diff --git a/src/FirebirdSql.Data.FirebirdClient.Tests/DecimalCodecTests.cs b/src/FirebirdSql.Data.FirebirdClient.Tests/DecimalCodecTests.cs
new file mode 100644
index 00000000..3c222bd0
--- /dev/null
+++ b/src/FirebirdSql.Data.FirebirdClient.Tests/DecimalCodecTests.cs
@@ -0,0 +1,106 @@
+/*
+ *    The contents of this file are subject to the Initial
+ *    Developer's Public License Version 1.0 (the "License");
+ *    you may not use this file except in compliance with the
+ *    License. You may obtain a copy of the License at
+ *    https://github.com/FirebirdSQL/NETProvider/raw/master/license.txt.
+ *
+ *    Software distributed under the License is distributed on
+ *    an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, either
+ *    express or implied. See the License for the specific
+ *    language governing rights and limitations under the License.
+ *
+ *    All Rights Reserved.
+ */
+
+//$Authors = F.D.Castel
+
+using System;
+using System.Numerics;
+using FirebirdSql.Data.Common;
+using FirebirdSql.Data.TestsBase;
+using FirebirdSql.Data.Types;
+using NUnit.Framework;
+
+namespace FirebirdSql.Data.FirebirdClient.Tests;
+
+[NoServerCategory]
+public class DecimalCodecTests
+{
+	static readonly object[] RepresentableDecFloat16Values = new object[]
+	{
+		new FbDecFloat(BigInteger.Parse("9999999999999999"), 369),
+		new FbDecFloat(BigInteger.Parse("-9999999999999999"), 369),
+		new FbDecFloat(1, -398),
+		new FbDecFloat(-1, -398),
+		// representable only after normalization
+		new FbDecFloat(1, 384),
+		new FbDecFloat(-1, 384),
+		new FbDecFloat(BigInteger.Parse("100000000000000000000"), -20),
+		new FbDecFloat(BigInteger.Parse("12345678901234560"), 0),
+		new FbDecFloat(100, -400),
+		new FbDecFloat(0, 1000),
+		new FbDecFloat(0, -1000),
+	};
+	[TestCaseSource(nameof(RepresentableDecFloat16Values))]
+	public void DecFloat16EncodesRepresentableValueExactly(FbDecFloat value)
+	{
+		var result = DecimalCodec.DecFloat16.ParseBytes(DecimalCodec.DecFloat16.EncodeDecimal(value));
+		Assert.AreEqual(value, result);
+	}
+
+	static readonly object[] RepresentableDecFloat34Values = new object[]
+	{
+		new FbDecFloat(BigInteger.Parse("9999999999999999999999999999999999"), 6111),
+		new FbDecFloat(1, -6176),
+		// representable only after normalization
+		new FbDecFloat(1, 6144),
+		new FbDecFloat(BigInteger.Parse("12345678901234567890123456789012340"), 0),
+		new FbDecFloat(BigInteger.Parse("-12345678901234567890123456789012340000"), -40),
+		new FbDecFloat(1000, -6179),
+		new FbDecFloat(0, int.MaxValue),
+		new FbDecFloat(0, int.MinValue),
+	};
+	[TestCaseSource(nameof(RepresentableDecFloat34Values))]
+	public void DecFloat34EncodesRepresentableValueExactly(FbDecFloat value)
+	{
+		var result = DecimalCodec.DecFloat34.ParseBytes(DecimalCodec.DecFloat34.EncodeDecimal(value));
+		Assert.AreEqual(value, result);
+	}
+
+	[Test]
+	public void ZeroExponentIsClamped()
+	{
+		var result = DecimalCodec.DecFloat16.ParseBytes(DecimalCodec.DecFloat16.EncodeDecimal(new FbDecFloat(0, 1000)));
+		Assert.AreEqual(369, result.Exponent);
+	}
+
+	static readonly object[] UnrepresentableDecFloat16Values = new object[]
+	{
+		new FbDecFloat(BigInteger.Parse("30000000000000004"), -17),
+		new FbDecFloat(BigInteger.Parse("-12345678901234567"), 0),
+		new FbDecFloat(1, 385),
+		new FbDecFloat(BigInteger.Parse("9999999999999999"), 370),
+		new FbDecFloat(1, -399),
+		new FbDecFloat(15, -400),
+		new FbDecFloat(1, int.MaxValue),
+		new FbDecFloat(1, int.MinValue),
+	};
+	[TestCaseSource(nameof(UnrepresentableDecFloat16Values))]
+	public void DecFloat16ThrowsInsteadOfRounding(FbDecFloat value)
+	{
+		Assert.Throws<OverflowException>(() => DecimalCodec.DecFloat16.EncodeDecimal(value));
+	}
+
+	static readonly object[] UnrepresentableDecFloat34Values = new object[]
+	{
+		new FbDecFloat(BigInteger.Parse("12345678901234567890123456789012345"), 0),
+		new FbDecFloat(1, 6145),
+		new FbDecFloat(1, -6177),
+	};
+	[TestCaseSource(nameof(UnrepresentableDecFloat34Values))]
+	public void DecFloat34ThrowsInsteadOfRounding(FbDecFloat value)
+	{
+		Assert.Throws<OverflowException>(() => DecimalCodec.DecFloat34.EncodeDecimal(value));
+	}
+}
diff --git a/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloat16SupportTests.cs b/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloat16SupportTests.cs
index 1fba5362..c0570c6e 100644
--- a/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloat16SupportTests.cs
+++ b/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloat16SupportTests.cs
@@ -63,8 +63,25 @@ public override async Task SetUp()
 			new object[] { new FbDecFloat(-10, 34), "-100000000000000000000000000000000000" },
 			new object[] { new FbDecFloat(BigInteger.Parse("123000000001"), -10), "123.000000001E-1" },
 			new object[] { new FbDecFloat(BigInteger.Parse("-123000000001"), -10), "-123.000000001E-1" },
+			new object[] { new FbDecFloat(1, 384), "1E384" },
+			new object[] { new FbDecFloat(BigInteger.Parse("100000000000000000000"), -20), "1" },
+			new object[] { new FbDecFloat(100, -400), "1E-398" },
+			new object[] { (FbDecFloat)2.0, "2" },
+			new object[] { (FbDecFloat)1e-20, "1E-20" },
+			new object[] { (FbDecFloat)2.5f, "2.5" },
+			new object[] { (FbDecFloat)100m, "100" },
+			new object[] { (FbDecFloat)1.10m, "1.10" },
 		};
 
