summaryrefslogtreecommitdiffstats
path: root/vendor/der/src/asn1/real.rs
diff options
context:
space:
mode:
Diffstat (limited to 'vendor/der/src/asn1/real.rs')
-rw-r--r--vendor/der/src/asn1/real.rs990
1 files changed, 990 insertions, 0 deletions
diff --git a/vendor/der/src/asn1/real.rs b/vendor/der/src/asn1/real.rs
new file mode 100644
index 0000000..b9f2e67
--- /dev/null
+++ b/vendor/der/src/asn1/real.rs
@@ -0,0 +1,990 @@
+//! ASN.1 `REAL` support.
+
+// TODO(tarcieri): checked arithmetic
+#![allow(
+ clippy::cast_lossless,
+ clippy::cast_sign_loss,
+ clippy::integer_arithmetic
+)]
+
+use crate::{
+ BytesRef, DecodeValue, EncodeValue, FixedTag, Header, Length, Reader, Result, StrRef, Tag,
+ Writer,
+};
+
+use super::integer::uint::strip_leading_zeroes;
+
+impl<'a> DecodeValue<'a> for f64 {
+ fn decode_value<R: Reader<'a>>(reader: &mut R, header: Header) -> Result<Self> {
+ let bytes = BytesRef::decode_value(reader, header)?.as_slice();
+
+ if header.length == Length::ZERO {
+ Ok(0.0)
+ } else if is_nth_bit_one::<7>(bytes) {
+ // Binary encoding from section 8.5.7 applies
+ let sign: u64 = u64::from(is_nth_bit_one::<6>(bytes));
+
+ // Section 8.5.7.2: Check the base -- the DER specs say that only base 2 should be supported in DER
+ let base = mnth_bits_to_u8::<5, 4>(bytes);
+
+ if base != 0 {
+ // Real related error: base is not DER compliant (base encoded in enum)
+ return Err(Tag::Real.value_error());
+ }
+
+ // Section 8.5.7.3
+ let scaling_factor = mnth_bits_to_u8::<3, 2>(bytes);
+
+ // Section 8.5.7.4
+ let mantissa_start;
+ let exponent = match mnth_bits_to_u8::<1, 0>(bytes) {
+ 0 => {
+ mantissa_start = 2;
+ let ebytes = (i16::from_be_bytes([0x0, bytes[1]])).to_be_bytes();
+ u64::from_be_bytes([0x0, 0x0, 0x0, 0x0, 0x0, 0x0, ebytes[0], ebytes[1]])
+ }
+ 1 => {
+ mantissa_start = 3;
+ let ebytes = (i16::from_be_bytes([bytes[1], bytes[2]])).to_be_bytes();
+ u64::from_be_bytes([0x0, 0x0, 0x0, 0x0, 0x0, 0x0, ebytes[0], ebytes[1]])
+ }
+ _ => {
+ // Real related error: encoded exponent cannot be represented on an IEEE-754 double
+ return Err(Tag::Real.value_error());
+ }
+ };
+ // Section 8.5.7.5: Read the remaining bytes for the mantissa
+ let mut n_bytes = [0x0; 8];
+ for (pos, byte) in bytes[mantissa_start..].iter().rev().enumerate() {
+ n_bytes[7 - pos] = *byte;
+ }
+ let n = u64::from_be_bytes(n_bytes);
+ // Multiply byt 2^F corresponds to just a left shift
+ let mantissa = n << scaling_factor;
+ // Create the f64
+ Ok(encode_f64(sign, exponent, mantissa))
+ } else if is_nth_bit_one::<6>(bytes) {
+ // This either a special value, or it's the value minus zero is encoded, section 8.5.9 applies
+ match mnth_bits_to_u8::<1, 0>(bytes) {
+ 0 => Ok(f64::INFINITY),
+ 1 => Ok(f64::NEG_INFINITY),
+ 2 => Ok(f64::NAN),
+ 3 => Ok(-0.0_f64),
+ _ => Err(Tag::Real.value_error()),
+ }
+ } else {
+ let astr = StrRef::from_bytes(&bytes[1..])?;
+ match astr.inner.parse::<f64>() {
+ Ok(val) => Ok(val),
+ // Real related error: encoding not supported or malformed
+ Err(_) => Err(Tag::Real.value_error()),
