2013-03-31 22:56:45 +00:00
|
|
|
|
|
|
|
|
|
var ct2type = {
|
|
|
|
|
"application/vnd.openxmlformats-officedocument.spreadsheetml.sheet.main+xml": "workbooks",
|
|
|
|
|
"application/vnd.openxmlformats-package.core-properties+xml": "coreprops",
|
|
|
|
|
"application/vnd.openxmlformats-officedocument.extended-properties+xml": "extprops",
|
|
|
|
|
"application/vnd.openxmlformats-officedocument.spreadsheetml.calcChain+xml": "calcchains",
|
|
|
|
|
"application/vnd.openxmlformats-officedocument.spreadsheetml.worksheet+xml":"sheets",
|
|
|
|
|
"application/vnd.openxmlformats-officedocument.spreadsheetml.sharedStrings+xml": "strs",
|
|
|
|
|
"application/vnd.openxmlformats-officedocument.spreadsheetml.styles+xml":"styles",
|
|
|
|
|
"application/vnd.openxmlformats-officedocument.theme+xml":"themes",
|
2014-01-16 03:42:25 +00:00
|
|
|
|
"application/vnd.openxmlformats-officedocument.spreadsheetml.comments+xml": "comments",
|
2013-03-31 22:56:45 +00:00
|
|
|
|
"foo": "bar"
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/* 18.2.28 (CT_WorkbookProtection) Defaults */
|
|
|
|
|
var WBPropsDef = {
|
|
|
|
|
allowRefreshQuery: '0',
|
|
|
|
|
autoCompressPictures: '1',
|
|
|
|
|
backupFile: '0',
|
|
|
|
|
checkCompatibility: '0',
|
|
|
|
|
codeName: '',
|
|
|
|
|
date1904: '0',
|
|
|
|
|
dateCompatibility: '1',
|
|
|
|
|
//defaultThemeVersion: '0',
|
|
|
|
|
filterPrivacy: '0',
|
|
|
|
|
hidePivotFieldList: '0',
|
|
|
|
|
promptedSolutions: '0',
|
|
|
|
|
publishItems: '0',
|
|
|
|
|
refreshAllConnections: false,
|
|
|
|
|
saveExternalLinkValues: '1',
|
|
|
|
|
showBorderUnselectedTables: '1',
|
|
|
|
|
showInkAnnotation: '1',
|
|
|
|
|
showObjects: 'all',
|
|
|
|
|
showPivotChartFilter: '0'
|
|
|
|
|
//updateLinks: 'userSet'
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/* 18.2.30 (CT_BookView) Defaults */
|
|
|
|
|
var WBViewDef = {
|
|
|
|
|
activeTab: '0',
|
|
|
|
|
autoFilterDateGrouping: '1',
|
|
|
|
|
firstSheet: '0',
|
|
|
|
|
minimized: '0',
|
|
|
|
|
showHorizontalScroll: '1',
|
|
|
|
|
showSheetTabs: '1',
|
|
|
|
|
showVerticalScroll: '1',
|
|
|
|
|
tabRatio: '600',
|
|
|
|
|
visibility: 'visible'
|
|
|
|
|
//window{Height,Width}, {x,y}Window
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/* 18.2.19 (CT_Sheet) Defaults */
|
|
|
|
|
var SheetDef = {
|
|
|
|
|
state: 'visible'
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/* 18.2.2 (CT_CalcPr) Defaults */
|
|
|
|
|
var CalcPrDef = {
|
|
|
|
|
calcCompleted: 'true',
|
|
|
|
|
calcMode: 'auto',
|
|
|
|
|
calcOnSave: 'true',
|
|
|
|
|
concurrentCalc: 'true',
|
|
|
|
|
fullCalcOnLoad: 'false',
|
|
|
|
|
fullPrecision: 'true',
|
|
|
|
|
iterate: 'false',
|
|
|
|
|
iterateCount: '100',
|
|
|
|
|
iterateDelta: '0.001',
|
|
|
|
|
refMode: 'A1'
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
/* 18.2.3 (CT_CustomWorkbookView) Defaults */
|
|
|
|
|
var CustomWBViewDef = {
|
|
|
|
|
autoUpdate: 'false',
|
|
|
|
|
changesSavedWin: 'false',
|
|
|
|
|
includeHiddenRowCol: 'true',
|
|
|
|
|
includePrintSettings: 'true',
|
|
|
|
|
maximized: 'false',
|
|
|
|
|
minimized: 'false',
|
|
|
|
|
onlySync: 'false',
|
|
|
|
|
personalView: 'false',
|
|
|
|
|
showComments: 'commIndicator',
|
|
|
|
|
showFormulaBar: 'true',
|
|
|
|
|
showHorizontalScroll: 'true',
|
|
|
|
|
showObjects: 'all',
|
|
|
|
|
showSheetTabs: 'true',
|
|
|
|
|
showStatusbar: 'true',
|
|
|
|
|
showVerticalScroll: 'true',
|
|
|
|
|
tabRatio: '600',
|
|
|
|
|
xWindow: '0',
|
|
|
|
|
yWindow: '0'
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
var XMLNS_CT = 'http://schemas.openxmlformats.org/package/2006/content-types';
