/*! Parser: metric *//*
* Demo: http://jsfiddle.net/Mottie/abkNM/382/
* Set the metric name in the header (defaults to 'm|meter'), e.g.
*
HDD Size |
* Distance |
*/
/*jshint jquery:true */
;( function( $ ) {
'use strict';
var prefixes = {
// 'prefix' : [ base 10, base 2 ]
// skipping IEEE 1541 defined prefixes: kibibyte, mebibyte, etc, for now.
'Y|Yotta|yotta' : [ 1e24, Math.pow(1024, 8) ], // 1024^8
'Z|Zetta|zetta' : [ 1e21, Math.pow(1024, 7) ], // 1024^7
'E|Exa|exa' : [ 1e18, Math.pow(1024, 6) ], // 1024^6
'P|Peta|peta' : [ 1e15, Math.pow(1024, 5) ], // 1024^5
'T|Tera|tera' : [ 1e12, Math.pow(1024, 4) ], // 1024^4
'G|Giga|giga' : [ 1e9, Math.pow(1024, 3) ], // 1024^3
'M|Mega|mega' : [ 1e6, Math.pow(1024, 2) ], // 1024^2
'k|Kilo|kilo' : [ 1e3, 1024 ], // 1024
// prefixes below here are rarely, if ever, used in binary
'h|hecto' : [ 1e2, 1e2 ],
'da|deka' : [ 1e1, 1e1 ],
'd|deci' : [ 1e-1, 1e-1 ],
'c|centi' : [ 1e-2, 1e-2],
'm|milli' : [ 1e-3, 1e-3 ],
'µ|micro' : [ 1e-6, 1e-6 ],
'n|nano' : [ 1e-9, 1e-9 ],
'p|pico' : [ 1e-12, 1e-12 ],
'f|femto' : [ 1e-15, 1e-15 ],
'a|atto' : [ 1e-18, 1e-18 ],
'z|zepto' : [ 1e-21, 1e-21 ],
'y|yocto' : [ 1e-24, 1e-24 ]
},
// the \\d+ will not catch digits with spaces, commas or decimals; so use the value from n instead
RegLong = '(\\d+)(\\s+)?([Zz]etta|[Ee]xa|[Pp]eta|[Tt]era|[Gg]iga|[Mm]ega|kilo|hecto|deka|deci|centi|milli|micro|nano|pico|femto|atto|zepto|yocto)(',
RegAbbr = '(\\d+)(\\s+)?(Z|E|P|T|G|M|k|h|da|d|c|m|µ|n|p|f|a|z|y)(',
// make these case-insensitive because we all forget the case for these binary values
byteTest = /^[b|bit|byte|o|octet]/i;
$.tablesorter.addParser({
id: 'metric',
is: function() {
return false;
},
format: function(txt, table, cell, cellIndex) {
var unit, isBinary, nameLong, nameAbbr,
// default base unit name
base = 'm|meter',
// process number here to get a numerical format (us or eu)
num = $.tablesorter.formatFloat( txt.replace(/[^\w,. \-()]/g, ''), table ),
$t = table.config.$headerIndexed[ cellIndex ],
regex = $t.data( 'metric' );
if ( !regex ) {
// stored values
unit = ( $t.attr('data-metric-name') || base ).split( '|' );
nameLong = $t.attr( 'data-metric-name-full' ) || '';
nameAbbr = $t.attr( 'data-metric-name-abbr' ) || '';
regex = [ nameLong || unit[1] || unit[0].substring(1), nameAbbr || unit[0] ];
isBinary = byteTest.test( regex.join( '' ) );
// adding 'data-metric-name-full' which would contain 'byte|BYTE|Byte' etc
regex[2] = new RegExp( RegLong + (
( nameLong === '' ? '' : nameLong + '|' + nameAbbr ) ||
// with data-metric-name='b|byte', we end up with 'b|B|byte|BYTE' - maybe not the best solution for case-insensitivity
( ( isBinary ? regex[0].toLowerCase() + '|' + regex[0].toUpperCase() : regex[0] ) + '|' +
( isBinary ? regex[1].toLowerCase() + '|' + regex[1].toUpperCase() : regex[1] ) ) ) +
')' );
// adding 'data-metric-name-abbr' which would contain 'b|B' etc
regex[3] = new RegExp( RegAbbr + ( nameAbbr ||
( ( isBinary ? regex[1].toLowerCase() + '|' + regex[1].toUpperCase() : regex[1] ) ) ) +
')' );
$t.data( 'metric', regex );
}
// find match to full name or abbreviation
unit = txt.match( regex[2] ) || txt.match( regex[3] );
if ( unit ) {
for ( base in prefixes ) {
if ( unit[3].match( base ) ) {
// exception when using binary prefix
// change base for binary use
isBinary = byteTest.test( unit[4] ) ? 1 : 0;
return num * prefixes[ base ][ isBinary ];
}
}
}
return num;
},
type: 'numeric'
});
})(jQuery);