jouleclaw/jouleclaw-rs/Cargo.lock

@@ -4022,7 +4022,6 @@ name = "jouleclaw-jcs"
 version = "0.1.0"
 dependencies = [
  "blake3",
- "ryu",
  "serde",
  "serde_json",
  "thiserror 2.0.20",

jouleclaw/jouleclaw-rs/crates/jouleclaw-jcs/Cargo.toml

@@ -13,4 +13,3 @@ serde      = { workspace = true }
 serde_json = { workspace = true, features = ["preserve_order"] }
 thiserror  = { workspace = true }
 blake3     = { workspace = true }
-ryu        = "1.0"

jouleclaw/jouleclaw-rs/crates/jouleclaw-jcs/src/lib.rs

@@ -13,9 +13,13 @@
 //!
 //! ## Algorithm (per RFC 8785)
 //!
-//! 1. **Numbers** — finite floats serialized per ECMA-262 ToString
-//!    (shortest round-trip via `ryu`); integers within safe range
-//!    use plain integer formatting; non-finite floats are an error.
+//! 1. **Numbers** — every number is the IEEE-754 double it reads
+//!    as, written as ECMA-262's Number::toString writes it: the
+//!    fewest digits that read back as the double (the nearest of
+//!    those, a tie going to the even), in plain decimals from 10⁻⁶
+//!    up to 10²¹ and in exponent form outside. An integer past
+//!    ±(2⁵³ − 1), which a double cannot hold exactly, has no
+//!    canonical form and is an error, as is a non-finite number.
 //! 2. **Strings** — escape `"`, `\`, control chars (with shortcuts
 //!    for `\b \t \n \f \r`), `\u00XX` for other U+0000–U+001F;
 //!    everything else emitted as UTF-8.
@@ -45,9 +49,10 @@ pub enum JcsError {
     /// these values.
     #[error("non-finite number in JCS input")]
     NonFiniteNumber,
-    /// A number with no representable form. Should never happen
-    /// with `serde_json::Number` but covers the unreachable arm.
-    #[error("unrepresentable number")]
+    /// An integer past ±(2⁵³ − 1): a double cannot hold it exactly,
+    /// so two different integers would share one canonical text
+    /// (I-JSON, RFC 7493 §2.2).
+    #[error("unrepresentable number: an integer past ±(2^53 − 1)")]
     UnrepresentableNumber,
 }
 
@@ -119,34 +124,125 @@ fn write_value(v: &serde_json::Value, out: &mut Vec<u8>) -> Result<(), JcsError>
 }
 