+	static readonly object[] UnrepresentableValues = new object[]
+	{
+		new FbDecFloat(BigInteger.Parse("30000000000000004"), -17),
+		new FbDecFloat(1, 385),
+		new FbDecFloat(1, -399),
+		(FbDecFloat)(0.1 + 0.2),
+		(FbDecFloat)decimal.MaxValue,
+	};
+
 	[TestCaseSource(nameof(TestValues))]
 	public async Task ReadsValueCorrectly(FbDecFloat value, string castValue)
 	{
@@ -88,6 +105,22 @@ public async Task PassesValueCorrectly(FbDecFloat value, string dummy)
 		}
 	}
 
+	[TestCaseSource(nameof(UnrepresentableValues))]
+	public async Task PassesUnrepresentableValueThrows(FbDecFloat value)
+	{
+		await using (var cmd = Connection.CreateCommand())
+		{
+			cmd.CommandText = "select cast(@value as decfloat(16)) from rdb$database";
+			cmd.Parameters.AddWithValue("value", value);
+			Assert.ThrowsAsync<OverflowException>(() => cmd.ExecuteScalarAsync());
+		}
+		await using (var cmd = Connection.CreateCommand())
+		{
+			cmd.CommandText = "select cast(1 as decfloat(16)) from rdb$database";
+			Assert.AreEqual((FbDecFloat)1, (FbDecFloat)await cmd.ExecuteScalarAsync(), "connection must remain usable");
+		}
+	}
+
 	[Test]
 	public async Task ReadsValueNullCorrectly()
 	{
diff --git a/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloat34SupportTests.cs b/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloat34SupportTests.cs
index b1c2bf39..32d0a26e 100644
--- a/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloat34SupportTests.cs
+++ b/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloat34SupportTests.cs
@@ -63,8 +63,25 @@ public override async Task SetUp()
 			new object[] { new FbDecFloat(-10, 34), "-100000000000000000000000000000000000" },
 			new object[] { new FbDecFloat(BigInteger.Parse("123000000001"), -10), "123.000000001E-1" },
 			new object[] { new FbDecFloat(BigInteger.Parse("-123000000001"), -10), "-123.000000001E-1" },
+			new object[] { new FbDecFloat(1, 6144), "1E6144" },
+			new object[] { new FbDecFloat(BigInteger.Parse("12345678901234567890123456789012340"), 0), "12345678901234567890123456789012340" },
+			new object[] { new FbDecFloat(1000, -6179), "1E-6176" },
+			new object[] { (FbDecFloat)2.0, "2" },
+			new object[] { (FbDecFloat)1e-20, "1E-20" },
+			new object[] { (FbDecFloat)2.5f, "2.5" },
+			new object[] { (FbDecFloat)(0.1 + 0.2), "0.30000000000000004" },
+			new object[] { (FbDecFloat)100m, "100" },
+			new object[] { (FbDecFloat)1.10m, "1.10" },
+			new object[] { (FbDecFloat)decimal.MaxValue, "79228162514264337593543950335" },
 		};
 
