无法读取NetCDF结构变量数组子范围

Question

I have an NCDF structure being retrieved from an HDF5 file, and I'm trying to read a subrange of an array variable using the Java NetCDF 4.3.16 library. I find the variable like this:

Variable netCDFVariable = netCDFFile.findVariable("/group/struct.var");

At this point I can read the entire variable if I want:

netCDFArray = netCDFVariable.read();

But let's say that the variable is an integer array of length 10, and I want to read just the subrange of indexes 3, 4, 5, 6, and 7:

Section section=new Section(new int[]{3}, new int[]{5});
netCDFArray = netCDFVariable.read(section);

But here Variable.read(Section) throws an InvalidRangeException . Tracing through the code, the variable checks its own internal shape variable, which happens to be [] and finds that to be invalid.

So why doesn't the variable have an appropriate shape ? Do I have to read the variable first to get its correct shape? (That would of course defeat the purpose of trying to read a subrange in the first place.)

Answer 1

As of NetCDF 4.3.17 the implementations of Variable.read() and Variable.read(Section) differ considerably when reading a member of an array of structures, neither adhering to the API documentation. Below I present a workaround.

For illustration, let's say I have the following:

Structure {
  int foo;
    :_Unsigned = "true";
  int bar;
    :_Unsigned = "true";
} example(24);

This is a 24-element array named example of structures, each structure containing two members, foo and bar . Let us assume I get a reference to the foo member like so:

final Variable fooVariable = netcdfFile.findVariable("/blah/example.foo");

If I call fooVariable.read() , the API documentation says that I will get back the first value, because foo is part of a structure that is an element in an array. This is not what happens; instead, the library actually does some clever reading and returns the foo member from all the structures as a single array of foo values. This is the behavior I desire.

Unfortunately the fooVariable.read(Section) implementation does not have the same bit of clever code as does fooVariable.read() , and instead is written to throw an UnsupportedOperationException . (Through the lack of an added check the code doesn't even get that far, throwing an InvalidRangeException because it thinks the given Section is invalid. This is a pity, because (once a check is added to avoid an InvalidRangeException ) the clever code from the Variable.read() implementation would work equally well for Variable.read(Section) with the insertion of only a single method argument on a single line!

Using the clever code from Variable.read() I have created a method that works around this problem, allowing a caller to ask for a section of an array with any variable. If the variable is a member of a structure in an array, the subrange of just that member will be read, effectively creating another version of the fooVariable.read(Section) method that has the same behavior as fooVariable.read() for structure members:

  /**
   * Reads an array of data from the provided NetCDF variable. If the variable is a member of a structure, that member is read
   * from all the structures in the array and returned as a single array containing that member from each structure.
   * @param variable The NetCDF variable to read.
   * @param section The section indicating the element of the array to read
   * @param indexEnd The ending source index, exclusive, or -1 if all available values should be read.
   * @return An array representing the requested range of values read for the given variable.
   * @throws IOException if there is an error reading the data.
   */
  public static Array readArray(final Variable variable, final Section section) throws IOException, InvalidRangeException {
    if (variable.isMemberOfStructure()) { //if the variable is member of a structure
      final Structure parentStructure = variable.getParentStructure().select(variable.getShortName()); //select just the member variable
      final ArrayStructure arrayStructure = (ArrayStructure) parentStructure.read(section); //read the array of structures
      return arrayStructure.extractMemberArray(arrayStructure.findMember(variable.getShortName())); //extract just the member into an array
    } else { //if the variable is not a member of a structure
      return variable.read(section); //just read the section directly from the variable
    }
  }

I reported this problem (now identified as NRW-974703) with Unidata NetCDF Support . At first I was told that these methods didn't apply to HDF5 and only apply to Unidata's own NetCDF file format. (This is completely incorrect.) Then I was told that I didn't understand the Java NetCDF API (although from the previous answer I question who exactly lacks understanding of the Unidata library). I have not yet received a response from Unidata after tracing down the specific problem code and providing the above workaround.