 fn write_number(n: &serde_json::Number, out: &mut Vec<u8>) -> Result<(), JcsError> {
-    if let Some(i) = n.as_i64() {
-        // Integer in safe range — direct formatting.
-        let s = format!("{}", i);
-        out.extend_from_slice(s.as_bytes());
-        return Ok(());
-    }
-    if let Some(u) = n.as_u64() {
-        let s = format!("{}", u);
-        out.extend_from_slice(s.as_bytes());
-        return Ok(());
-    }
-    if let Some(f) = n.as_f64() {
-        if !f.is_finite() {
-            return Err(JcsError::NonFiniteNumber);
+    const SAFE: u64 = (1 << 53) - 1;
+    let x = if let Some(i) = n.as_i64() {
+        if i.unsigned_abs() > SAFE {
+            return Err(JcsError::UnrepresentableNumber);
         }
-        // Integer-valued floats serialize as integers (RFC 8785
-        // / ECMA-262 ToString); other floats use ryu shortest
-        // round-trip.
-        if f.fract() == 0.0 && f.abs() < 1e16 {
-            let s = format!("{}", f as i64);
-            out.extend_from_slice(s.as_bytes());
-        } else {
-            let mut buf = ryu::Buffer::new();
-            out.extend_from_slice(buf.format(f).as_bytes());
+        i as f64
+    } else if let Some(u) = n.as_u64() {
+        if u > SAFE {
+            return Err(JcsError::UnrepresentableNumber);
         }
-        return Ok(());
+        u as f64
+    } else {
+        n.as_f64().ok_or(JcsError::UnrepresentableNumber)?
+    };
+    if !x.is_finite() {
+        return Err(JcsError::NonFiniteNumber);
     }
-    Err(JcsError::UnrepresentableNumber)
+    out.extend_from_slice(es_number(x).as_bytes());
+    Ok(())
+}
+
+/// A finite double as ECMA-262's Number::toString writes it (6.1.6.1.20).
+fn es_number(x: f64) -> String {
+    if x == 0.0 {
+        return "0".into();
+    }
+    let (s, n) = shortest(x.abs());
+    let digits: String = s.iter().map(|d| char::from(b'0' + d)).collect();
+    let k = s.len() as i32;
+    let mut out = String::new();
+    if x < 0.0 {
+        out.push('-');
+    }
+    if k <= n && n <= 21 {
+        out.push_str(&digits);
+        out.extend(std::iter::repeat_n('0', (n - k) as usize));
+    } else if 0 < n && n <= 21 {
+        out.push_str(&digits[..n as usize]);
+        out.push('.');
+        out.push_str(&digits[n as usize..]);
+    } else if -6 < n && n <= 0 {
+        out.push_str("0.");
+        out.extend(std::iter::repeat_n('0', (-n) as usize));
+        out.push_str(&digits);
+    } else {
+        out.push_str(&digits[..1]);
+        if k > 1 {
+            out.push('.');
+            out.push_str(&digits[1..]);
+        }
+        let _ = write!(out, "e{}{}", if n - 1 < 0 { '-' } else { '+' }, (n - 1).abs());
+    }
+    out
+}
+
+/// The digits s and the exponent n, x = 0.s × 10ⁿ, that ECMA-262
+/// chooses for `x` > 0: the fewest that read back as `x`, and of those
+/// the nearest to `x`, a tie going to the even.
+fn shortest(x: f64) -> (Vec<u8>, i32) {
+    let e = format!("{x:e}");
+    let (mant, exp) = e.split_once('e').unwrap_or((&e, "0"));
+    let mut s: Vec<u8> = mant.bytes().filter(u8::is_ascii_digit).map(|b| b - b'0').collect();
+    while s.len() > 1 && s.last() == Some(&0) {
+        s.pop();
+    }
+    let n = exp.parse::<i32>().unwrap_or(0) + 1;
+    // Rust's digits are the fewest, but a tie between two equally near
+    // ones rounds up (RFC 8785's 1424953923781206.25). Only when another
+    // of the same length also reads back as x (near 17 digits) is there a
+    // choice: then take x's exact expansion (a double's has at most 767
+    // significant digits) rounded to that length.
+    if ![false, true].into_iter().any(|up| step(&s, up).is_some_and(|t| reads_as(&t, n, x))) {
+        return (s, n);
+    }
+    let k = s.len();
+    let exact = format!("{x:.800e}");
+    let (em, ee) = exact.split_once('e').unwrap_or((&exact, "0"));
+    let all: Vec<u8> = em.bytes().filter(u8::is_ascii_digit).map(|b| b - b'0').collect();
+    let mut cn = ee.parse::<i32>().unwrap_or(0) + 1;
+    let mut c = all[..k].to_vec();
+    let rest = &all[k..];
+    let up = match rest.first().copied().unwrap_or(0) {
+        d if d > 5 => true,
+        5 => rest[1..].iter().any(|&d| d != 0) || c[k - 1] % 2 == 1,
+        _ => false,
+    };
+    if up {
+        match step(&c, true) {
+            Some(t) => c = t,
+            None => {
+                c = std::iter::once(1).chain(std::iter::repeat_n(0, k - 1)).collect();
+                cn += 1;
+            }
+        }
+    }
+    if reads_as(&c, cn, x) { (c, cn) } else { (s, n) }
+}
+
+/// The decimal one unit above or below `s` in its last digit, if it
+/// still has as many digits.
+fn step(s: &[u8], up: bool) -> Option<Vec<u8>> {
+    let mut t = s.to_vec();
+    for i in (0..t.len()).rev() {
+        match (up, t[i]) {
+            (true, 9) => t[i] = 0,
+            (false, 0) => t[i] = 9,
+            (_, d) => {
+                t[i] = if up { d + 1 } else { d - 1 };
+                return (t[0] != 0).then_some(t);
+            }
+        }
+    }
+    None
+}
+
+/// Whether 0.s × 10ⁿ reads as `x` (std's reader is correctly rounded).
+fn reads_as(s: &[u8], n: i32, x: f64) -> bool {
+    let digits: String = s.iter().map(|d| char::from(b'0' + d)).collect();
+    format!("0.{digits}e{n}").parse::<f64>().is_ok_and(|y| y == x)
 }
 
