GNU Octave wrapper around RapidJSON

These wrapper doesn't try to be compatible with jsondecode/jsonencode or JSONlab.

Instead it tries to map GNU Octave data structures as close as possible to JSON and back. So the major design goal is that in GNU Octave

assert (load_json (save_json (foo)), foo)

works for almost any type of "foo".

Importing data from a JSON string tries to preserve the datatype whenever possible and if this doesn't slown down the import.

octave> x = load_json ('{"a":5, "b":3.14156}')
x =

  scalar structure containing the fields:

    a = 5
    b =  3.1416

octave> typeinfo (x.a)
ans = uint32 scalar
octave> typeinfo (x.b)
ans = scalar

N-dimensional matrices are returned as matrix (aka "double matrix")

octave> x = load_json ('[3,4,5]')
x =

   3   4   5

octave> typeinfo (x)
ans = matrix

As soon as one element in a JSON array can't mapped to a double value, a cell is returned:

octave> x = load_json ('[2, "foo", 3.14]')
x =
{
  [1,1] =  2
  [1,2] = foo
  [1,3] =  3.1400
}

If you want to parse from a file use load_json (fileread (filename))

• Saving a matrix always gives double values in JSON

Converting a struct array to JSON "unpacks" the struct array but I've been told by javascript programmers that this would be
the expected behaviour...

octave> foo(1).a = 3;
octave> foo(1).b = pi;
octave> foo(2).a = 5;
octave> foo(2).b = 2.718;
octave> save_json (foo)
ans = [
    {
        "a": 3.0,
        "b": 3.141592653589793
    },
    {
        "a": 5.0,
        "b": 2.718
    }
]

If you want to read it back (and collapse the unpacked struct) you can do:

octave> load_json (ans)
ans =
{
  [1,1] =

    scalar structure containing the fields:

      a =  3
      b =  3.1416

  [1,2] =

    scalar structure containing the fields:

      a =  5
      b =  2.7180

}

octave> x = [ans{:}]
x =

  1x2 struct array containing the fields:

    a
    b

octave:1> save_json (NA)
ans = null
octave:2> save_json ([])
ans = []
octave:3> save_json (NaN)
ans = NaN
octave:4> load_json ("null")
ans = NA
octave:4> load_json ("[]")
ans = [](0x0)
octave:5> load_json ("NaN")
ans = NaN

Using "NaN" in JSON is an extension and may not be comaptible with other libraries.

If you want to start using the C++ API liboctave here are some links to useful resources:

• chapter "Oct-Files" in the GNU Octave manual
• Doxygen generated documentation of liboctave
• Manually generated class overview for liboctave on the wiki
• C++ Tips and Tricks section on the wiki has a table with quivalent C++ code. Very usefull if you already know the m-file syntax
• C++ code styleguide on the wiki
• Developer page on the wiki

• https://github.com/fangq/jsonlab (pure .m files)
• https://github.com/gllmflndn/JSONio (uses C lib jsmn)