// Copyright Carlos Scheidegger, 2016

// License: Apache 2.0 (same as Feather itself)

//

// A minimal JavaScript implementation of feather file reading. Notably,

// this only currently supports plain encodings, doesn't support null values,

// and doesn't support int64, uint64, or boolean data.

//

// This uses https://github.com/google/flatbuffers/blob/master/js/flatbuffers.js

// and metadata_generated.js

//

// if metadata.fbs changes, you will need to regenerate metadata_generated.js,

// by

//

// $ flatc --js metadata.fbs

//

//////////////////////////////////////////////////////////////////////////////

// If you want to bowerize/browserify/complexify this thing, go right ahead.

(function(module) {

var typeEnum = {

1: Int8Array,

2: Int16Array,

3: Int32Array,

5: Uint8Array,

6: Uint16Array,

7: Uint32Array,

9: Float32Array,

10: Float64Array

};

var sizeEnum = {

1: 1,

2: 2,

3: 4,

5: 1,

6: 2,

7: 4,

9: 2,

10: 4

};

/**

* convert a feather file loaded as an ArrayBuffer to an object whose

* slots are the columns of the feather file, and whose values are Typed Arrays

* which share the buffer.

*

* This should be about as efficient as we can get in JavaScript.

*

* NB: This means that if you mutate the buffer, the values in the

* typed arrays will change (and vice versa!)

*

* @param {ArrayBuffer} buffer The buffer containing the feather file to convert.

*/

function convertToDataFrame(buffer)

{

var FEA1 = 826361158;

var l = buffer.byteLength;

if (l < 8) {

throw new Error("Feather file needs to be at least 8 bytes, got " + l + " instead.");

}

var byteView = new Uint8Array(buffer, 0, l);

var intView = new Uint32Array(buffer, 0, l/4);

var beginMagic = intView[0];

var endMagic = intView[intView.length-1];

if (beginMagic !== FEA1) {

var str = String.fromCharCode(byteView[0]) +

String.fromCharCode(byteView[1]) +

String.fromCharCode(byteView[2]) +

String.fromCharCode(byteView[3]);

throw new Error("Expected magic number at BOF to be FEA1, got " + str + " instead.");

}

if (endMagic !== FEA1) {

var str = String.fromCharCode(byteView[l-4]) +

String.fromCharCode(byteView[l-3]) +

String.fromCharCode(byteView[l-2]) +

String.fromCharCode(byteView[l-1]);

throw new Error("Expected magic number at EOF to be FEA1, got " + str + " instead.");

}

var metadataLength = intView[intView.length-2];

var metadata = buffer.slice(buffer.byteLength-8-metadataLength, buffer.byteLength-8);

var bb = new flatbuffers.ByteBuffer(new Uint8Array(metadata));

var r = feather.fbs.CTable.getRootAsCTable(bb);

var nCols = r.columnsLength();

var d = {};

for (var i=0; i<nCols; ++i) {

var c = r.columns(i);

var name = c.name();

var values = c.values();

if (values.encoding() !== 0) {

throw new Error("Only PLAIN encoding is currently supported.");

}

var offset = values.offset().toFloat64();

var nullCount = values.nullCount().toFloat64();

if (nullCount !== 0) {

throw new Error("Null values not currently supported");

}

var nElements = values.length().toFloat64();

var type = values.type();

if (typeEnum[type] === undefined) {

throw new Error("Unsupported datatype");

}

d[name] = new typeEnum[type](buffer, offset, nElements);

}

return d;

}

module.convertToDataFrame = convertToDataFrame;

}(window));