Skip to content

Improve floating-point fraction and negative constant reconstruction #4022

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
siegfriedpammer merged 4 commits into from
Aug 18, 2026
+186 −31
Merged

Improve floating-point fraction and negative constant reconstruction #4022

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
b303568
Recognize negated floating-point special constants
SychicBoy Aug 16, 2026
59361d9
Prefer common machine-scale floating-point fractions
SychicBoy Aug 16, 2026
b209b39
Drive negated special constants from the lookup table
siegfriedpammer Aug 17, 2026
2311b64
Cover the fractions that were not written as fractions
siegfriedpammer Aug 18, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
21 changes: 21 additions & 0 deletions ICSharpCode.Decompiler.Tests/TestCases/Pretty/WellKnownConstants.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,6 +61,7 @@ public class WellKnownConstants
public const float FloatMaxValue = float.MaxValue;
public const float FloatMinValue = float.MinValue;
public const float FloatEpsilon = float.Epsilon;
public const float FloatNegativeEpsilon = -float.Epsilon;

public const double DoubleZero = 0.0;
public const double DoubleMinusZero = -0.0;
Expand All @@ -70,6 +71,7 @@ public class WellKnownConstants
public const double DoubleMaxValue = double.MaxValue;
public const double DoubleMinValue = double.MinValue;
public const double DoubleEpsilon = double.Epsilon;
public const double DoubleNegativeEpsilon = -double.Epsilon;

public const decimal DecimalMaxValue = decimal.MaxValue;
public const decimal DecimalMinValue = decimal.MinValue;
Expand All @@ -82,6 +84,25 @@ public class WellKnownConstants
public const double Double_Sixth = 1.0 / 6.0;
public const float Float_Tenth = 0.1f;
public const double Double_Tenth = 0.1;
public const float Float_Third = 1f / 3f;
public const float Float_PowerOfTwoFraction = 21f / 32f;
public const float Float_SmoothFraction = 2f / 15f;
public const float Float_UnitFraction = 1f / 85f;
public const float Float_ByteScale_225 = 225f / 255f;
public const float Float_ByteScale_200 = 200f / 255f;
public const float Float_ByteScale_150 = 150f / 255f;
public const float Float_NegativeByteScale = -200f / 255f;
public const double Double_ByteScale = 200.0 / 255.0;
public const float Float_KScale = 123f / 1024f;
public const float Float_MScale = 123f / 1048576f;
public const float Float_UnitKScale = 1f / 1024f;
public const float Float_SmallKScale = 17f / 1024f;
public const float Float_UnreducedKScale = 126f / 1024f;
public const float Float_ShortBinaryFraction = 13f / 256f;
public const float Float_NearOnePowerOfTwo = 225f / 256f;
public const float Float_TinyPowerOfTwoScale = 1f / 32768f;
public const float Float_NotAPreferredScale = 0.123456f;
public const double Double_NotAPreferredScale = 0.123456789;

#if ROSLYN2 && !NET40
public const float Float_PI = MathF.PI;
Expand Down
196 changes: 165 additions & 31 deletions ICSharpCode.Decompiler/CSharp/Syntax/TypeSystemAstBuilder.cs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1245,47 +1245,60 @@ bool IsSpecialConstant(IType expectedType, object constant, [NotNullWhen(true)]
if (AddResolveResultAnnotations)
expression.AddAnnotation(new MemberResolveResult(new TypeResolveResult(constantType), field));

if (info.Negate)
{
expression = new UnaryOperatorExpression(UnaryOperatorType.Minus, expression);
if (AddResolveResultAnnotations)
expression.AddAnnotation(new ConstantResolveResult(constantType, constant));
}