+	static readonly object[] UnrepresentableValues = new object[]
+	{
+		new FbDecFloat(BigInteger.Parse("12345678901234567890123456789012345"), 0),
+		new FbDecFloat(1, 6145),
+		new FbDecFloat(1, -6177),
+	};
+
 	[TestCaseSource(nameof(TestValues))]
 	public async Task ReadsValueCorrectly(FbDecFloat value, string castValue)
 	{
@@ -88,6 +105,22 @@ public async Task PassesValueCorrectly(FbDecFloat value, string dummy)
 		}
 	}
 
+	[TestCaseSource(nameof(UnrepresentableValues))]
+	public async Task PassesUnrepresentableValueThrows(FbDecFloat value)
+	{
+		await using (var cmd = Connection.CreateCommand())
+		{
+			cmd.CommandText = "select cast(@value as decfloat(34)) from rdb$database";
+			cmd.Parameters.AddWithValue("value", value);
+			Assert.ThrowsAsync<OverflowException>(() => cmd.ExecuteScalarAsync());
+		}
+		await using (var cmd = Connection.CreateCommand())
+		{
+			cmd.CommandText = "select cast(1 as decfloat(34)) from rdb$database";
+			Assert.AreEqual((FbDecFloat)1, (FbDecFloat)await cmd.ExecuteScalarAsync(), "connection must remain usable");
+		}
+	}
+
 	[Test]
 	public async Task ReadsValueNullCorrectly()
 	{
diff --git a/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloatTypeTests.cs b/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloatTypeTests.cs
index 4df15b2d..ad41cca5 100644
--- a/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloatTypeTests.cs
+++ b/src/FirebirdSql.Data.FirebirdClient.Tests/FbDecFloatTypeTests.cs
@@ -15,6 +15,8 @@
 
 //$Authors = Jiri Cincura (jiri@cincura.net)
 