|
|
|
|
|
var XMLNS_WB = 'http://schemas.openxmlformats.org/spreadsheetml/2006/main';
|
|
|
|
|
|
|
|
|
|
var strs = {}; // shared strings
|
|
|
|
|
var styles = {}; // shared styles
|
2013-04-03 23:03:39 +00:00
|
|
|
|
var _ssfopts = {}; // spreadsheet formatting options
|
2013-03-31 22:56:45 +00:00
|
|
|
|
|
|
|
|
|
/* 18.3 Worksheets */
|
|
|
|
|
function parseSheet(data) {
|
2013-11-12 16:29:53 +00:00
|
|
|
|
if(!data) return data;
|
2013-03-31 22:56:45 +00:00
|
|
|
|
/* 18.3.1.99 worksheet CT_Worksheet */
|
|
|
|
|
var s = {};
|
|
|
|
|
|
|
|
|
|
/* 18.3.1.35 dimension CT_SheetDimension ? */
|
|
|
|
|
var ref = data.match(/<dimension ref="([^"]*)"\s*\/>/);
|
2013-10-28 20:28:17 +00:00
|
|
|
|
if(ref && ref.length == 2 && ref[1].indexOf(":") !== -1) s["!ref"] = ref[1];
|
2013-03-31 22:56:45 +00:00
|
|
|
|
|
|
|
|
|
var refguess = {s: {r:1000000, c:1000000}, e: {r:0, c:0} };
|
|
|
|
|
var q = ["v","f"];
|
2013-05-01 04:55:54 +00:00
|
|
|
|
var sidx = 0;
|
2013-03-31 22:56:45 +00:00
|
|
|
|
/* 18.3.1.80 sheetData CT_SheetData ? */
|
|
|
|
|
if(!data.match(/<sheetData *\/>/))
|
2013-04-12 05:41:26 +00:00
|
|
|
|
data.match(/<sheetData>([^\u2603]*)<\/sheetData>/m)[1].split("</row>").forEach(function(x) {
|
2013-03-31 22:56:45 +00:00
|
|
|
|
if(x === "" || x.trim() === "") return;
|
|
|
|
|
|
|
|
|
|
/* 18.3.1.73 row CT_Row */
|
|
|
|
|
var row = parsexmltag(x.match(/<row[^>]*>/)[0]);
|
|
|
|
|
if(refguess.s.r > row.r - 1) refguess.s.r = row.r - 1;
|
|
|
|
|
if(refguess.e.r < row.r - 1) refguess.e.r = row.r - 1;
|
|
|
|
|
|
|
|
|
|
/* 18.3.1.4 c CT_Cell */
|
|
|
|
|
var cells = x.substr(x.indexOf('>')+1).split(/<c/);
|
|
|
|
|
cells.forEach(function(c, idx) { if(c === "" || c.trim() === "") return;
|
2013-10-28 20:28:17 +00:00
|
|
|
|
var cref = c.match(/r="([^"]*)"/);
|
2013-03-31 22:56:45 +00:00
|
|
|
|
c = "<c" + c;
|
2013-10-28 20:28:17 +00:00
|
|
|
|
if(cref && cref.length == 2) {
|
|
|
|
|
var cref_cell = decode_cell(cref[1]);
|
|
|
|
|
idx = cref_cell.c;
|
|
|
|
|
}
|
|
|
|
|
if(refguess.s.c > idx) refguess.s.c = idx;
|
|
|
|
|
if(refguess.e.c < idx) refguess.e.c = idx;
|
2013-03-31 22:56:45 +00:00
|
|
|
|
var cell = parsexmltag((c.match(/<c[^>]*>/)||[c])[0]); delete cell[0];
|
|
|
|
|
var d = c.substr(c.indexOf('>')+1);
|
|
|
|
|
var p = {};
|
|
|
|
|
q.forEach(function(f){var x=d.match(matchtag(f));if(x)p[f]=unescapexml(x[1]);});
|
|
|
|
|
|
|
|
|
|
/* SCHEMA IS ACTUALLY INCORRECT HERE. IF A CELL HAS NO T, EMIT "" */
|
|
|
|
|
if(cell.t === undefined && p.v === undefined) { p.t = "str"; p.v = undefined; }
|
|
|
|
|
else p.t = (cell.t ? cell.t : "n"); // default is "n" in schema
|
|
|
|
|
switch(p.t) {
|
|
|
|
|
case 'n': p.v = parseFloat(p.v); break;
|
2013-05-01 04:55:54 +00:00
|
|
|
|
case 's': {
|
|
|
|
|
sidx = parseInt(p.v, 10);
|
|
|
|
|
p.v = strs[sidx].t;
|
|
|
|
|
p.r = strs[sidx].r;
|
|
|
|
|
} break;
|
2013-03-31 22:56:45 +00:00
|
|
|
|
case 'str': if(p.v) p.v = utf8read(p.v); break; // normal string
|
|
|
|
|
case 'inlineStr':
|
2013-11-12 16:29:53 +00:00
|
|
|
|
p.t = 'str'; p.v = unescapexml((d.match(matchtag('t'))||["",""])[1]);
|
2013-03-31 22:56:45 +00:00
|
|
|
|
break; // inline string
|
|
|
|
|
case 'b':
|
|
|
|
|
switch(p.v) {
|
|
|
|
|
case '0': case 'FALSE': case "false": case false: p.v=false; break;
|
|
|
|
|
case '1': case 'TRUE': case "true": case true: p.v=true; break;
|
|
|
|
|
default: throw "Unrecognized boolean: " + p.v;
|
|
|
|
|
} break;
|
|
|
|
|
/* in case of error, stick value in .raw */
|
|
|
|
|
case 'e': p.raw = p.v; p.v = undefined; break;
|
|
|
|
|