return true;
}

static readonly Dictionary<object, (KnownTypeCode Type, string Member)> specialConstants = new Dictionary<object, (KnownTypeCode Type, string Member)>() {
// Negate wraps the member reference in a unary minus. The negative counterpart of most
// members is itself a member (-MaxValue is exactly MinValue, and both infinities have
// their own), so only Epsilon needs it. It cannot be derived from the sign of the
// constant: MinValue and NegativeInfinity are negative, but must not be negated.
static readonly Dictionary<object, (KnownTypeCode Type, string Member, bool Negate)> specialConstants = new Dictionary<object, (KnownTypeCode Type, string Member, bool Negate)>() {
// byte:
{ byte.MaxValue, (KnownTypeCode.Byte, "MaxValue") },
{ byte.MaxValue, (KnownTypeCode.Byte, "MaxValue", false) },
// sbyte:
{ sbyte.MinValue, (KnownTypeCode.SByte, "MinValue") },
{ sbyte.MaxValue, (KnownTypeCode.SByte, "MaxValue") },
{ sbyte.MinValue, (KnownTypeCode.SByte, "MinValue", false) },
{ sbyte.MaxValue, (KnownTypeCode.SByte, "MaxValue", false) },
// short:
{ short.MinValue, (KnownTypeCode.Int16, "MinValue") },
{ short.MaxValue, (KnownTypeCode.Int16, "MaxValue") },
{ short.MinValue, (KnownTypeCode.Int16, "MinValue", false) },
{ short.MaxValue, (KnownTypeCode.Int16, "MaxValue", false) },
// ushort:
{ ushort.MaxValue, (KnownTypeCode.UInt16, "MaxValue") },
{ ushort.MaxValue, (KnownTypeCode.UInt16, "MaxValue", false) },
// int:
{ int.MinValue, (KnownTypeCode.Int32, "MinValue") },
{ int.MaxValue, (KnownTypeCode.Int32, "MaxValue") },
{ int.MinValue, (KnownTypeCode.Int32, "MinValue", false) },
{ int.MaxValue, (KnownTypeCode.Int32, "MaxValue", false) },
// uint:
{ uint.MaxValue, (KnownTypeCode.UInt32, "MaxValue") },
{ uint.MaxValue, (KnownTypeCode.UInt32, "MaxValue", false) },
// long:
{ long.MinValue, (KnownTypeCode.Int64, "MinValue") },
{ long.MaxValue, (KnownTypeCode.Int64, "MaxValue") },
{ long.MinValue, (KnownTypeCode.Int64, "MinValue", false) },
{ long.MaxValue, (KnownTypeCode.Int64, "MaxValue", false) },
// ulong:
{ ulong.MaxValue, (KnownTypeCode.UInt64, "MaxValue") },
{ ulong.MaxValue, (KnownTypeCode.UInt64, "MaxValue", false) },
// float:
{ float.NaN, (KnownTypeCode.Single, "NaN") },
{ float.NegativeInfinity, (KnownTypeCode.Single, "NegativeInfinity") },
{ float.PositiveInfinity, (KnownTypeCode.Single, "PositiveInfinity") },
{ float.MinValue, (KnownTypeCode.Single, "MinValue") },
{ float.MaxValue, (KnownTypeCode.Single, "MaxValue") },
{ float.Epsilon, (KnownTypeCode.Single, "Epsilon") },
{ float.NaN, (KnownTypeCode.Single, "NaN", false) },
{ float.NegativeInfinity, (KnownTypeCode.Single, "NegativeInfinity", false) },
{ float.PositiveInfinity, (KnownTypeCode.Single, "PositiveInfinity", false) },
{ float.MinValue, (KnownTypeCode.Single, "MinValue", false) },
{ float.MaxValue, (KnownTypeCode.Single, "MaxValue", false) },
{ float.Epsilon, (KnownTypeCode.Single, "Epsilon", false) },
{ -float.Epsilon, (KnownTypeCode.Single, "Epsilon", true) },
// double:
{ double.NaN, (KnownTypeCode.Double, "NaN") },
{ double.NegativeInfinity, (KnownTypeCode.Double, "NegativeInfinity") },
{ double.PositiveInfinity, (KnownTypeCode.Double, "PositiveInfinity") },
{ double.MinValue, (KnownTypeCode.Double, "MinValue") },
{ double.MaxValue, (KnownTypeCode.Double, "MaxValue") },
{ double.Epsilon, (KnownTypeCode.Double, "Epsilon") },
{ double.NaN, (KnownTypeCode.Double, "NaN", false) },
{ double.NegativeInfinity, (KnownTypeCode.Double, "NegativeInfinity", false) },
{ double.PositiveInfinity, (KnownTypeCode.Double, "PositiveInfinity", false) },
{ double.MinValue, (KnownTypeCode.Double, "MinValue", false) },
{ double.MaxValue, (KnownTypeCode.Double, "MaxValue", false) },
{ double.Epsilon, (KnownTypeCode.Double, "Epsilon", false) },
{ -double.Epsilon, (KnownTypeCode.Double, "Epsilon", true) },
// decimal:
{ decimal.MinValue, (KnownTypeCode.Decimal, "MinValue") },
{ decimal.MaxValue, (KnownTypeCode.Decimal, "MaxValue") },
{ decimal.MinValue, (KnownTypeCode.Decimal, "MinValue", false) },
{ decimal.MaxValue, (KnownTypeCode.Decimal, "MaxValue", false) },
};

bool IsFlagsEnum(ITypeDefinition type)
Expand Down Expand Up @@ -1495,6 +1508,103 @@ static bool IsEqual(long num, long den, object constantValue, bool isDouble)
const int MAX_DENOMINATOR_DOUBLE = 1000;
const int MAX_DENOMINATOR_FLOAT = 360;