 fn write_string(s: &str, out: &mut Vec<u8>) {
@@ -214,12 +310,51 @@ mod tests {
 
     #[test]
     fn floats_use_shortest_round_trip() {
-        // ryu shortest = "1.5"
         assert_eq!(enc(json!(1.5)), "1.5");
-        // ryu shortest = "0.1"
         assert_eq!(enc(json!(0.1)), "0.1");
     }
 
+    #[test]
+    fn numbers_are_laid_out_as_ecmascript_lays_them_out() {
+        // Plain decimals from 10^-6 up to 10^21, exponent form outside.
+        for (x, want) in [
+            (0.000001, "0.000001"),
+            (0.0000015, "0.0000015"),
+            (1e-7, "1e-7"),
+            (1.5e-5, "0.000015"),
+            (1e16, "10000000000000000"),
+            (1e20, "100000000000000000000"),
+            (1e21, "1e+21"),
+            (-0.0, "0"),
+            (5e-324, "5e-324"),
+        ] {
+            assert_eq!(enc(json!(x)), want, "{x:e}");
+        }
+    }
+
+    #[test]
+    fn numbers_match_rfc_8785_appendix_b_and_a_javascript_engine() {
+        // Bit patterns and what a JavaScript engine's Number::toString
+        // writes for each: RFC 8785's appendix B, the edges of each
+        // layout, and patterns spread over every exponent.
+        let mut n = 0;
+        for line in include_str!("../testdata/es-number-tostring.txt").lines() {
+            let (bits, want) = line.split_once(' ').unwrap();
+            let x = f64::from_bits(u64::from_str_radix(bits, 16).unwrap());
+            assert_eq!(es_number(x), want, "{bits}");
+            n += 1;
+        }
+        assert!(n > 3000);
+    }
+
+    #[test]
+    fn integers_a_double_cannot_hold_have_no_canonical_form() {
+        assert_eq!(enc(json!(-9_007_199_254_740_991_i64)), "-9007199254740991");
+        for v in [json!(9_007_199_254_740_992_u64), json!(-9_007_199_254_740_992_i64), json!(u64::MAX)] {
+            assert!(matches!(canonicalize(&v), Err(JcsError::UnrepresentableNumber)), "{v}");
+        }
+    }
+
     #[test]
     fn non_finite_number_errors() {
         // serde_json can't construct NaN/Inf directly, so we test

sandbox/SPEC.md

@@ -87,7 +87,7 @@ A signature over an object binds its content address, not the reverse (§7.3).
 