default: throw "Unrecognized cell type: " + p.t;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* formatting */
|
2013-11-12 16:29:53 +00:00
|
|
|
|
if(cell.s && styles.CellXf) { /* TODO: second check is a hacked guard */
|
2013-03-31 22:56:45 +00:00
|
|
|
|
var cf = styles.CellXf[cell.s];
|
2013-04-17 01:32:20 +00:00
|
|
|
|
if(cf && cf.numFmtId && cf.numFmtId !== 0) {
|
2013-03-31 22:56:45 +00:00
|
|
|
|
p.raw = p.v;
|
|
|
|
|
p.rawt = p.t;
|
|
|
|
|
try {
|
2013-04-03 23:03:39 +00:00
|
|
|
|
p.v = SSF.format(cf.numFmtId,p.v,_ssfopts);
|
2013-03-31 22:56:45 +00:00
|
|
|
|
p.t = 'str';
|
|
|
|
|
} catch(e) { p.v = p.raw; }
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
s[cell.r] = p;
|
|
|
|
|
});
|
|
|
|
|
});
|
|
|
|
|
if(!s["!ref"]) s["!ref"] = encode_range(refguess);
|
|
|
|
|
return s;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
function parseProps(data) {
|
|
|
|
|
var p = { Company:'' }, q = {};
|
|
|
|
|
var strings = ["Application", "DocSecurity", "Company", "AppVersion"];
|
|
|
|
|
var bools = ["HyperlinksChanged","SharedDoc","LinksUpToDate","ScaleCrop"];
|
|
|
|
|
var xtra = ["HeadingPairs", "TitlesOfParts"];
|
|
|
|
|
var xtracp = ["category", "contentStatus", "lastModifiedBy", "lastPrinted", "revision", "version"];
|
|
|
|
|
var xtradc = ["creator", "description", "identifier", "language", "subject", "title"];
|
|
|
|
|
var xtradcterms = ["created", "modified"];
|
|
|
|
|
xtra = xtra.concat(xtracp.map(function(x) { return "cp:" + x; }));
|
|
|
|
|
xtra = xtra.concat(xtradc.map(function(x) { return "dc:" + x; }));
|
|
|
|
|
xtra = xtra.concat(xtradcterms.map(function(x) { return "dcterms:" + x; }));
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
strings.forEach(function(f){p[f] = (data.match(matchtag(f))||[])[1];});
|
|
|
|
|
bools.forEach(function(f){p[f] = (data.match(matchtag(f))||[])[1] == "true";});
|
|
|
|
|
xtra.forEach(function(f) {
|
|
|
|
|
var cur = data.match(new RegExp("<" + f + "[^>]*>(.*)<\/" + f + ">"));
|
|
|
|
|
if(cur && cur.length > 0) q[f] = cur[1];
|
|
|
|
|
});
|
|
|
|
|
|
2013-04-17 01:32:20 +00:00
|
|
|
|
if(q.HeadingPairs && q.TitlesOfParts) {
|
|
|
|
|
var v = parseVector(q.HeadingPairs);
|
2013-03-31 22:56:45 +00:00
|
|
|
|
var j = 0, widx = 0;
|
|
|
|
|
for(var i = 0; i !== v.length; ++i) {
|
|
|
|
|
switch(v[i].v) {
|
2013-04-17 01:32:20 +00:00
|
|
|
|
case "Worksheets": widx = j; p.Worksheets = +v[++i]; break;
|
2013-03-31 22:56:45 +00:00
|
|
|
|
case "Named Ranges": ++i; break; // TODO: Handle Named Ranges
|
|
|
|
|
}
|
|
|
|
|
}
|
2013-04-17 01:32:20 +00:00
|
|
|
|
var parts = parseVector(q.TitlesOfParts).map(utf8read);
|
|
|
|
|
p.SheetNames = parts.slice(widx, widx + p.Worksheets);
|
2013-03-31 22:56:45 +00:00
|
|
|
|
}
|
2013-04-17 01:32:20 +00:00
|
|
|
|
p.Creator = q["dc:creator"];
|
|
|
|
|
p.LastModifiedBy = q["cp:lastModifiedBy"];
|
|
|
|
|
p.CreatedDate = new Date(q["dcterms:created"]);
|
|
|
|
|
p.ModifiedDate = new Date(q["dcterms:modified"]);
|
2013-03-31 22:56:45 +00:00
|
|
|
|
return p;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* 18.6 Calculation Chain */
|
|
|
|
|
function parseDeps(data) {
|
|
|
|
|
var d = [];
|
|
|
|
|
var l = 0, i = 1;
|
2013-11-06 06:05:32 +00:00
|
|
|
|
(data.match(/<[^>]*>/g)||[]).forEach(function(x) {
|
2013-03-31 22:56:45 +00:00
|
|
|
|
var y = parsexmltag(x);
|
|
|
|
|
switch(y[0]) {
|
|
|
|
|
case '<?xml': break;
|
|
|
|
|
/* 18.6.2 calcChain CT_CalcChain 1 */
|
|
|
|
|
case '<calcChain': case '<calcChain>': case '</calcChain>': break;
|
|
|
|
|
/* 18.6.1 c CT_CalcCell 1 */
|
|
|
|
|
case '<c': delete y[0]; if(y.i) i = y.i; else y.i = i; d.push(y); break;
