0.0637 |     - |     - |     200 B |
|       ToSnakeCaseNewtonsoftJson |    85.72 ns |  1.6418 ns |  1.4554 ns |    2 | 0.0484 |     - |     - |     152 B |
| ToSnakeCaseNewtonsoftJsonBySpan |    86.96 ns |  1.7060 ns |  1.5958 ns |    2 | 0.0484 |     - |     - |     152 B |
|                 ToSnakeCaseLinq |   353.42 ns |  3.9670 ns |  3.7108 ns |    3 | 0.1450 |     - |     - |     456 B |
|                ToSnakeCaseRegex | 2,056.69 ns | 29.5694 ns | 26.2125 ns |    4 | 0.1526 |     - |     - |     496 B |

\cc @karelz

[hez2010]

@xsoheilalizadeh I think the name of a property is usually not too long, so maybe you can use stackalloc to reduce memory allocation on heap to reduce GC pressure? (It may achieve zero allocation)

public unsafe ReadOnlySpan<char> ToSnakeCaseBySpan()
{
    ...
    Span<char> buffer = stackalloc char[bufferSize];
    ...
}

[xsoheilalizadeh]

Nice suggestion, but here is a question how I should return buffer, I've got following:

Cannot use local 'buffer' in this context because it may expose referenced variables outside of their declaration scope

[Gnbrkm41]

Should I open a new PR for this?

Nope, not yet 😅 not until the API is approved.

[hez2010]

Because buffer now is allocated on the stack, you cannot return it directly since it will be disposed after return. To solve it, you can simply change `return buffer` to `return buffer.ToString()`; Sent from Mail<https://go.microsoft.com/fwlink/?LinkId=550986> for Windows 10

…

________________________________ From: Soheil Alizadeh <notifications@github.com> Sent: Friday, September 13, 2019 3:48:38 PM To: dotnet/corefx <corefx@noreply.github.com> Cc: Steve <hez2010@outlook.com>; Author <author@noreply.github.com> Subject: Re: [dotnet/corefx] Add UnderScoreCase support for System.Text.Json (#39564) Nice suggestion, but here is a question how I should return buffer, I've got following: Cannot use local 'buffer' in this context because it may expose referenced variables outside of their declaration scope — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<https://eur04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fdotnet%2Fcorefx%2Fissues%2F39564%3Femail_source%3Dnotifications%26email_token%3DADSENWIDJVGBMFT3T4V3U5LQJNAVNA5CNFSM4IEL7W4KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD6UIBGI%23issuecomment-531136665&data=02%7C01%7C%7C68574ff9a79b43fbaf2a08d7381ecaaf%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637039577204910801&sdata=PhAc8%2BvEwr9AHgXI1vNdlLENvuKw1jDC6x%2F%2FNuc7PDk%3D&reserved=0>, or mute the thread<https://eur04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FADSENWIHQJ7DC64RZH2RQFDQJNAVNANCNFSM4IEL7W4A&data=02%7C01%7C%7C68574ff9a79b43fbaf2a08d7381ecaaf%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637039577204920805&sdata=M9kuqi6W4P3Msx76loMm8q2laC%2FrNgb4ZwIozAwcDZ4%3D&reserved=0>.

[karelz]

@ahsonkhan are we ready to review the API? @xsoheilalizadeh is eager to implement it ...

[xsoheilalizadeh]

I've figured out that the ToSnakeCaseBySpan can't pass these tests, it's just compatible with pascal case properties, I've tried to make it compatible with other formats but unfortunately, I've failed and it'd be unreadable. I was hurrying to contribute and got a bad result, sorry.

[ahsonkhan]

This looks like a reasonable API addition. I was considering how common such formats are to warrant adding support built-in (rather than folks creating their own naming policy to support it), and looks like quite a few places use snake_case including Facebook, AWS, and Twitter APIs (aside from the fact that Json.NET supports it):
https://stackoverflow.com/questions/5543490/json-naming-convention

And of course, what @khellang said:

I think snake case naming is pretty common, especially in the Ruby world, and should be supported out of the box. GitHub's API is probably the most popular example using this naming convention off the top of my head.

cc @steveharter

[terrajobst]

Video

Makes sense. We also considered others (such as kebap casing) but it seems we should defer that until someone actually needs that.

[Gnbrkm41]

@xsoheilalizadeh, are you still working on it? :D you should be able to open PRs for it.

[xsoheilalizadeh]

Unfortunately no till I pass tests, I tried to pass tests but my implementation only support pascal-case naming for example FirstName will be first_name but in other cases like IsCIA it will be Is_c_i_a that isn't expected.