+ }
+ }
+ }
+}
+
+impl EncodeValue for f64 {
+ fn value_len(&self) -> Result<Length> {
+ if self.is_sign_positive() && (*self) < f64::MIN_POSITIVE {
+ // Zero: positive yet smaller than the minimum positive number
+ Ok(Length::ZERO)
+ } else if self.is_nan()
+ || self.is_infinite()
+ || (self.is_sign_negative() && -self < f64::MIN_POSITIVE)
+ {
+ // NaN, infinite (positive or negative), or negative zero (negative but its negative is less than the min positive number)
+ Ok(Length::ONE)
+ } else {
+ // The length is that of the first octets plus those needed for the exponent plus those needed for the mantissa
+ let (_sign, exponent, mantissa) = decode_f64(*self);
+
+ let exponent_len = if exponent == 0 {
+ // Section 8.5.7.4: there must be at least one octet for exponent encoding
+ // But, if the exponent is zero, it'll be skipped, so we make sure force it to 1
+ Length::ONE
+ } else {
+ let ebytes = exponent.to_be_bytes();
+ Length::try_from(strip_leading_zeroes(&ebytes).len())?
+ };
+
+ let mantissa_len = if mantissa == 0 {
+ Length::ONE
+ } else {
+ let mbytes = mantissa.to_be_bytes();
+ Length::try_from(strip_leading_zeroes(&mbytes).len())?
+ };
+
+ exponent_len + mantissa_len + Length::ONE
+ }
+ }
+
+ fn encode_value(&self, writer: &mut impl Writer) -> Result<()> {
+ // Check if special value
+ // Encode zero first, if it's zero
+ // Special value from section 8.5.9 if non zero
+ if self.is_nan()
+ || self.is_infinite()
+ || (self.is_sign_negative() && -self < f64::MIN_POSITIVE)
+ || (self.is_sign_positive() && (*self) < f64::MIN_POSITIVE)
+ {
+ if self.is_sign_positive() && (*self) < f64::MIN_POSITIVE {
+ // Zero
+ return Ok(());
+ } else if self.is_nan() {
+ // Not a number
+ writer.write_byte(0b0100_0010)?;
+ } else if self.is_infinite() {
+ if self.is_sign_negative() {
+ // Negative infinity
+ writer.write_byte(0b0100_0001)?;
+ } else {
+ // Plus infinity
+ writer.write_byte(0b0100_0000)?;
+ }
+ } else {
+ // Minus zero
+ writer.write_byte(0b0100_0011)?;
+ }
+ } else {
+ // Always use binary encoding, set bit 8 to 1
+ let mut first_byte = 0b1000_0000;
+
+ if self.is_sign_negative() {
+ // Section 8.5.7.1: set bit 7 to 1 if negative
+ first_byte |= 0b0100_0000;
+ }
+
+ // Bits 6 and 5 are set to 0 to specify that binary encoding is used
+ //
+ // NOTE: the scaling factor is only used to align the implicit point of the mantissa.
+ // This is unnecessary in DER because the base is 2, and therefore necessarily aligned.
+ // Therefore, we do not modify the mantissa in anyway after this function call, which
+ // already adds the implicit one of the IEEE 754 representation.
+ let (_sign, exponent, mantissa) = decode_f64(*self);
+
+ // Encode the exponent as two's complement on 16 bits and remove the bias
+ let exponent_bytes = exponent.to_be_bytes();
+ let ebytes = strip_leading_zeroes(&exponent_bytes);
+
+ match ebytes.len() {
+ 0 | 1 => {}
+ 2 => first_byte |= 0b0000_0001,
+ 3 => first_byte |= 0b0000_0010,
+ _ => {
+ // TODO: support multi octet exponent encoding?