|
|
|
|
|
}
|
|
|
|
|
});
|
|
|
|
|
return d;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
var ctext = {};
|
|
|
|
|
|
|
|
|
|
function parseCT(data) {
|
2013-11-12 16:29:53 +00:00
|
|
|
|
if(!data || !data.match) return data;
|
2013-03-31 22:56:45 +00:00
|
|
|
|
var ct = { workbooks: [], sheets: [], calcchains: [], themes: [], styles: [],
|
2014-01-16 03:42:25 +00:00
|
|
|
|
coreprops: [], extprops: [], strs:[], comments: [], xmlns: "" };
|
2013-11-12 16:29:53 +00:00
|
|
|
|
(data.match(/<[^>]*>/g)||[]).forEach(function(x) {
|
2013-03-31 22:56:45 +00:00
|
|
|
|
var y = parsexmltag(x);
|
|
|
|
|
switch(y[0]) {
|
|
|
|
|
case '<?xml': break;
|
|
|
|
|
case '<Types': ct.xmlns = y.xmlns; break;
|
|
|
|
|
case '<Default': ctext[y.Extension] = y.ContentType; break;
|
|
|
|
|
case '<Override':
|
|
|
|
|
if(y.ContentType in ct2type)ct[ct2type[y.ContentType]].push(y.PartName);
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
});
|
2013-10-30 22:26:31 +00:00
|
|
|
|
if(ct.xmlns !== XMLNS_CT) throw new Error("Unknown Namespace: " + ct.xmlns);
|
2013-03-31 22:56:45 +00:00
|
|
|
|
ct.calcchain = ct.calcchains.length > 0 ? ct.calcchains[0] : "";
|
|
|
|
|
ct.sst = ct.strs.length > 0 ? ct.strs[0] : "";
|
|
|
|
|
ct.style = ct.styles.length > 0 ? ct.styles[0] : "";
|
|
|
|
|
delete ct.calcchains;
|
|
|
|
|
return ct;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
/* 18.2 Workbook */
|
|
|
|
|
function parseWB(data) {
|
|
|
|
|
var wb = { AppVersion:{}, WBProps:{}, WBView:[], Sheets:[], CalcPr:{}, xmlns: "" };
|
|
|
|
|
var pass = false;
|
|
|
|
|
data.match(/<[^>]*>/g).forEach(function(x) {
|
|
|
|
|
var y = parsexmltag(x);
|
|
|
|
|
switch(y[0]) {
|
|
|
|
|
case '<?xml': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.27 workbook CT_Workbook 1 */
|
|
|
|
|
case '<workbook': wb.xmlns = y.xmlns; break;
|
|
|
|
|
case '</workbook>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.13 fileVersion CT_FileVersion ? */
|
|
|
|
|
case '<fileVersion': delete y[0]; wb.AppVersion = y; break;
|
|
|
|
|
case '<fileVersion/>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.12 fileSharing CT_FileSharing ? */
|
|
|
|
|
case '<fileSharing': case '<fileSharing/>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.28 workbookPr CT_WorkbookPr ? */
|
|
|
|
|
case '<workbookPr': delete y[0]; wb.WBProps = y; break;
|
|
|
|
|
case '<workbookPr/>': delete y[0]; wb.WBProps = y; break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.29 workbookProtection CT_WorkbookProtection ? */
|
|
|
|
|
case '<workbookProtection/>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.1 bookViews CT_BookViews ? */
|
|
|
|
|
case '<bookViews>': case '</bookViews>': break;
|
|
|
|
|
/* 18.2.30 workbookView CT_BookView + */
|
|
|
|
|
case '<workbookView': delete y[0]; wb.WBView.push(y); break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.20 sheets CT_Sheets 1 */
|
|
|
|
|
case '<sheets>': case '</sheets>': break; // aggregate sheet
|
|
|
|
|
/* 18.2.19 sheet CT_Sheet + */
|
|
|
|
|
case '<sheet': delete y[0]; y.name = utf8read(y.name); wb.Sheets.push(y); break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.15 functionGroups CT_FunctionGroups ? */
|
|
|
|
|
case '<functionGroups': case '<functionGroups/>': break;
|
|
|
|
|
/* 18.2.14 functionGroup CT_FunctionGroup + */
|
|
|
|
|
case '<functionGroup': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.9 externalReferences CT_ExternalReferences ? */
|
|
|
|
|
case '<externalReferences': case '</externalReferences>': break;
|
|
|
|
|
/* 18.2.8 externalReference CT_ExternalReference + */
|
|
|
|
|