 ## 4. Canonical serialization
 
-The canonical form of receipt and attestation cores is **deterministic CBOR** (RFC 8949 §4.2 core deterministic encoding) for the JCR-1 receipt, and **JCS** (RFC 8785) for the JSON attestation. A conforming implementation **MUST** produce byte-identical canonical encodings for equal objects across platforms. **Floats MUST NOT appear in any canonical form or decision path.** All energy figures are integer microjoules; all budgets, counts, and limits are unsigned integers.
+The canonical form of receipt and attestation cores is **deterministic CBOR** (RFC 8949 §4.2 core deterministic encoding) for the JCR-1 receipt, and **JCS** (RFC 8785) for the JSON attestation. A conforming implementation **MUST** produce byte-identical canonical encodings for equal objects across platforms. **Floats MUST NOT appear in any canonical form or decision path.** All energy figures are integer microjoules; all budgets, counts, and limits are unsigned integers. In the JCS form every integer **MUST** lie within ±(2⁵³ − 1): RFC 8785 holds numbers as IEEE-754 doubles (I-JSON, RFC 7493 §2.2), so a larger one has no encoding that every platform writes alike, and an implementation **MUST** refuse a core that holds one.
 
 ---
 

sandbox/sandbox-rs/crates/sandbox-attest/src/fleet.rs

@@ -129,7 +129,7 @@ impl FleetRecord {
         let mut idx = index;
         let mut siblings = Vec::new();
         while level.len() > 1 {
-            let sib = if idx % 2 == 0 {
+            let sib = if idx.is_multiple_of(2) {
                 // sibling on the right (or self, duplicated, at an odd end)
                 (*level.get(idx + 1).unwrap_or(&level[idx]), true)
             } else {

sandbox/sandbox-rs/crates/sandbox-attest/src/jcs.rs

@@ -4,7 +4,10 @@
 //! the same primitive the ARL Sandbox uses, so the two interoperate at the
 //! attestation layer (SPEC §7.3). Our cores contain only objects, arrays,
 //! strings, and unsigned integers; **no floats** appear (SPEC §4), which keeps
-//! number canonicalization trivial (integers serialize verbatim).
+//! number canonicalization trivial: an integer within ±(2^53 − 1) is written as
+//! its digits, which is what RFC 8785 writes for it on every platform. A larger
+//! one is refused: RFC 8785 holds numbers as doubles (I-JSON, RFC 7493 §2.2), so
+//! an implementation that reads it as one would write other bytes.
 
 use serde_json::Value;
 
@@ -13,6 +16,8 @@ use serde_json::Value;
 pub enum JcsError {
     #[error("floats are not permitted in a canonical attestation core (SPEC §4)")]
     FloatNotPermitted,
+    #[error("an integer past ±(2^53 − 1) has no canonical form that every platform writes alike (SPEC §4)")]
+    IntegerOutOfRange,
 }
 
 /// Canonicalize a JSON value to its RFC 8785 bytes. Object keys are sorted by
@@ -31,6 +36,10 @@ fn write_value(v: &Value, out: &mut String) -> Result<(), JcsError> {
             if n.is_f64() {
                 return Err(JcsError::FloatNotPermitted);
             }
+            const SAFE: u64 = (1 << 53) - 1;
+            if n.as_u64().is_some_and(|u| u > SAFE) || n.as_i64().is_some_and(|i| i.unsigned_abs() > SAFE) {
+                return Err(JcsError::IntegerOutOfRange);
+            }
             out.push_str(&n.to_string());
         }
         Value::String(s) => write_string(s, out),
@@ -97,14 +106,21 @@ mod tests {
 
     #[test]
     fn integers_serialize_verbatim() {
-        let v = json!({ "n": 18446744073709551615u64 });
+        let v = json!({ "n": 9007199254740991u64, "m": -9007199254740991i64 });
         let bytes = canonicalize(&v).unwrap();
         assert_eq!(
             String::from_utf8(bytes).unwrap(),
-            r#"{"n":18446744073709551615}"#
+            r#"{"m":-9007199254740991,"n":9007199254740991}"#
         );
     }
 
+    #[test]
+    fn integers_a_double_cannot_hold_are_refused() {
+        for v in [json!({ "n": 9007199254740992u64 }), json!({ "n": u64::MAX }), json!([-9007199254740992i64])] {
+            assert!(matches!(canonicalize(&v), Err(JcsError::IntegerOutOfRange)), "{v}");
+        }
+    }
+
     #[test]
     fn floats_are_rejected() {
         let v = json!({ "x": 1.5 });