The general structure of utility programs, such as the NCO utilities or programs to copy netCDF data from a source file to a target file is straightforward:

• Find out how many dimensions are in the source file using a file inquire function.
• For each dimension, find out its name and length and create a corresponding dimension in the output file. Preserve the unlimited dimension, if needed.
• Find out how many variables are in the source file using a file inquire function.
• For each source variable, find out its name, type, shape, and number of attributes. Create a corresponding variable in the output file. Loop through its attributes in the source file and create corresponding attributes, including all their values in the target file.
• For each global (file-level) attribute, find its name, type, and values. Create corresponding global attribute in the target file.
• For each variable, copy all its data values from the source file to the target file.

This last step can be tricky, because:

• The code must work for a variable with any number of dimensions, so a simple set of nested loops with one loop for each dimension cannot be used.
• A variable may have more data than can fit in memory, so it is impractical to read all the values into memory at once and then write them all out.
• Even a single row of variable values may be too large to fit in memory, for example a variable of type double with a most rapidly varying dimension of size 2³¹ . One solution: treat dimension indices like an odometer to iterate through variable data in a single loop.

NetCDF-4 note:  The enhanced data model for netCDF-4 adds arbitrarily nested groups and nested user-defined data types that include variable length arrays. These add a need for recursion and object-oriented programming to deal with a potentially infinite number of types.

2010 Unidata NetCDF Workshop > Introduction to the NetCDF APIs and Example Programs