case '<externalReference': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.6 definedNames CT_DefinedNames ? */
|
|
|
|
|
case '<definedNames/>': break;
|
|
|
|
|
case '<definedNames>': pass=true; break;
|
|
|
|
|
case '</definedNames>': pass=false; break;
|
|
|
|
|
/* 18.2.5 definedName CT_DefinedName + */
|
|
|
|
|
case '<definedName': case '<definedName/>': case '</definedName>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.2 calcPr CT_CalcPr ? */
|
|
|
|
|
case '<calcPr': delete y[0]; wb.CalcPr = y; break;
|
|
|
|
|
case '<calcPr/>': delete y[0]; wb.CalcPr = y; break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.16 oleSize CT_OleSize ? (ref required) */
|
|
|
|
|
case '<oleSize': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.4 customWorkbookViews CT_CustomWorkbookViews ? */
|
|
|
|
|
case '<customWorkbookViews>': case '</customWorkbookViews>': case '<customWorkbookViews': break;
|
|
|
|
|
/* 18.2.3 customWorkbookView CT_CustomWorkbookView + */
|
|
|
|
|
case '<customWorkbookView': case '</customWorkbookView>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.18 pivotCaches CT_PivotCaches ? */
|
|
|
|
|
case '<pivotCaches>': case '</pivotCaches>': case '<pivotCaches': break;
|
|
|
|
|
/* 18.2.17 pivotCache CT_PivotCache ? */
|
|
|
|
|
case '<pivotCache': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.21 smartTagPr CT_SmartTagPr ? */
|
|
|
|
|
case '<smartTagPr': case '<smartTagPr/>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.23 smartTagTypes CT_SmartTagTypes ? */
|
|
|
|
|
case '<smartTagTypes': case '<smartTagTypes>': case '</smartTagTypes>': break;
|
|
|
|
|
/* 18.2.22 smartTagType CT_SmartTagType ? */
|
|
|
|
|
case '<smartTagType': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.24 webPublishing CT_WebPublishing ? */
|
|
|
|
|
case '<webPublishing': case '<webPublishing/>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.11 fileRecoveryPr CT_FileRecoveryPr ? */
|
|
|
|
|
case '<fileRecoveryPr': case '<fileRecoveryPr/>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.26 webPublishObjects CT_WebPublishObjects ? */
|
|
|
|
|
case '<webPublishObjects>': case '<webPublishObjects': case '</webPublishObjects>': break;
|
|
|
|
|
/* 18.2.25 webPublishObject CT_WebPublishObject ? */
|
|
|
|
|
case '<webPublishObject': break;
|
|
|
|
|
|
|
|
|
|
/* 18.2.10 extLst CT_ExtensionList ? */
|
|
|
|
|
case '<extLst>': case '</extLst>': case '<extLst/>': break;
|
|
|
|
|
/* 18.2.7 ext CT_Extension + */
|
|
|
|
|
case '<ext': pass=true; break; //TODO: check with versions of excel
|
|
|
|
|
case '</ext>': pass=false; break;
|
|
|
|
|
|
|
|
|
|
/* Others */
|
|
|
|
|
case '<mx:ArchID': break;
|
|
|
|
|
case '<mc:AlternateContent': pass=true; break;
|
|
|
|
|
case '</mc:AlternateContent>': pass=false; break;
|
|
|
|
|
}
|
|
|
|
|
});
|
2013-10-30 22:26:31 +00:00
|
|
|
|
if(wb.xmlns !== XMLNS_WB) throw new Error("Unknown Namespace: " + wb.xmlns);
|
2013-03-31 22:56:45 +00:00
|
|
|
|
|
|
|
|
|
var z;
|
|
|
|
|
/* defaults */
|
2013-04-17 01:32:20 +00:00
|
|
|
|
for(z in WBPropsDef) if(typeof wb.WBProps[z] === 'undefined') wb.WBProps[z] = WBPropsDef[z];
|
|
|
|
|
for(z in CalcPrDef) if(typeof wb.CalcPr[z] === 'undefined') wb.CalcPr[z] = CalcPrDef[z];
|
2013-03-31 22:56:45 +00:00
|
|
|
|
|
2013-04-17 01:32:20 +00:00
|
|
|
|
wb.WBView.forEach(function(w){for(var z in WBViewDef) if(typeof w[z] === 'undefined') w[z]=WBViewDef[z]; });
|
|
|
|
|
wb.Sheets.forEach(function(w){for(var z in SheetDef) if(typeof w[z] === 'undefined') w[z]=SheetDef[z]; });
|
2013-03-31 22:56:45 +00:00
|
|
|
|
|
2013-04-06 21:56:20 +00:00
|
|
|
|
