[Bug] Simple division Expression returns invalid value

[darxis]

MathParser.org-mXparser verssion: v.5.2.1 Framework: .NET Framework 4.7.2

I came into a bug today, please see the following test case that I created to experience the bug:

[TestMethod]
public void testExpr1403(){
    TestCommonTools.testExprSettingsInit();
    bool testResult = false;
    String expStr = "0.0000004566969933 / 36370.51937825058";
    TestCommonTools.consolePrintTestExprStart(1, expStr);
    Expression testExp = new Expression(expStr);
    double value = testExp.calculate(); // it returns 1E-11 instead of 1.2556790530000044E-11
    double reg = 0.0000004566969933 / 36370.51937825058;
    if (value == reg)
        testResult = true;
    TestCommonTools.consolePrintTestExprEnd(value, reg, testResult, testExp);
    Assert.IsTrue(testResult);
}

The bug is that the division returns 1E-11 instead of 1.2556790530000044E-11

Please fix this, I will add a PR for the test case

[mariuszgromada]

Thanks.

Have you been trying switching off smart rounding settings? Below the link tot he section in the tutorial:

https://mathparser.org/mxparser-tutorial/smart-rounding-settings/

Best regards

[darxis]

@mariuszgromada I didn't change any default settings

mXparser.checkIfAlmostIntRounding(); // true
mXparser.checkIfUlpRounding(); // false
mXparser.checkIfCanonicalRounding(); // true

I encountered this bug when I upgraded from 4.4.2 to 5.2.1. On version 4.4.2 this bug doesn't occur but when I upgraded to 5.2.1 this bug is present. I didn't found any breaking changes in the change log, on the nuget package additional details it says: API of versions 5.x are fully backward compatible with all 4.x. Technical migrating to 5.x should be hassle free.. But it isn't as I can see :/

[mariuszgromada]

I belive there was a bug in 4.4 with default settings. Can you switch off all the roundings?

[darxis]

@mariuszgromada On both 4.4.2 and 5.2.1 only AlmostIntRounding and CanonicalRounding are enabled by default.

Version 4.4.2
mXparser.checkIfAlmostIntRounding(); // true
mXparser.checkIfUlpRounding(); // false
mXparser.checkIfCanonicalRounding(); // true

Version 5.2.1
mXparser.checkIfAlmostIntRounding(); // true
mXparser.checkIfUlpRounding(); // false
mXparser.checkIfCanonicalRounding(); // true

Also I can see that this bug is present since version v5.0.7, on v5.0.6 the bug is not present.

When I call mXparser.disableCanonicalRounding() the bug disappears. Why with CanonicalRounding enabled it gives 0.0000004566969933 / 36370.51937825058 = 0.00000000001?

[mariuszgromada]

Can you run this code for me?

mXparser.disableAlmostIntRounding();
mXparser.disableCanonicalRounding();
mXparser.disableUlpRounding();
String expStr = "0.0000004566969933 / 36370.51937825058";
Expression testExp = new Expression(expStr);
double value = testExp.calculate();
double reg = 0.0000004566969933 / 36370.51937825058;
System.Console.WriteLine("value = " + value);
System.Console.WriteLine("reg = " + reg);

[mariuszgromada]

Can you also make a test using more fresh .net framework?

[darxis]

Can you run this code for me?

mXparser.disableAlmostIntRounding();
mXparser.disableCanonicalRounding();
mXparser.disableUlpRounding();
String expStr = "0.0000004566969933 / 36370.51937825058";
Expression testExp = new Expression(expStr);
double value = testExp.calculate();
double reg = 0.0000004566969933 / 36370.51937825058;
System.Console.WriteLine("value = " + value);
System.Console.WriteLine("reg = " + reg);

It returns the following output, clearly disabling CanonicalRounding fixes this issue.

value = 1,2556790530000044E-11
reg = 1,2556790530000044E-11

[mariuszgromada]

Ok, now pls see my result when canonical rounding is enabled

mXparser.disableAlmostIntRounding();
mXparser.enableCanonicalRounding();
mXparser.disableUlpRounding();
String expStr = "0.0000004566969933 / 36370.51937825058";
Expression testExp = new Expression(expStr);
double value = testExp.calculate();
double reg = 0.0000004566969933 / 36370.51937825058;
System.Console.WriteLine("value = " + value);
System.Console.WriteLine("reg = " + reg);

Result

value = 1.2556791E-11
reg = 1.2556790530000044E-11

I am using a newer version of .net, I suspect the implementation difference in .net between 4.7.2 (your older version) and others. Can you test this on e.g. .net6 or some other version?

[darxis]

I am executing the code you sent me in the mXparser source UnitTests projects.

I can see that it fails only for .NET Framework and .NET Standard libraries, for .NET Core it works properly.

I created another test case for the code you just sent to execute it in the mXparser source UnitTests projects (4-Unit-Tests-Core, 4-Unit-Tests-Net, 4-Unit-Tests-Stand)

[TestMethod]
public void testExpr1405()
{
    mXparser.disableAlmostIntRounding();
    mXparser.enableCanonicalRounding();
    mXparser.disableUlpRounding();
    String expStr = "0.0000004566969933 / 36370.51937825058";
    Expression testExp = new Expression(expStr);
    double value = testExp.calculate();
    double reg = 0.0000004566969933 / 36370.51937825058;
    System.Console.WriteLine("value = " + value);
    System.Console.WriteLine("reg = " + reg);
}

The code you sent gives the different outputs for .NET Core and .NET Framework/Standard.

4-Unit-Tests-Core
value = 1,2556791E-11
reg = 1,2556790530000044E-11

4-Unit-Tests-Net
value = 1E-11
reg = 1,255679053E-11

4-Unit-Tests-Stand
value = 1E-11
reg = 1,255679053E-11

[mariuszgromada]

I believe I am facing consequence of this https://stackoverflow.com/questions/65339515/double-to-string-conversion-different-for-net-framework-and-net-core

[mariuszgromada]

I am using to string conversion in canonical rounding.

[darxis]

I am using to string conversion in canonical rounding.

That could be the cause, I think we should take a look at commits between v5.0.6 and v5.0.7 because the bug is not present in v5.0.6 and is present in v5.0.7.

[mariuszgromada]

No need. I have fixed the bug in the canonical rounding that was present since 4.4.2.

As of now please disable canonical rounding and you are good to go.

I will try do identify how to solve the difference between frameworks.

Best regards

[mariuszgromada]

Fixed