JSON to C# Class and Record Converter

Every value found at the same position is merged into one description before any C# is written, and what the samples show is what the types promise. The file starts with #nullable enable, so string means never null and string? means may be null whatever the project setting, and the compiler flags code that forgets to check. A key present in every sample is a required property (C# 11): code that builds the object with new must set it, and System.Text.Json, from .NET 7, refuses a document that leaves it out, with "JSON deserialization for type 'Root' was missing required properties". Newtonsoft.Json ignores the keyword, so for it the attribute carries the same rule, Required = Required.Always, or Required.AllowNull where some sample was null. A key missing from some samples is an optional nullable property, such as long? or string?, left null when absent and written back as null.

Integers are long, not int, because a sample says nothing about the largest value still to come. A fraction or an exponent makes a property double, which is binary: 0.1 + 0.2 is 0.30000000000000004, and 19.99 is stored as the nearest double. For money choose decimal, which is base 10 and keeps 19.99, and even the trailing zero of 1.50, exactly; both serializers write 1.50 back. An integer beyond long is decimal, which holds 28 digits. A string property is DateTimeOffset only when every sample is an ISO 8601 date-time with a Z or an offset. DateTimeOffset keeps the offset; DateTime turns 10:00+05:30 into the local time of whichever machine reads it, DateTimeKind.Local, so two servers read two different values from the same JSON. A time with no zone stays string, because both serializers would quietly attach the reader’s own offset, and a date on its own is DateOnly. System.Text.Json refuses a lowercase t or z that Newtonsoft.Json accepts, so only the strict form is typed as a date.

Property names are PascalCase from the key, keeping its own capitals as the .NET naming guidelines do: user_id, user-id and userId all give UserId, and the attribute keeps the original key, [JsonPropertyName("user_id")] or [JsonProperty("user_id")]. The attribute matters more than it looks, because by default System.Text.Json matches names case-sensitively and would never put "user_id" into UserId, while Newtonsoft.Json matches case-insensitively. C# forbids a property with the name of its own class, error CS0542, and JSON produces that easily, as in {"status": {"status": "shipped"}}; such a property is renamed StatusValue, with a warning. So is a name every object already has, such as ToString or Equals, which in a record is a compile error. Keys that give the same name are numbered UserId2, a keyword in the namespace is escaped with @, and a digit at the start gets Field in front. Records are init-only, { get; init; }, rather than positional, so they carry the same attributes and required checks as the classes. Nested objects are separate classes, shared when identical, and a root array has no class of its own: deserialize it as the List<T> the tool shows.