3 tests fail

[yurivict]

          Start  356: NumberToString.DecimalNotationUpTo21Digits
 360/2979 Test  #356: NumberToString.DecimalNotationUpTo21Digits .............................................................***Failed    0.09 sec
Running main() from /wrkdirs/usr/ports/devel/googletest/work/googletest-release-1.12.1/googletest/src/gtest_main.cc
Note: Google Test filter = NumberToString.DecimalNotationUpTo21Digits
[==========] Running 1 test from 1 test suite.
[----------] Global test environment set-up.
[----------] 1 test from NumberToString
[ RUN      ] NumberToString.DecimalNotationUpTo21Digits
/disk-samsung/freebsd-ports/science/InsightToolkit/work/ITK-5.2.1/Modules/Core/Common/test/itkNumberToStringGTest.cxx:87: Failure
Expected equality of these values:
  numberToString(power_of_ten)
    Which is: "100000010000"
  '1' + std::string(exponent, '0')
    Which is: "100000000000"
/disk-samsung/freebsd-ports/science/InsightToolkit/work/ITK-5.2.1/Modules/Core/Common/test/itkNumberToStringGTest.cxx:88: Failure
Expected equality of these values:
  numberToString(-power_of_ten)
    Which is: "-100000010000"
  "-1" + std::string(exponent, '0')
    Which is: "-100000000000"
/disk-samsung/freebsd-ports/science/InsightToolkit/work/ITK-5.2.1/Modules/Core/Common/test/itkNumberToStringGTest.cxx:96: Failure
Expected equality of these values:
  numberToString(power_of_ten)
    Which is: "0.000009999999999999999"
  "0." + std::string(-1 - exponent, '0') + '1'
    Which is: "0.00001"
/disk-samsung/freebsd-ports/science/InsightToolkit/work/ITK-5.2.1/Modules/Core/Common/test/itkNumberToStringGTest.cxx:97: Failure
Expected equality of these values:
  numberToString(-power_of_ten)
    Which is: "-0.000009999999999999999"
  "-0." + std::string(-1 - exponent, '0') + '1'
    Which is: "-0.00001"
[  FAILED  ] NumberToString.DecimalNotationUpTo21Digits (0 ms)
[----------] 1 test from NumberToString (0 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test suite ran. (0 ms total)
[  PASSED  ] 0 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] NumberToString.DecimalNotationUpTo21Digits

 1 FAILED TEST

          Start 1201: itkTIFFImageIOPlannarConfig2
1205/2979 Test #1201: itkTIFFImageIOPlannarConfig2 ...........................................................................***Exception: SegFault  0.08 sec
Trying reader->Update()             

          Start 1204: itkTIFFImageIOInfoTest2
1208/2979 Test #1204: itkTIFFImageIOInfoTest2 ................................................................................***Exception: SegFault  0.08 sec

Version: 5.2.1
clang-14
FreeBSD 13.1

[dzenanz]

@N-Dekker If I remember correctly, you might be fresh with NumberToString?

[N-Dekker]

@yurivict Interesting 🤔 Regarding the itkNumberToStringGTest failures: Did you possibly do anything to tweak the floating point behavior of the compiler? As in https://clang.llvm.org/docs/UsersManual.html#controlling-floating-point-behavior ?

[N-Dekker]

@yurivict It would be helpful if you could reproduce those float-to-string conversion errors (for example from 100000000000.0 to "100000010000" by using the double-conversion library from https://github.com/google/double-conversion directly (without ITK). Doing something like the following, but outside of ITK:

ITK/Modules/Core/Common/src/itkNumberToString.cxx

Lines 43 to 60 in 38696b2

	template <typename TValue>
	std::string
	FloatingPointNumberToString(const TValue val)
	{
	// Declare a buffer, large enough for strings like:
	// "-100000000000000000000" (-1e20, either float or double, 23 chars)
	// "-1.7976931348623157e+308" (-DBL_MAX, 25 chars)
	// "-0.0000033333333333333333" (-3e-005/9.0, 26 chars)
	char buf[32];
	double_conversion::StringBuilder builder(buf, sizeof(buf));
	if (!ConvertToShortest(double_conversion::DoubleToStringConverter::EcmaScriptConverter(), val, builder))
	{
	itkGenericExceptionMacro(<< "Conversion failed for " << val);
	}
	return std::string(builder.Finalize());
	}

Would you be able to?

