define ['angular', 'niftiParser', 'ndarray'], (ng, niftiParser, ndarray) ->
  nifti = ng.module 'qiprofile.nifti', []

  nifti.factory 'Nifti', ->
    # Unquoted whitespace matcher (doesn't handle escapes).
    WHITESPACE_REGEX = /\s+(?=([^"]*"[^"]*")*[^"]*$)/g
    
    # Parses the image file content. Returns the
    # {header, data}, where:
    # * *header* is an object {nifti, nrrd, dicom}, where *nifti* is
    #   is the NIfTI header, NRRD is the NRRD header and *dicom* is
    #   the embedded DICOM meta-data object
    # * *data* is the image binary ndarray in dimension order
    #   [x, y, z, time], where *z* is the slice index and *time*
    #   is the volume index.
    # 
    # @buffer the binary image file ArrayBuffer 
    # @returns the parsed image {header, data} object
    parse: (buffer) ->
      # Extract the NIfTI header.
      niftiHeader = niftiParser.parseNIfTIHeader(buffer)
      # Extract the NRRD header.
      nrrdHeader = niftiParser.parseNRRDHeader(buffer)
      
      # Extract the embedded DICOM metadata. 
      extensions = niftiParser.parseHeaderExtensions(buffer)
      if extensions.length > 1
        throw new Error("The NIfTI file has more than one extension")
      if extensions.length == 1
        extension = extensions[0]
        # Note: the conventional idiom:
        #   String.fromCharCode.apply(null, data)
        # results in a stack overflow. The work-around is to convert
        # the characters one byte at a time.
        chars = (String.fromCharCode(c) for c in extension.data)
        # Get rid of whitespace.
        json = chars.join('').replace(WHITESPACE_REGEX, '')
        # Note: JSON parse fails with message that it can't find
        # a JSON object. The work-around to this work-around bug is
        # to search for the substrings we want and hammer together
        # a minimal JSON string.
        # TODO - unravel this mess.
        windowCenterRegex = /("WindowCenter":\[[^\]]+\])/
        windowWidthRegex = /("WindowWidth":\[[^\]]+\])/
        matches = windowCenterRegex.exec(json)
        if matches?
          item1 = matches[1]
          matches = windowWidthRegex.exec(json)
          if matches?
            item2 = matches[1]
            json = '{' + [item1, item2].join(',') + '}'
            # Parse the JSON.
            dicom = JSON.parse(json)
          else
            dicom = null
        else
          dicom = null
      else
        dicom = null

      # Extract the image binary data.
      parsed = niftiParser.parse(buffer)
      # The binary image ndarray. Per the nifti-js site, the NIfTI
      # header sizes attribute is the inverse of the raw NIfTI
      # ndarray order, and suggests reversing the sizes in the
      # ndarray constructor argument below. There is (of course)
      # no rationale given for this. We dispense with this advice
      # and retain the input sizes, which results in a data shape
      # consistent with the input shape.
      # 
      # The dcmstack layout is in [time, x, y, z] order. This differs
      # from the expected [x, y, z, time] order described
      # in, e.g., https://brainder.org/2012/09/23/the-nifti-file-format/,
      # We retain the input layout, which has ramifications for the
      # array slicing and dicing in sliceDisplay.coffee.
      #
      # Note: the js ndarray default stride differs from the expected
      # nibabel default stride, after accounting for the difference
      # in the underlying datum size (byte for nibabel, int16 for js).
      # Reversing the sizes as described above still results in an
      # incorrect stride. The work-around is to calculate the stride
      # by hand.
      # TODO - bring this up with the ndarray dev team.
      stride = (parsed.sizes[0...i].reduce(_.multiply, 1) for i in [0...parsed.sizes.length]) 
      data = ndarray(parsed.data, parsed.sizes, stride)
      
      # Return the image {header, data} object.
      header:
        nifti: niftiHeader
        nrrd: nrrdHeader
        dicom: dicom
      data: data