+using System;
+using System.Globalization;
 using System.Numerics;
 using FirebirdSql.Data.TestsBase;
 using FirebirdSql.Data.Types;
@@ -72,4 +74,115 @@ public void EqualityFalse(FbDecFloat expected, FbDecFloat actual)
 	{
 		Assert.AreNotEqual(expected, actual);
 	}
+
+	static readonly object[] DecimalConversionSource = new object[]
+	{
+			new object[] { 0m, "0", 0 },
+			new object[] { 1m, "1", 0 },
+			new object[] { 100m, "100", 0 },
+			new object[] { -5m, "-5", 0 },
+			new object[] { 1.5m, "15", -1 },
+			new object[] { 1.10m, "110", -2 },
+			new object[] { -0.001m, "-1", -3 },
+			new object[] { decimal.MaxValue, "79228162514264337593543950335", 0 },
+			new object[] { decimal.MinValue, "-79228162514264337593543950335", 0 },
+			new object[] { 0.0000000000000000000000000001m, "1", -28 },
+			new object[] { -7.9228162514264337593543950335m, "-79228162514264337593543950335", -28 },
+	};
+	[TestCaseSource(nameof(DecimalConversionSource))]
+	public void ConvertsDecimalExactly(decimal value, string coefficient, int exponent)
+	{
+		var result = (FbDecFloat)value;
+		Assert.AreEqual(BigInteger.Parse(coefficient), result.Coefficient);
+		Assert.AreEqual(exponent, result.Exponent);
+	}
+
+	static readonly object[] DoubleConversionSource = new object[]
+	{
+			new object[] { 0d, "0", 0 },
+			new object[] { 2d, "2", 0 },
+			new object[] { 100d, "100", 0 },
+			new object[] { 0.1, "1", -1 },
+			new object[] { -6.6, "-66", -1 },
+			new object[] { 1e-20, "1", -20 },
+			new object[] { 1e300, "1", 300 },
+			new object[] { 0.1 + 0.2, "30000000000000004", -17 },
+			new object[] { 123456789.123456789, "12345678912345679", -8 },
+			new object[] { double.MaxValue, "17976931348623157", 292 },
+			new object[] { double.MinValue, "-17976931348623157", 292 },
+			new object[] { double.Epsilon, "5", -324 },
+	};
+	[TestCaseSource(nameof(DoubleConversionSource))]
+	public void ConvertsDoubleToShortestRoundTripValue(double value, string coefficient, int exponent)
+	{
+		var result = (FbDecFloat)value;
+		Assert.AreEqual(BigInteger.Parse(coefficient), result.Coefficient);
+		Assert.AreEqual(exponent, result.Exponent);
+	}
+
+	static readonly object[] SingleConversionSource = new object[]
+	{
+			new object[] { 2f, "2", 0 },
+			new object[] { 0.1f, "1", -1 },
+			new object[] { -6.6f, "-66", -1 },
+			new object[] { 1e-20f, "1", -20 },
+			new object[] { float.MaxValue, "34028235", 31 },
+			new object[] { float.Epsilon, "1", -45 },
+	};
+	[TestCaseSource(nameof(SingleConversionSource))]
+	public void ConvertsSingleToShortestRoundTripValue(float value, string coefficient, int exponent)
+	{
+		var result = (FbDecFloat)value;
+		Assert.AreEqual(BigInteger.Parse(coefficient), result.Coefficient);
+		Assert.AreEqual(exponent, result.Exponent);
+	}
+
+	[Test]
+	public void ConvertsNegativeZero()
+	{
+		Assert.AreEqual(FbDecFloat.NegativeZero, (FbDecFloat)(-0d));
+		Assert.AreEqual(FbDecFloat.NegativeZero, (FbDecFloat)(-0f));
+		Assert.AreEqual(FbDecFloat.NegativeZero, (FbDecFloat)decimal.Negate(0m));
+		Assert.AreNotEqual(FbDecFloat.NegativeZero, (FbDecFloat)0d);
+		Assert.AreNotEqual(FbDecFloat.NegativeZero, (FbDecFloat)0m);
+	}
+
+	[Test]
+	public void ConvertsSpecialValues()
+	{
+		Assert.AreEqual(FbDecFloat.PositiveNaN, (FbDecFloat)double.NaN);
+		Assert.AreEqual(FbDecFloat.PositiveInfinity, (FbDecFloat)double.PositiveInfinity);
+		Assert.AreEqual(FbDecFloat.NegativeInfinity, (FbDecFloat)double.NegativeInfinity);
+		Assert.AreEqual(FbDecFloat.PositiveNaN, (FbDecFloat)float.NaN);
+		Assert.AreEqual(FbDecFloat.PositiveInfinity, (FbDecFloat)float.PositiveInfinity);
+		Assert.AreEqual(FbDecFloat.NegativeInfinity, (FbDecFloat)float.NegativeInfinity);
+	}
+
+	[Test]
+	public void DoubleAndSingleConversionsRoundTrip()
+	{
+		var random = new Random(20261107);
+		var negativeDoubles = 0;
+		var negativeSingles = 0;
+		for (var i = 0; i < 10_000; i++)
+		{
+			// random bit patterns, so all exponents and both signs are covered
+			var d = BitConverter.Int64BitsToDouble(random.NextInt64(long.MinValue, long.MaxValue));
+			if (double.IsFinite(d))
+			{
+				var result = (FbDecFloat)d;
+				Assert.AreEqual(d, double.Parse(result.ToString(), CultureInfo.InvariantCulture), $"{d:R} -> {result}");
+				negativeDoubles += d < 0 ? 1 : 0;
+			}
+			var f = BitConverter.Int32BitsToSingle(random.Next(int.MinValue, int.MaxValue));
+			if (float.IsFinite(f))
+			{
+				var result = (FbDecFloat)f;
+				Assert.AreEqual(f, float.Parse(result.ToString(), CultureInfo.InvariantCulture), $"{f:R} -> {result}");
+				negativeSingles += f < 0 ? 1 : 0;
+			}
+		}
+		Assert.Greater(negativeDoubles, 1000);
+		Assert.Greater(negativeSingles, 1000);
+	}
 }
diff --git a/src/FirebirdSql.Data.FirebirdClient/Common/DecimalCodec.cs b/src/FirebirdSql.Data.FirebirdClient/Common/DecimalCodec.cs
index 5ff63dde..8a7f8f9c 100644
--- a/src/FirebirdSql.Data.FirebirdClient/Common/DecimalCodec.cs
+++ b/src/FirebirdSql.Data.FirebirdClient/Common/DecimalCodec.cs
@@ -17,6 +17,7 @@
 