+ return Err(Tag::Real.value_error());
+ }
+ }
+
+ writer.write_byte(first_byte)?;
+
+ // Encode both bytes or just the last one, handled by encode_bytes directly
+ // Rust already encodes the data as two's complement, so no further processing is needed
+ writer.write(ebytes)?;
+
+ // Now, encode the mantissa as unsigned binary number
+ let mantissa_bytes = mantissa.to_be_bytes();
+ let mbytes = strip_leading_zeroes(&mantissa_bytes);
+ writer.write(mbytes)?;
+ }
+
+ Ok(())
+ }
+}
+
+impl FixedTag for f64 {
+ const TAG: Tag = Tag::Real;
+}
+
+/// Is the N-th bit 1 in the first octet?
+/// NOTE: this function is zero indexed
+pub(crate) fn is_nth_bit_one<const N: usize>(bytes: &[u8]) -> bool {
+ if N < 8 {
+ bytes
+ .first()
+ .map(|byte| byte & (1 << N) != 0)
+ .unwrap_or(false)
+ } else {
+ false
+ }
+}
+
+/// Convert bits M, N into a u8, in the first octet only
+pub(crate) fn mnth_bits_to_u8<const M: usize, const N: usize>(bytes: &[u8]) -> u8 {
+ let bit_m = is_nth_bit_one::<M>(bytes);
+ let bit_n = is_nth_bit_one::<N>(bytes);
+ (bit_m as u8) << 1 | bit_n as u8
+}
+
+/// Decode an f64 as its sign, exponent, and mantissa in u64 and in that order, using bit shifts and masks.
+/// Note: this function **removes** the 1023 bias from the exponent and adds the implicit 1
+#[allow(clippy::cast_possible_truncation)]
+pub(crate) fn decode_f64(f: f64) -> (u64, u64, u64) {
+ let bits = f.to_bits();
+ let sign = bits >> 63;
+ let exponent = bits >> 52 & 0x7ff;
+ let exponent_bytes_no_bias = (exponent as i16 - 1023).to_be_bytes();
+ let exponent_no_bias = u64::from_be_bytes([
+ 0x0,
+ 0x0,
+ 0x0,
+ 0x0,
+ 0x0,
+ 0x0,
+ exponent_bytes_no_bias[0],
+ exponent_bytes_no_bias[1],
+ ]);
+ let mantissa = bits & 0xfffffffffffff;
+ (sign, exponent_no_bias, mantissa + 1)
+}
+
+/// Encode an f64 from its sign, exponent (**without** the 1023 bias), and (mantissa - 1) using bit shifts as received by ASN1
+pub(crate) fn encode_f64(sign: u64, exponent: u64, mantissa: u64) -> f64 {
+ // Add the bias to the exponent
+ let exponent_with_bias =
+ (i16::from_be_bytes([exponent.to_be_bytes()[6], exponent.to_be_bytes()[7]]) + 1023) as u64;
+ let bits = sign << 63 | exponent_with_bias << 52 | (mantissa - 1);
+ f64::from_bits(bits)
+}
+
+#[cfg(test)]
+mod tests {
+ use crate::{Decode, Encode};
+
+ #[test]
+ fn decode_subnormal() {
+ assert!(f64::from_der(&[0x09, 0x01, 0b0100_0010]).unwrap().is_nan());
+ let plus_infty = f64::from_der(&[0x09, 0x01, 0b0100_0000]).unwrap();
+ assert!(plus_infty.is_infinite() && plus_infty.is_sign_positive());
+ let neg_infty = f64::from_der(&[0x09, 0x01, 0b0100_0001]).unwrap();
+ assert!(neg_infty.is_infinite() && neg_infty.is_sign_negative());
+ let neg_zero = f64::from_der(&[0x09, 0x01, 0b0100_0011]).unwrap();
+ assert!(neg_zero.is_sign_negative() && neg_zero.abs() < f64::EPSILON);
+ }
+
+ #[test]
+ fn encode_subnormal() {
+ // All subnormal fit in three bytes
+ let mut buffer = [0u8; 3];
+ assert_eq!(
+ &[0x09, 0x01, 0b0100_0010],
+ f64::NAN.encode_to_slice(&mut buffer).unwrap()
+ );
+ assert_eq!(
+ &[0x09, 0x01, 0b0100_0000],
+ f64::INFINITY.encode_to_slice(&mut buffer).unwrap()
+ );
+ assert_eq!(
+ &[0x09, 0x01, 0b0100_0001],
+ f64::NEG_INFINITY.encode_to_slice(&mut buffer).unwrap()
+ );
+ assert_eq!(
+ &[0x09, 0x01, 0b0100_0011],
+ (-0.0_f64).encode_to_slice(&mut buffer).unwrap()
+ );
+ }
+
+ #[test]
+ fn encdec_normal() {
+ // The comments correspond to the decoded value from the ASN.1 playground when the bytes are inputed.