_ssfopts.date1904 = parsexmlbool(wb.WBProps.date1904, 'date1904');
|
2013-04-03 23:03:39 +00:00
|
|
|
|
|
2013-03-31 22:56:45 +00:00
|
|
|
|
return wb;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* 18.8.31 numFmts CT_NumFmts */
|
|
|
|
|
function parseNumFmts(t) {
|
|
|
|
|
styles.NumberFmt = [];
|
2013-04-17 01:32:20 +00:00
|
|
|
|
for(var y in SSF._table) styles.NumberFmt[y] = SSF._table[y];
|
2013-03-31 22:56:45 +00:00
|
|
|
|
t[0].match(/<[^>]*>/g).forEach(function(x) {
|
|
|
|
|
var y = parsexmltag(x);
|
|
|
|
|
switch(y[0]) {
|
|
|
|
|
case '<numFmts': case '</numFmts>': case '<numFmts/>': break;
|
|
|
|
|
case '<numFmt': {
|
|
|
|
|
var f=unescapexml(y.formatCode), i=parseInt(y.numFmtId,10);
|
|
|
|
|
styles.NumberFmt[i] = f; SSF.load(f,i);
|
|
|
|
|
} break;
|
|
|
|
|
default: throw 'unrecognized ' + y[0] + ' in numFmts';
|
|
|
|
|
}
|
|
|
|
|
});
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* 18.8.10 cellXfs CT_CellXfs */
|
|
|
|
|
function parseCXfs(t) {
|
|
|
|
|
styles.CellXf = [];
|
|
|
|
|
t[0].match(/<[^>]*>/g).forEach(function(x) {
|
|
|
|
|
var y = parsexmltag(x);
|
|
|
|
|
switch(y[0]) {
|
|
|
|
|
case '<cellXfs': case '<cellXfs/>': case '</cellXfs>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.8.45 xf CT_Xf */
|
|
|
|
|
case '<xf': if(y.numFmtId) y.numFmtId = parseInt(y.numFmtId, 10);
|
|
|
|
|
styles.CellXf.push(y); break;
|
|
|
|
|
case '</xf>': break;
|
|
|
|
|
|
|
|
|
|
/* 18.8.1 alignment CT_CellAlignment */
|
|
|
|
|
case '<alignment': break;
|
|
|
|
|
|
|
|
|
|
/* 18.8.33 protection CT_CellProtection */
|
2013-10-30 22:26:31 +00:00
|
|
|
|
case '<protection': case '</protection>': case '<protection/>': break;
|
2013-03-31 22:56:45 +00:00
|
|
|
|
|
|
|
|
|
case '<extLst': case '</extLst>': break;
|
|
|
|
|
case '<ext': break;
|
|
|
|
|
default: throw 'unrecognized ' + y[0] + ' in cellXfs';
|
|
|
|
|
}
|
|
|
|
|
});
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* 18.8 Styles CT_Stylesheet*/
|
|
|
|
|
function parseStyles(data) {
|
|
|
|
|
/* 18.8.39 styleSheet CT_Stylesheet */
|
|
|
|
|
var t;
|
|
|
|
|
|
|
|
|
|
/* numFmts CT_NumFmts ? */
|
2013-04-17 01:32:20 +00:00
|
|
|
|
if((t=data.match(/<numFmts([^>]*)>.*<\/numFmts>/))) parseNumFmts(t);
|
2013-03-31 22:56:45 +00:00
|
|
|
|
|
|
|
|
|
/* fonts CT_Fonts ? */
|
|
|
|
|
/* fills CT_Fills ? */
|
|
|
|
|
/* borders CT_Borders ? */
|
|
|
|
|
/* cellStyleXfs CT_CellStyleXfs ? */
|
|
|
|
|
|
|
|
|
|
/* cellXfs CT_CellXfs ? */
|
2013-04-17 01:32:20 +00:00
|
|
|
|
if((t=data.match(/<cellXfs([^>]*)>.*<\/cellXfs>/))) parseCXfs(t);
|
2013-03-31 22:56:45 +00:00
|
|
|
|
|
|
|
|
|
/* dxfs CT_Dxfs ? */
|
|
|
|
|
/* tableStyles CT_TableStyles ? */
|
|
|
|
|
/* colors CT_Colors ? */
|
|
|
|
|
/* extLst CT_ExtensionList ? */
|
|
|
|
|
|
|
|
|
|
return styles;
|
|
|
|
|
}
|
|
|
|
|
|
2014-01-16 03:42:25 +00:00
|
|
|
|
/* 9.3.2 OPC Relationships Markup */
|
|
|
|
|
function parseRels(data, currentFilePath) {
|
|
|
|
|
if (!data) return data;
|
|
|
|
|
if (currentFilePath.charAt(0) !== '/') {
|
|
|
|
|
currentFilePath = '/'+currentFilePath;
|
|
|
|
|
}
|
|
|
|
|
var rels = {};
|
|
|
|
|
|
|
|
|
|
var resolveRelativePathIntoAbsolute = function (to) {
|
|
|
|
|
var toksFrom = currentFilePath.split('/');
|
|
|
|
|
toksFrom.pop(); // folder path
|
|
|
|
|
var toksTo = to.split('/');
|
|
|
|
|
var reversed = [];
|
|
|
|
|
while (toksTo.length !== 0) {
|
|
|
|
|
var tokTo = toksTo.shift();
|
|
|
|
|
if (tokTo === '..') {
|
|
|
|
|
toksFrom.pop();
|
|
|
|
|
} else if (tokTo !== '.') {
|
|
|
|
|
toksFrom.push(tokTo);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
