sheetjs/bits/36_xlsxutils.js
SheetJS 972897c9ae version bump 0.5.1: field reorg
- cell.r holds the raw rich text, cell.h now holds the html
  (version 0.5.0 cell.raw -> cell.r, cell.r -> cell.h)
- similar changes to comments and XLSB generation
- removed some unused binary methods
2014-02-05 08:39:21 -05:00

83 lines
2.3 KiB
JavaScript

var attregexg=/(\w+)=((?:")([^"]*)(?:")|(?:')([^']*)(?:'))/g;
var attregex=/(\w+)=((?:")(?:[^"]*)(?:")|(?:')(?:[^']*)(?:'))/;
function parsexmltag(tag) {
var words = tag.split(/\s+/);
var z = {'0': words[0]};
if(words.length === 1) return z;
(tag.match(attregexg) || []).map(
function(x){var y=x.match(attregex); z[y[1]] = y[2].substr(1,y[2].length-2); });
return z;
}
function evert(obj) {
var o = {};
Object.keys(obj).forEach(function(k) { if(obj.hasOwnProperty(k)) o[obj[k]] = k; });
return o;
}
var encodings = {
'"': '"',
''': "'",
'>': '>',
'&lt;': '<',
'&amp;': '&'
};
var rencoding = evert(encodings);
var rencstr = "&<>'\"".split("");
// TODO: CP remap (need to read file version to determine OS)
function unescapexml(text){
var s = text + '';
for(var y in encodings) s = s.replace(new RegExp(y,'g'), encodings[y]);
return s.replace(/_x([0-9a-fA-F]*)_/g,function(m,c) {return _chr(parseInt(c,16));});
}
function escapexml(text){
var s = text + '';
rencstr.forEach(function(y){s=s.replace(new RegExp(y,'g'), rencoding[y]);});
return s;
}
function parsexmlbool(value, tag) {
switch(value) {
case '0': case 0: case 'false': case 'FALSE': return false;
case '1': case 1: case 'true': case 'TRUE': return true;
default: throw "bad boolean value " + value + " in "+(tag||"?");
}
}
var utf8read = function(orig) {
var out = "", i = 0, c = 0, c1 = 0, c2 = 0, c3 = 0;
while (i < orig.length) {
c = orig.charCodeAt(i++);
if (c < 128) out += _chr(c);
else {
c2 = orig.charCodeAt(i++);
if (c>191 && c<224) out += _chr((c & 31) << 6 | c2 & 63);
else {
c3 = orig.charCodeAt(i++);
out += _chr((c & 15) << 12 | (c2 & 63) << 6 | c3 & 63);
}
}
}
return out;
};
// matches <foo>...</foo> extracts content
function matchtag(f,g) {return new RegExp('<'+f+'(?: xml:space="preserve")?>([^\u2603]*)</'+f+'>',(g||"")+"m");}
function parseVector(data) {
var h = parsexmltag(data);
var matches = data.match(new RegExp("<vt:" + h.baseType + ">(.*?)</vt:" + h.baseType + ">", 'g'))||[];
if(matches.length != h.size) throw "unexpected vector length " + matches.length + " != " + h.size;
var res = [];
matches.forEach(function(x) {
var v = x.replace(/<[/]?vt:variant>/g,"").match(/<vt:([^>]*)>(.*)</);
res.push({v:v[2], t:v[1]});
});
return res;
}
function isval(x) { return typeof x !== "undefined" && x !== null; }