 using System;
 using System.Diagnostics;
+using System.Numerics;
 using FirebirdSql.Data.Types;
 
 namespace FirebirdSql.Data.Common;
@@ -41,6 +42,9 @@ public DecimalFormat(int formatBitLength, int coefficientDigits)
 			ELimit = 3 * (1 << ExponentContinuationBits) - 1;
 			EMin = -ELimit / 2;
 			ExponentBias = -EMin + CoefficientDigits - 1;
+			MaxCoefficient = BigInteger.Pow(10, CoefficientDigits) - 1;
+			MinExponent = UnbiasedExponent(0);
+			MaxExponent = UnbiasedExponent(ELimit);
 		}
 
 		public int FormatBitLength { get; }
@@ -51,6 +55,10 @@ public DecimalFormat(int formatBitLength, int coefficientDigits)
 		public int ELimit { get; }
 		public int EMin { get; }
 		public int ExponentBias { get; }
+		public BigInteger MaxCoefficient { get; }
+		// Range of the exponent applied to the integral coefficient (value = coefficient * 10^exponent).
+		public int MinExponent { get; }
+		public int MaxExponent { get; }
 
 		public void ValidateByteLength(byte[] decBytes)
 		{
@@ -171,8 +179,8 @@ public byte[] EncodeDecimal(FbDecFloat @decimal)
 
 	void EncodeFinite(FbDecFloat @decimal, byte[] decBytes)
 	{
-		var biasedExponent = _decimalFormat.BiasedExponent(@decimal.Exponent);
-		var coefficient = @decimal.Coefficient;
+		var (coefficient, exponent) = FitToFormat(@decimal);
+		var biasedExponent = _decimalFormat.BiasedExponent(exponent);
 		var mostSignificantDigit = _coefficientCoder.EncodeValue(coefficient, decBytes);
 		var expMSB = biasedExponent >>> _decimalFormat.ExponentContinuationBits;
 		var expTwoBitCont = (biasedExponent >>> _decimalFormat.ExponentContinuationBits - 2) & 0b011;
@@ -192,6 +200,52 @@ void EncodeFinite(FbDecFloat @decimal, byte[] decBytes)
 		EncodeExponentContinuation(decBytes, biasedExponent, _decimalFormat.ExponentContinuationBits - 2);
 	}
 