return toksFrom.join('/');
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
data.match(/<[^>]*>/g).forEach(function(x) {
|
|
|
|
|
var y = parsexmltag(x);
|
|
|
|
|
/* 9.3.2.2 OPC_Relationships */
|
|
|
|
|
if (y[0] === '<Relationship') {
|
|
|
|
|
var rel = {}; rel.Type = y.Type; rel.Target = y.Target; rel.Id = y.Id; rel.TargetMode = y.TargetMode;
|
|
|
|
|
var canonictarget = resolveRelativePathIntoAbsolute(y.Target);
|
|
|
|
|
rels[canonictarget] = rel;
|
|
|
|
|
}
|
|
|
|
|
});
|
|
|
|
|
|
|
|
|
|
return rels;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/* 18.7.3 CT_Comment */
|
|
|
|
|
function parseComments(data) {
|
|
|
|
|
if(data.match(/<comments *\/>/)) {
|
|
|
|
|
throw new Error('Not a valid comments xml');
|
|
|
|
|
}
|
|
|
|
|
var authors = [];
|
|
|
|
|
var commentList = [];
|
|
|
|
|
data.match(/<authors>([^\u2603]*)<\/authors>/m)[1].split('</author>').forEach(function(x) {
|
|
|
|
|
if(x === "" || x.trim() === "") return;
|
|
|
|
|
authors.push(x.match(/<author[^>]*>(.*)/)[1]);
|
|
|
|
|
});
|
|
|
|
|
data.match(/<commentList>([^\u2603]*)<\/commentList>/m)[1].split('</comment>').forEach(function(x, index) {
|
|
|
|
|
if(x === "" || x.trim() === "") return;
|
|
|
|
|
var y = parsexmltag(x.match(/<comment[^>]*>/)[0]);
|
2014-01-18 13:45:49 +00:00
|
|
|
|
var comment = { author: y.authorId && authors[y.authorId] ? authors[y.authorId] : undefined, ref: y.ref, guid: y.guid };
|
2014-01-18 08:29:51 +00:00
|
|
|
|
var textMatch = x.match(/<text>([^\u2603]*)<\/text>/m);
|
|
|
|
|
if (!textMatch || !textMatch[1]) return; // a comment may contain an empty text tag.
|
2014-01-22 04:58:26 +00:00
|
|
|
|
var rt = parse_si(textMatch[1]);
|
2014-01-18 13:45:49 +00:00
|
|
|
|
comment.raw = rt.raw;
|
|
|
|
|
comment.t = rt.t;
|
|
|
|
|
comment.r = rt.r;
|
2014-01-16 03:42:25 +00:00
|
|
|
|
commentList.push(comment);
|
|
|
|
|
});
|
|
|
|
|
return commentList;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
function parseCommentsAddToSheets(zip, dirComments, sheets, sheetRels) {
|
|
|
|
|
for(var i = 0; i != dirComments.length; ++i) {
|
|
|
|
|
var canonicalpath=dirComments[i];
|
|
|
|
|
var comments=parseComments(getdata(getzipfile(zip, canonicalpath.replace(/^\//,''))));
|
|
|
|
|
// find the sheets targeted by these comments
|
|
|
|
|
var sheetNames = Object.keys(sheets);
|
|
|
|
|
for(var j = 0; j != sheetNames.length; ++j) {
|
|
|
|
|
var sheetName = sheetNames[j];
|
|
|
|
|
var rels = sheetRels[sheetName];
|
|
|
|
|
if (rels) {
|
|
|
|
|
var rel = rels[canonicalpath];
|
|
|
|
|
if (rel) {
|
|
|
|
|
insertCommentsIntoSheet(sheetName, sheets[sheetName], comments);
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
function insertCommentsIntoSheet(sheetName, sheet, comments) {
|
|
|
|
|
comments.forEach(function(comment) {
|
|
|
|
|
var cell = sheet[comment.ref];
|
|
|
|
|
if (!cell) {
|
|
|
|
|
cell = {};
|
|
|
|
|
sheet[comment.ref] = cell;
|
2014-01-20 23:05:00 +00:00
|
|
|
|
var range = decode_range(sheet["!ref"]);
|
|
|
|
|
var thisCell = decode_cell(comment.ref);
|
|
|
|
|
if(range.s.r > thisCell.r) range.s.r = thisCell.r;
|
|
|
|
|
if(range.e.r < thisCell.r) range.e.r = thisCell.r;
|
|
|
|
|
if(range.s.c > thisCell.c) range.s.c = thisCell.c;
|
|
|
|
|
if(range.e.c < thisCell.c) range.e.c = thisCell.c;
|
|
|
|
|
var encoded = encode_range(range);
|
|
|
|
|
if (encoded !== sheet["!ref"]) sheet["!ref"] = encoded;
|
2014-01-22 04:58:26 +00:00
|
|
|
|
}
|
2014-01-16 03:42:25 +00:00
|
|
|
|
|
|
|
|
|
if (!cell.c) {
|
|
|
|
|
cell.c = [];
|
|
|
|
|
}
|
2014-01-18 13:45:49 +00:00
|
|
|
|