+ {
+ // rec1value R ::= 0
+ let val = 0.0;
+ let expected = &[0x09, 0x0];
+ let mut buffer = [0u8; 2];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ // rec1value R ::= { mantissa 1, base 2, exponent 0 }
+ let val = 1.0;
+ let expected = &[0x09, 0x03, 0x80, 0x00, 0x01];
+ let mut buffer = [0u8; 5];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ // rec1value R ::= { mantissa -1, base 2, exponent 0 }
+ let val = -1.0;
+ let expected = &[0x09, 0x03, 0xc0, 0x00, 0x01];
+ let mut buffer = [0u8; 5];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ // rec1value R ::= { mantissa -1, base 2, exponent 1 }
+ let val = -1.0000000000000002;
+ let expected = &[0x09, 0x03, 0xc0, 0x00, 0x02];
+ let mut buffer = [0u8; 5];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ // rec1value R ::= { mantissa 1, base 2, exponent -1022 }
+ // NOTE: f64::MIN_EXP == -1021 so the exponent decoded by ASN.1 is what we expect
+ let val = f64::MIN_POSITIVE;
+ let expected = &[0x09, 0x04, 0x81, 0xfc, 0x02, 0x01];
+ let mut buffer = [0u8; 7];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ // rec4value R ::= { mantissa 1, base 2, exponent 3 }
+ let val = 1.0000000000000016;
+ let expected = &[0x09, 0x03, 0x80, 0x00, 0x08];
+ let mut buffer = [0u8; 5];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ // rec5value R ::= { mantissa 4222124650659841, base 2, exponent 4 }
+ let val = 31.0;
+ let expected = &[
+ 0x9, 0x9, 0x80, 0x04, 0x0f, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01,
+ ];
+ let mut buffer = [0u8; 11];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+ }
+
+ #[test]
+ fn encdec_irrationals() {
+ {
+ let val = core::f64::consts::PI;
+ let expected = &[
+ 0x09, 0x09, 0x80, 0x01, 0x09, 0x21, 0xfb, 0x54, 0x44, 0x2d, 0x19,
+ ];
+ let mut buffer = [0u8; 11];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = core::f64::consts::E;
+ let expected = &[
+ 0x09, 0x09, 0x80, 0x01, 0x05, 0xbf, 0x0a, 0x8b, 0x14, 0x57, 0x6a,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+ {
+ let val = core::f64::consts::LN_2;
+ let expected = &[
+ 0x09, 0x0a, 0x81, 0xff, 0xff, 0x6, 0x2e, 0x42, 0xfe, 0xfa, 0x39, 0xf0,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+ }
+
+ #[test]
+ fn encdec_reasonable_f64() {
+ // Tests the encoding and decoding of reals with some arbitrary numbers
+ {
+ // rec1value R ::= { mantissa 2414341043715239, base 2, exponent 21 }
+ let val = 3221417.1584163485;
+ let expected = &[
+ 0x9, 0x9, 0x80, 0x15, 0x8, 0x93, 0xd4, 0x94, 0x46, 0xfc, 0xa7,
+ ];
+ let mut buffer = [0u8; 11];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ // rec1value R ::= { mantissa 2671155248072715, base 2, exponent 23 }
+ let val = 13364022.365665454;
+ let expected = &[
+ 0x09, 0x09, 0x80, 0x17, 0x09, 0x7d, 0x66, 0xcb, 0xb3, 0x88, 0x0b,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ // rec1value R ::= { mantissa -4386812962460287, base 2, exponent 14 }
+ let val = -32343.132588105735;
+ let expected = &[
+ 0x09, 0x09, 0xc0, 0x0e, 0x0f, 0x95, 0xc8, 0x7c, 0x52, 0xd2, 0x7f,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = -27084.866751869475;