[yurivict]

The conversion prints: 100000000000.0 -> 100000000000:

#include <double-conversion/double-to-string.h>
#include <iostream>

bool
ConvertToShortest(const double_conversion::DoubleToStringConverter & converter,
                  const float                                        val,
                  double_conversion::StringBuilder &                 builder)
{
  // Call the converter member function that is specific for single-precision `float`.
  return converter.ToShortestSingle(val, &builder);
}


template <typename TValue>
std::string
FloatingPointNumberToString(const TValue val)
{
  // Declare a buffer, large enough for strings like:
  // "-100000000000000000000" (-1e20, either float or double, 23 chars)
  // "-1.7976931348623157e+308" (-DBL_MAX, 25 chars)
  // "-0.0000033333333333333333" (-3e-005/9.0, 26 chars)
  char buf[32];

  double_conversion::StringBuilder builder(buf, sizeof(buf));

  if (!ConvertToShortest(double_conversion::DoubleToStringConverter::EcmaScriptConverter(), val, builder))
    std::cout << "Conversion failed for " << val << std::endl;

  return std::string(builder.Finalize());
}

int main() {
        std::cout << "100000000000.0 -> " << FloatingPointNumberToString<double>(100000000000.0) << std::endl;
}

[N-Dekker]

@yurivict Thanks for trying. 👍 So it looks like converter.ToShortestSingle(val, &builder) works fine. Can you please also try converter.ToShortest(val, &builder)? (ToShortestSingle is specific for 32-bit float, while ToShortest is for 64-bit double.) So just the same test, only without the word "Single" in the code 😃

[yurivict]

Without Single it prints this:

100000000000.0 -> 99999997952

[N-Dekker]

Sorry, my bad. When you remove "Single", also change the type of the val parameter from float to double, please! Just like the ConvertToShortest overload for a double val, in itkNumberToString.cxx

ITK/Modules/Core/Common/src/itkNumberToString.cxx

Lines 26 to 32 in 38696b2

	bool
	ConvertToShortest(const double_conversion::DoubleToStringConverter & converter,
	const double val,
	double_conversion::StringBuilder & builder)
	{
	return converter.ToShortest(val, &builder);
	}

[yurivict]

100000000000.0 -> 100000000000

[yurivict]

5.3rc04 has 4 cases failing on FreeBSD 13.1 STABLE:

	129 - ITKMeshInDoxygenGroup (Failed)
	365 - NumberToString.DecimalNotationUpTo21Digits (Failed)
	1264 - itkTIFFImageIOPlannarConfig2 (SEGFAULT)
	1267 - itkTIFFImageIOInfoTest2 (SEGFAULT)

[N-Dekker]

@yurivict Can you please check if the NumberToString.DecimalNotationUpTo21Digits failures may be solved, just by adding one extra #include to Modules/Core/Common/test/itkNumberToStringGTest.cxx, as follows:

#include <cmath> // For std::pow.

(At line 22, right after #include <gtest/gtest.h>)

In other words, does pull request #3751 indeed fix those NumberToString.DecimalNotationUpTo21Digits failures?

[yurivict]

After more modules were enabled more tests are now failing: test.log

[N-Dekker]

@yurivict Now that PR #3751 is merged, can you please rerun the NumberToString GoogleTest? PR #3751 might not have fixed the issue, but it should at least provide some more information. Specifically: do the failures occur with 32-bit float, or 64-bit double, or both?