BUG: VTKPolyDataMeshIO should not hang on inf and NaN in ASCII .vtk file

N-Dekker · hjmjohnson · commit 678b97eaebed · 2022-08-12T09:07:49.000-05:00
`MeshFileReader`/`VTKPolyDataMeshIO` did hang forever when trying to read an ASCII .vtk file that contained non-numeric chars in "POINTS" section, for example "Infinity" or "NaN". With this commit, "inf" and "Infinity" will be interpreted as the floating point value infinity, while "nan" and "NaN" will be interpreted as Not-A-Number. Other strings of non-numeric characters in the list of point components (coordinates) will cause the reader to throw an appropriate `itk::ExceptionObject`. Addresses issue #3539 "MeshFileReader/VTKPolyDataMeshIO hangs forever on non-numeric chars in "POINTS" section (including "Infinity" and "NaN")"
diff --git a/Modules/IO/MeshVTK/include/itkVTKPolyDataMeshIO.h b/Modules/IO/MeshVTK/include/itkVTKPolyDataMeshIO.h
@@ -205,11 +205,7 @@ class ITKIOMeshVTK_EXPORT VTKPolyDataMeshIO : public MeshIOBase
       if (line.find("POINTS") != std::string::npos)
       {
         /**  Load the point coordinates into the itk::Mesh */
-        SizeValueType numberOfComponents = this->m_NumberOfPoints * this->m_PointDimension;
-        for (SizeValueType ii = 0; ii < numberOfComponents; ++ii)
-        {
-          inputFile >> buffer[ii];
-        }
+        Self::ReadComponentsAsASCII(inputFile, buffer, this->m_NumberOfPoints * this->m_PointDimension);
       }
     }
   }
@@ -281,11 +277,8 @@ class ITKIOMeshVTK_EXPORT VTKPolyDataMeshIO : public MeshIOBase
         }
 
         /** for VECTORS or NORMALS or TENSORS, we could read them directly */
-        SizeValueType numberOfComponents = this->m_NumberOfPointPixels * this->m_NumberOfPointPixelComponents;
-        for (SizeValueType ii = 0; ii < numberOfComponents; ++ii)
-        {
-          inputFile >> buffer[ii];
-        }
+        Self::ReadComponentsAsASCII(
+          inputFile, buffer, this->m_NumberOfPointPixels * this->m_NumberOfPointPixelComponents);
       }
     }
   }
@@ -376,11 +369,8 @@ class ITKIOMeshVTK_EXPORT VTKPolyDataMeshIO : public MeshIOBase
         }
 
         /** for VECTORS or NORMALS or TENSORS, we could read them directly */
-        SizeValueType numberOfComponents = this->m_NumberOfCellPixels * this->m_NumberOfCellPixelComponents;
-        for (SizeValueType ii = 0; ii < numberOfComponents; ++ii)
-        {
-          inputFile >> buffer[ii];
-        }
+        Self::ReadComponentsAsASCII(
+          inputFile, buffer, this->m_NumberOfCellPixels * this->m_NumberOfCellPixelComponents);
       }
     }
   }
@@ -1150,6 +1140,30 @@ class ITKIOMeshVTK_EXPORT VTKPolyDataMeshIO : public MeshIOBase
   /** Convenience method returns the IOComponentEnum corresponding to a string. */
   IOComponentEnum
   GetComponentTypeFromString(const std::string & pointType);
+
+private:
+  /** Reads the specified number of components from the specified input file into the specified buffer.
+   * \note This member function is overloaded for `float` and `double`, in order to support reading infinity and NaN
+   * values.
+   */
+  template <typename T>
+  static void
+  ReadComponentsAsASCII(std::ifstream & inputFile, T * const buffer, const SizeValueType numberOfComponents)
+  {
+    for (SizeValueType i = 0; i < numberOfComponents; ++i)
+    {
+      if (!(inputFile >> buffer[i]))
+      {
+        itkGenericExceptionMacro("Failed to read a component from the specified ASCII input file!");
+      }
+    }
+  }
+
+  static void
+  ReadComponentsAsASCII(std::ifstream & inputFile, float * const buffer, const SizeValueType numberOfComponents);
+
+  static void
+  ReadComponentsAsASCII(std::ifstream & inputFile, double * const buffer, const SizeValueType numberOfComponents);
 };
 } // end namespace itk
 
diff --git a/Modules/IO/MeshVTK/src/itkVTKPolyDataMeshIO.cxx b/Modules/IO/MeshVTK/src/itkVTKPolyDataMeshIO.cxx
@@ -19,10 +19,69 @@
 #include "itkVTKPolyDataMeshIO.h"
 
 #include "itksys/SystemTools.hxx"
+
+#include <double-conversion/string-to-double.h>
+
 #include <fstream>
+#include <limits>
 
 namespace itk
 {
+
+namespace
+{
+template <typename TFloatingPoint>
+void
+ReadFloatingPointsAsASCII(std::ifstream &        inputFile,
+                          TFloatingPoint * const buffer,
+                          const SizeValueType    numberOfFloatingPoints)
+{
+  std::string str;
+
+  for (SizeValueType i = 0; i < numberOfFloatingPoints; ++i)
+  {
+    buffer[i] = [&str, &inputFile] {
+      if (inputFile >> str)
+      {
+        using NumericLimits = std::numeric_limits<TFloatingPoint>;
+
+        if ((str == "NaN") || (str == "nan"))
+        {
+          return NumericLimits::quiet_NaN();
+        }
+        if (str == "Infinity")
+        {
+          return NumericLimits::infinity();
+        }
+        if (str == "-Infinity")
+        {
+          return -NumericLimits::infinity();
+        }
+        const auto numberOfChars = str.size();
+
+        if (numberOfChars <= std::numeric_limits<int>::max())
+        {
+          constexpr auto                                   double_NaN = std::numeric_limits<double>::quiet_NaN();
+          int                                              processedCharCount{ 0 };
+          const double_conversion::StringToDoubleConverter converter(0, double_NaN, double_NaN, "inf", "nan");
+          const auto                                       conversionResult =
+            converter.StringTo<TFloatingPoint>(str.c_str(), static_cast<int>(numberOfChars), &processedCharCount);
+
+          if (processedCharCount == static_cast<int>(numberOfChars) && !std::isnan(conversionResult))
+          {
+            return conversionResult;
+          }
+        }
+      }
+      itkGenericExceptionMacro("Failed to read a floating point component from the specified ASCII input file!"
+                               << (str.empty() ? "" : (" Read characters: \"" + str + "\"")));
+    }();
+  }
+}
+
+} // namespace
+
+
 // Constructor
 VTKPolyDataMeshIO ::VTKPolyDataMeshIO()
 {
@@ -1584,4 +1643,21 @@ VTKPolyDataMeshIO ::PrintSelf(std::ostream & os, Indent indent) const
     os << indent << "cellTensorDataName : " << dataName << std::endl;
   }
 }
+
+void
+VTKPolyDataMeshIO::ReadComponentsAsASCII(std::ifstream &     inputFile,
+                                         float * const       buffer,
+                                         const SizeValueType numberOfComponents)
+{
+  ReadFloatingPointsAsASCII(inputFile, buffer, numberOfComponents);
+}
+
+void
+VTKPolyDataMeshIO::ReadComponentsAsASCII(std::ifstream &     inputFile,
+                                         double * const      buffer,
+                                         const SizeValueType numberOfComponents)
+{
+  ReadFloatingPointsAsASCII(inputFile, buffer, numberOfComponents);
+}
+
 } // end of namespace itk