+ let expected = &[
+ 0x09, 0x09, 0xc0, 0x0e, 0x0a, 0x73, 0x37, 0x78, 0xdc, 0xd5, 0x4a,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ // rec1value R ::= { mantissa -4372913134428149, base 2, exponent 7 }
+ let val = -252.28566647111404;
+ let expected = &[
+ 0x09, 0x09, 0xc0, 0x07, 0x0f, 0x89, 0x24, 0x2e, 0x02, 0xdf, 0xf5,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = -14.399709612928548;
+ let expected = &[
+ 0x09, 0x09, 0xc0, 0x03, 0x0c, 0xcc, 0xa6, 0xbd, 0x06, 0xd9, 0x92,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = -0.08340570261832964;
+ let expected = &[
+ 0x09, 0x0a, 0xc1, 0xff, 0xfc, 0x05, 0x5a, 0x13, 0x7d, 0x0b, 0xae, 0x3d,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = 0.00536851453803701;
+ let expected = &[
+ 0x09, 0x0a, 0x81, 0xff, 0xf8, 0x05, 0xfd, 0x4b, 0xa5, 0xe7, 0x4c, 0x93,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = 0.00045183525648866433;
+ let expected = &[
+ 0x09, 0x0a, 0x81, 0xff, 0xf4, 0x0d, 0x9c, 0x89, 0xa6, 0x59, 0x33, 0x39,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = 0.000033869092002682955;
+ let expected = &[
+ 0x09, 0x0a, 0x81, 0xff, 0xf1, 0x01, 0xc1, 0xd5, 0x23, 0xd5, 0x54, 0x7c,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = 0.0000011770891033600088;
+ let expected = &[
+ 0x09, 0x0a, 0x81, 0xff, 0xec, 0x03, 0xbf, 0x8f, 0x27, 0xf4, 0x62, 0x56,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = 0.00000005549514041997082;
+ let expected = &[
+ 0x09, 0x0a, 0x81, 0xff, 0xe7, 0x0d, 0xcb, 0x31, 0xab, 0x6e, 0xb8, 0xd7,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = 0.0000000012707044685547803;
+ let expected = &[
+ 0x09, 0x0a, 0x81, 0xff, 0xe2, 0x05, 0xd4, 0x9e, 0x0a, 0xf2, 0xff, 0x1f,
+ ];
+ let mut buffer = [0u8; 12];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+
+ {
+ let val = 0.00000000002969611878378562;
+ let expected = &[
+ 0x09, 0x09, 0x81, 0xff, 0xdd, 0x53, 0x5b, 0x6f, 0x97, 0xee, 0xb6,
+ ];
+ let mut buffer = [0u8; 11];
+ let encoded = val.encode_to_slice(&mut buffer).unwrap();
+ assert_eq!(
+ expected, encoded,
+ "invalid encoding of {}:\ngot {:x?}\nwant: {:x?}",
+ val, encoded, expected
+ );
+ let decoded = f64::from_der(encoded).unwrap();
+ assert!(
+ (decoded - val).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ val,
+ decoded
+ );
+ }
+ }
+
+ #[test]
+ fn reject_non_canonical() {
+ assert!(f64::from_der(&[0x09, 0x81, 0x00]).is_err());
+ }
+
+ #[test]
+ fn encdec_f64() {
+ use super::{decode_f64, encode_f64};
+ // Test that the extraction and recreation works
+ for val in [
+ 1.0,
+ 0.1,
+ -0.1,
+ -1.0,
+ 0.0,
+ f64::MIN_POSITIVE,
+ f64::MAX,
+ f64::MIN,
+ 3.1415,
+ 951.2357864,
+ -3.1415,
+ -951.2357864,
+ ] {
+ let (s, e, m) = decode_f64(val);
+ let val2 = encode_f64(s, e, m);
+ assert!(
+ (val - val2).abs() < f64::EPSILON,
+ "fail - want {}\tgot {}",
+ val,
+ val2
+ );
+ }
+ }
+
+ #[test]
+ fn validation_cases() {
+ // Caveat: these test cases are validated on the ASN.1 playground: https://asn1.io/asn1playground/ .
