Jcs.Net 0.1.0

Jcs.NET

RFC 8785 JSON Canonicalization Scheme (JCS) for .NET. Deterministic JSON serialization for hashing and signing. Zero external dependencies.

Two JSON documents can be semantically identical and byte-for-byte different: property order, whitespace, 1E30 vs 1e+30, é vs é. The moment you hash or sign JSON, those differences break everything. RFC 8785 fixes this by defining a single canonical byte sequence for any JSON document. The JavaScript implementation (canonicalize) is downloaded millions of times a month; on NuGet there has been no maintained, RFC-verified package. Jcs.NET is that implementation: strict, test-vector-verified, and dependency-free.

Where you need it:

• Webhook signature verification (HMAC over a canonical payload instead of fragile raw-body comparison)
• JWS/JWT with unencoded or detached payloads
• Verifiable credentials and decentralized identity (JCS is the canonicalization step in several W3C proof suites)
• Audit-trail and ledger hashing, content-addressed storage, deduplication

Install

dotnet add package Jcs.Net

Quickstart

using System;
using Jcs.Net;

string canonical = JsonCanonicalizer.Canonicalize("""
    {
      "numbers": [1E30, 4.50],
      "b": "2",
      "a": "1"
    }
    """);

Console.WriteLine(canonical);
// {"a":"1","b":"2","numbers":[1e+30,4.5]}

The API is a single static class:

Signing a webhook payload

using System;
using System.Security.Cryptography;
using Jcs.Net;

string payload = """{"amount":4500,"currency":"NGN","reference":"inv-1042"}""";
byte[] secret = "webhook-secret"u8.ToArray();

byte[] canonical = JsonCanonicalizer.CanonicalizeToUtf8(payload);
byte[] signature = HMACSHA256.HashData(secret, canonical);

Console.WriteLine(Convert.ToHexString(signature));

The receiver canonicalizes the body it received and recomputes the HMAC. Reordered properties, added whitespace, or re-encoded numbers no longer invalidate the signature, and any semantic change still does.

RFC 8785 compliance

Jcs.NET implements the full specification:

• Object members sorted recursively by the UTF-16 code units of the raw (unescaped) property name (section 3.2.3), including the surrogate-pair ordering cases
• Minimal string escaping (section 3.2.2.2): only \", \\, the short escapes \b \t \n \f \r, and lowercase \u00xx for the remaining control characters; everything else, including non-ASCII, is emitted literally
• ECMAScript Number::toString serialization for numbers (section 3.2.2.3)
• No inter-token whitespace, UTF-8 output (sections 3.2.1, 3.2.4)

The library is strict about input, because a canonicalizer that silently accepts ambiguous data produces hashes you cannot trust. It rejects, with a descriptive JcsException (derived from JsonException):

Verified against the RFC 8785 appendix examples and the official test suite from the RFC author's repository (cyberphone/json-canonicalization), including exact output bytes.

Numbers follow ECMAScript, not .NET

RFC 8785 requires numbers to serialize exactly as JavaScript's Number::toString would. That format differs from every stock .NET format ("R" gives 1E+30, exponent thresholds differ, negative zero survives). Jcs.NET implements the ECMAScript algorithm and verifies it against 10,000 committed test vectors generated from the V8 engine, plus the RFC appendix table. A further one million randomized IEEE 754 bit patterns were verified against V8 during development with zero mismatches.

Samples of what conformance actually means:

Notes and limitations

• Nesting depth is capped at 64 (the System.Text.Json default).
• The JsonElement overloads reject lone surrogates on the JsonDocument.Parse path: System.Text.Json parses such escapes lazily, and when Jcs.NET reads the value or property name the failure is translated into a JcsException. Elements produced by writer paths such as JsonSerializer.SerializeToElement are different: there System.Text.Json has already replaced unpaired surrogates with U+FFFD before Jcs.NET sees the data, so two distinct .NET strings can canonicalize identically. Use the string-input overloads for the strictest rejection semantics. Duplicate detection works on both paths.
• On the string-input path, a surrogate pair split across representations (a literal high surrogate followed by an escaped low surrogate, or the reverse) is rejected: the literal half makes the text invalid UTF-16 on its own, so there is no well-formed reading of such input.
• Numbers with more precision than an IEEE 754 double (money amounts, 64-bit IDs) are rounded by design, as in every JCS implementation. Follow RFC 8785 appendix D: transport such values as JSON strings.
• Property names are compared and sorted as UTF-16 code units, not by any locale or Unicode collation. That is what the RFC requires, so "péché" sorts after "peach" even though French collation says otherwise.
• Input strings are canonicalized as-is; no Unicode normalization is applied (the RFC forbids it).

Roadmap

• netstandard2.0 target for .NET Framework and older runtimes
• Streaming mode writing directly to Utf8JsonWriter / IBufferWriter<byte> to avoid intermediate strings
• Property-based fuzzing against the reference implementations in CI

License

MIT. See LICENSE.