// Common machine-scale denominators: powers of two used for binary scaling,
// and 2^n-1 values used when normalizing integers (for example, byte colors / 255).
// Keep this as a targeted candidate set rather than increasing the generic denominator
// limit, which would reintroduce accidental fraction matches for ordinary floating-point values.
static readonly int[] preferredFractionDenominators = {
127, 128,
255, 256,
1023, 1024,
4095, 4096,
8192,
16384,
32767, 32768,
65535, 65536,
1048576
};

static int GetIntegerLiteralLength(long value)
{
int length = value < 0 ? 1 : 0;
do
{
length++;
value /= 10;
} while (value != 0);
return length;
}

static int GetFractionDisplayLength(long num, long den, bool isDouble)
{
// Float integer constants use the `f` suffix; double constants need `.0`.
int numericSuffixLength = isDouble ? 2 : 1;
return GetIntegerLiteralLength(num) + GetIntegerLiteralLength(den)
+ 3 + 2 * numericSuffixLength; // `num / den`
}

// Unit fractions and denominators composed only of 2, 3 and 5 are already
// conventional forms; do not expand them just to reach a preferred scale.
static bool IsSimpleFraction(long num, long den)
{
Debug.Assert(den > 0);
if (num == 1 || num == -1)
return true;
while (den % 2 == 0)
den /= 2;
while (den % 3 == 0)
den /= 3;
while (den % 5 == 0)
den /= 5;
return den == 1;
}

static int GetPreferredFractionScore(long num, int den, bool isDouble)
{
// Powers of two are native to binary floating point and therefore more likely to
// match by coincidence. Values of the form 2^n-1 are a stronger normalization
// signal, so allow them a slightly larger readability bonus.
bool isPowerOfTwo = (den & (den - 1)) == 0;
Debug.Assert(isPowerOfTwo || ((den + 1) & den) == 0);
int readabilityBonus = isPowerOfTwo ? 1 : 2;
return GetFractionDisplayLength(num, den, isDouble) - readabilityBonus;
}

static bool TryGetPreferredFraction(object constantValue, bool isDouble, out long num, out long den, out int score)
{
num = 0;
den = 0;
score = int.MaxValue;

double value = isDouble ? (double)constantValue : (float)constantValue;
if (!(Math.Abs(value) < 1.0))
return false;

foreach (int candidateDen in preferredFractionDenominators)
{
long candidateNum = (long)Math.Round(value * candidateDen);
if (candidateNum == 0 || candidateNum <= -candidateDen || candidateNum >= candidateDen)
continue;
if (!IsEqual(candidateNum, candidateDen, constantValue, isDouble))
continue;

int candidateScore = GetPreferredFractionScore(candidateNum, candidateDen, isDouble);
if (candidateScore < score || (candidateScore == score && candidateDen < den))
{
num = candidateNum;
den = candidateDen;
score = candidateScore;
}
}

return den != 0;
}

Expression MakeFraction(IType type, long num, long den)
{
return new BinaryOperatorExpression(MakeConstant(type, num), BinaryOperatorType.Divide, MakeConstant(type, den));
}

Expression ConvertFloatingPointLiteral(IType type, object constantValue)
{
// Coerce constantValue to either float or double:
Expand Down Expand Up @@ -1536,11 +1646,35 @@ Expression ConvertFloatingPointLiteral(IType type, object constantValue)
? FractionApprox((double)constantValue, MAX_DENOMINATOR_DOUBLE)
: FractionApprox((float)constantValue, 200);

if (IsValidFraction(num, den) && IsEqual(num, den, constantValue, isDouble) && Math.Abs(den) != 1)
bool hasRegularFraction = IsValidFraction(num, den)
&& IsEqual(num, den, constantValue, isDouble)
&& Math.Abs(den) != 1;

if (TryGetPreferredFraction(constantValue, isDouble, out long preferredNum, out long preferredDen, out int preferredScore))
{
int baselineLength = hasRegularFraction
? GetFractionDisplayLength(num, den, isDouble)
: str.Length + (isDouble ? 0 : 1);
// Do not replace an already-simple fraction (for example 21 / 32 or 2 / 15)
// just to reach one of the larger preferred scales.
bool regularFractionIsSimple = hasRegularFraction && IsSimpleFraction(num, den);
if (!regularFractionIsSimple && preferredScore <= baselineLength)
{
if (hasRegularFraction)
{
num = preferredNum;
den = preferredDen;
}
else
{
expr = MakeFraction(type, preferredNum, preferredDen);
}
}
}

if (hasRegularFraction)
{
var left = MakeConstant(type, num);
var right = MakeConstant(type, den);
expr = new BinaryOperatorExpression(left, BinaryOperatorType.Divide, right);
expr = MakeFraction(type, num, den);
}
}

Expand Down
Loading