+	// Brings the coefficient and exponent into the ranges of the format without changing the numeric value
+	// (removing trailing zeros of the coefficient, or moving the exponent into the coefficient).
+	// Throws when that is impossible, i.e. when the value would have to be rounded, or overflows or underflows the format.
+	(BigInteger coefficient, int exponent) FitToFormat(FbDecFloat @decimal)
+	{
+		var coefficient = BigInteger.Abs(@decimal.Coefficient);
+		long exponent = @decimal.Exponent;
+		if (coefficient.IsZero)
+		{
+			return (coefficient, (int)Math.Clamp(exponent, _decimalFormat.MinExponent, _decimalFormat.MaxExponent));
+		}
+		while (coefficient > _decimalFormat.MaxCoefficient)
+		{
+			coefficient = BigInteger.DivRem(coefficient, 10, out var remainder);
+			if (!remainder.IsZero)
+			{
+				throw CannotRepresent(@decimal, $"more than {_decimalFormat.CoefficientDigits} significant digits");
+			}
+			exponent++;
+		}
+		while (exponent > _decimalFormat.MaxExponent)
+		{
+			coefficient *= 10;
+			if (coefficient > _decimalFormat.MaxCoefficient)
+			{
+				throw CannotRepresent(@decimal, "exponent overflow");
+			}
+			exponent--;
+		}
+		while (exponent < _decimalFormat.MinExponent)
+		{
+			coefficient = BigInteger.DivRem(coefficient, 10, out var remainder);
+			if (!remainder.IsZero)
+			{
+				throw CannotRepresent(@decimal, "exponent underflow");
+			}
+			exponent++;
+		}
+		return (coefficient, (int)exponent);
+	}
+
+	OverflowException CannotRepresent(FbDecFloat @decimal, string reason)
+	{
+		return new OverflowException($"Value {@decimal} cannot be represented exactly as DECFLOAT({_decimalFormat.CoefficientDigits}): {reason}.");
+	}
+
 	static void EncodeExponentContinuation(byte[] decBytes, int expAndBias, int expBitsRemaining)
 	{
 		var expByteIndex = 1;
diff --git a/src/FirebirdSql.Data.FirebirdClient/Types/FbDecFloat.cs b/src/FirebirdSql.Data.FirebirdClient/Types/FbDecFloat.cs
index 29c37c4e..7c2c22c5 100644
--- a/src/FirebirdSql.Data.FirebirdClient/Types/FbDecFloat.cs
+++ b/src/FirebirdSql.Data.FirebirdClient/Types/FbDecFloat.cs
@@ -64,16 +64,27 @@ public FbDecFloat(BigInteger coefficient, int exponent = 0)
 		float.NaN => PositiveNaN,
 		float.PositiveInfinity => PositiveInfinity,
 		float.NegativeInfinity => NegativeInfinity,
-		_ => ParseNumber(value, "0.#######"),
+		// shortest representation that round-trips to the same float
+		_ => ParseRoundTrip(value.ToString("R", CultureInfo.InvariantCulture)),
 	};
 	public static explicit operator FbDecFloat(double value) => value switch
 	{
 		double.NaN => PositiveNaN,
 		double.PositiveInfinity => PositiveInfinity,
 		double.NegativeInfinity => NegativeInfinity,
-		_ => ParseNumber(value, "0.###############"),
+		// shortest representation that round-trips to the same double
+		_ => ParseRoundTrip(value.ToString("R", CultureInfo.InvariantCulture)),
 	};
-	public static explicit operator FbDecFloat(decimal value) => ParseNumber(value, "0.############################");
+	public static explicit operator FbDecFloat(decimal value)
+	{
+		// exact: the decimal's 96-bit integer and scale are the coefficient and (negated) exponent
+		Span<int> bits = stackalloc int[4];
+		decimal.GetBits(value, bits);
+		BigInteger coefficient = ((UInt128)(uint)bits[2] << 64) | ((UInt128)(uint)bits[1] << 32) | (uint)bits[0];
+		var negative = bits[3] < 0;
+		var scale = (bits[3] >> 16) & 0xFF;
+		return new FbDecFloat(DecimalType.Finite, negative, negative ? -coefficient : coefficient, -scale);
+	}
 
 	public override string ToString()
 	{
@@ -155,10 +166,23 @@ public bool Equals(FbDecFloat other)
 		return !lhs.Equals(rhs);
 	}
 
-	static FbDecFloat ParseNumber(IFormattable formattable, string format)
+	// Parses the invariant "R" format of float/double, e.g. "2", "-0", "6.6", "1E-20", "1.7976931348623157E+308".
+	static FbDecFloat ParseRoundTrip(string s)
 	{
-		var s = formattable.ToString(format, CultureInfo.InvariantCulture);
-		var pos = s.IndexOf('.');
-		return new FbDecFloat(BigInteger.Parse(s.Remove(pos, 1), CultureInfo.InvariantCulture), -(s.Length - pos - 1));
+		var exponent = 0;
+		var exponentPos = s.IndexOf('E');
+		if (exponentPos >= 0)
+		{
+			exponent = int.Parse(s.AsSpan(exponentPos + 1), NumberStyles.AllowLeadingSign, CultureInfo.InvariantCulture);
+			s = s.Substring(0, exponentPos);
+		}
+		var pointPos = s.IndexOf('.');
+		if (pointPos >= 0)
+		{
+			exponent -= s.Length - pointPos - 1;
+			s = s.Remove(pointPos, 1);
+		}
+		var coefficient = BigInteger.Parse(s, NumberStyles.AllowLeadingSign, CultureInfo.InvariantCulture);
+		return new FbDecFloat(DecimalType.Finite, s[0] == '-', coefficient, exponent);
 	}
 }