+ // The test case consists in inputing the bytes in the "decode" field and checking that the decoded
+ // value corresponds to the one encoded here.
+ // This tool encodes _all_ values that are non-zero in the ISO 6093 NR3 representation.
+ // This does not seem to perfectly adhere to the ITU specifications, Special Cases section.
+ // The implementation of this crate correctly supports decoding such values. It will, however,
+ // systematically encode REALs in their base 2 form, with a scaling factor where needed to
+ // ensure that the mantissa is either odd or zero (as per section 11.3.1).
+
+ // Positive trivial numbers
+ {
+ let expect = 10.0;
+ let testcase = &[0x09, 0x05, 0x03, 0x31, 0x2E, 0x45, 0x31];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ {
+ let expect = 100.0;
+ let testcase = &[0x09, 0x05, 0x03, 0x31, 0x2E, 0x45, 0x32];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ {
+ let expect = 101.0;
+ let testcase = &[0x09, 0x08, 0x03, 0x31, 0x30, 0x31, 0x2E, 0x45, 0x2B, 0x30];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ {
+ let expect = 101.0;
+ let testcase = &[0x09, 0x08, 0x03, 0x31, 0x30, 0x31, 0x2E, 0x45, 0x2B, 0x30];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ {
+ let expect = 0.0;
+ let testcase = &[0x09, 0x00];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ {
+ let expect = 951.2357864;
+ let testcase = &[
+ 0x09, 0x0F, 0x03, 0x39, 0x35, 0x31, 0x32, 0x33, 0x35, 0x37, 0x38, 0x36, 0x34, 0x2E,
+ 0x45, 0x2D, 0x37,
+ ];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ // Negative trivial numbers
+ {
+ let expect = -10.0;
+ let testcase = &[0x09, 0x06, 0x03, 0x2D, 0x31, 0x2E, 0x45, 0x31];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ {
+ let expect = -100.0;
+ let testcase = &[0x09, 0x06, 0x03, 0x2D, 0x31, 0x2E, 0x45, 0x32];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ {
+ let expect = -101.0;
+ let testcase = &[
+ 0x09, 0x09, 0x03, 0x2D, 0x31, 0x30, 0x31, 0x2E, 0x45, 0x2B, 0x30,
+ ];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ {
+ let expect = -0.5;
+ let testcase = &[0x09, 0x07, 0x03, 0x2D, 0x35, 0x2E, 0x45, 0x2D, 0x31];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ {
+ let expect = -0.0;
+ let testcase = &[0x09, 0x03, 0x01, 0x2D, 0x30];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ {
+ // Test NR3 decoding
+ let expect = -951.2357864;
+ let testcase = &[
+ 0x09, 0x10, 0x03, 0x2D, 0x39, 0x35, 0x31, 0x32, 0x33, 0x35, 0x37, 0x38, 0x36, 0x34,
+ 0x2E, 0x45, 0x2D, 0x37,
+ ];
+ let decoded = f64::from_der(testcase).unwrap();
+ assert!(
+ (decoded - expect).abs() < f64::EPSILON,
+ "wanted: {}\tgot: {}",
+ expect,
+ decoded
+ );
+ }
+ }
+}