cell.c.push({a: comment.author, t: comment.t, raw: comment.raw, r: comment.r});
|
2014-01-16 03:42:25 +00:00
|
|
|
|
});
|
|
|
|
|
}
|
|
|
|
|
|
2013-11-06 06:05:32 +00:00
|
|
|
|
function getdata(data) {
|
2014-01-22 04:58:26 +00:00
|
|
|
|
if(!data) return null;
|
2013-11-06 06:05:32 +00:00
|
|
|
|
if(data.data) return data.data;
|
|
|
|
|
if(data._data && data._data.getContent) return Array.prototype.slice.call(data._data.getContent(),0).map(function(x) { return String.fromCharCode(x); }).join("");
|
2013-11-12 16:29:53 +00:00
|
|
|
|
return null;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
function getzipfile(zip, file) {
|
|
|
|
|
var f = file; if(zip.files[f]) return zip.files[f];
|
|
|
|
|
f = file.toLowerCase(); if(zip.files[f]) return zip.files[f];
|
|
|
|
|
f = f.replace(/\//g,'\\'); if(zip.files[f]) return zip.files[f];
|
2013-11-12 17:08:20 +00:00
|
|
|
|
throw new Error("Cannot find file " + file + " in zip");
|
2013-11-06 06:05:32 +00:00
|
|
|
|
}
|
|
|
|
|
|
2013-03-31 22:56:45 +00:00
|
|
|
|
function parseZip(zip) {
|
|
|
|
|
var entries = Object.keys(zip.files);
|
|
|
|
|
var keys = entries.filter(function(x){return x.substr(-1) != '/';}).sort();
|
2013-11-12 16:29:53 +00:00
|
|
|
|
var dir = parseCT(getdata(getzipfile(zip, '[Content_Types].xml')));
|
2013-12-27 03:15:16 +00:00
|
|
|
|
if(dir.workbooks.length === 0) throw new Error("Could not find workbook entry");
|
2013-03-31 22:56:45 +00:00
|
|
|
|
strs = {};
|
2013-11-12 16:29:53 +00:00
|
|
|
|
if(dir.sst) strs=parse_sst(getdata(getzipfile(zip, dir.sst.replace(/^\//,''))));
|
2013-03-31 22:56:45 +00:00
|
|
|
|
|
|
|
|
|
styles = {};
|
2013-11-12 16:29:53 +00:00
|
|
|
|
if(dir.style) styles = parseStyles(getdata(getzipfile(zip, dir.style.replace(/^\//,''))));
|
2013-03-31 22:56:45 +00:00
|
|
|
|
|
2013-11-12 16:29:53 +00:00
|
|
|
|
var wb = parseWB(getdata(getzipfile(zip, dir.workbooks[0].replace(/^\//,''))));
|
|
|
|
|
var propdata = dir.coreprops.length !== 0 ? getdata(getzipfile(zip, dir.coreprops[0].replace(/^\//,''))) : "";
|
|
|
|
|
propdata += dir.extprops.length !== 0 ? getdata(getzipfile(zip, dir.extprops[0].replace(/^\//,''))) : "";
|
2013-03-31 22:56:45 +00:00
|
|
|
|
var props = propdata !== "" ? parseProps(propdata) : {};
|
|
|
|
|
var deps = {};
|
2013-11-12 16:29:53 +00:00
|
|
|
|
if(dir.calcchain) deps=parseDeps(getdata(getzipfile(zip, dir.calcchain.replace(/^\//,''))));
|
2013-03-31 22:56:45 +00:00
|
|
|
|
var sheets = {}, i=0;
|
2014-01-16 03:42:25 +00:00
|
|
|
|
var sheetRels = {};
|
2013-03-31 22:56:45 +00:00
|
|
|
|
if(!props.Worksheets) {
|
2014-01-22 04:58:26 +00:00
|
|
|
|
/* Google Docs doesn't generate the appropriate metadata, so we impute: */
|
|
|
|
|
var wbsheets = wb.Sheets;
|
|
|
|
|
props.Worksheets = wbsheets.length;
|
|
|
|
|
props.SheetNames = [];
|
|
|
|
|
for(var j = 0; j != wbsheets.length; ++j) {
|
|
|
|
|
props.SheetNames[j] = wbsheets[j].name;
|
|
|
|
|
}
|
|
|
|
|
for(i = 0; i != props.Worksheets; ++i) {
|
|
|
|
|
try { /* TODO: remove these guards */
|
|
|
|
|
var path = 'xl/worksheets/sheet' + (i+1) + '.xml';
|
|
|
|
|
var relsPath = path.replace(/^(.*)(\/)([^\/]*)$/, "$1/_rels/$3.rels");
|
|
|
|
|
sheets[props.SheetNames[i]]=parseSheet(getdata(getzipfile(zip, path)));
|
|
|
|
|
sheetRels[props.SheetNames[i]]=parseRels(getdata(getzipfile(zip, relsPath)), path);
|
|
|
|
|
} catch(e) {}
|
|
|
|
|
}
|
|
|
|
|
} else {
|
|
|
|
|
for(i = 0; i != props.Worksheets; ++i) {
|
|
|
|
|
try {
|
|
|
|
|
var path = dir.sheets[i].replace(/^\//,'');
|
2014-01-16 03:42:25 +00:00
|
|
|
|
var relsPath = path.replace(/^(.*)(\/)([^\/]*)$/, "$1/_rels/$3.rels");
|
2014-01-22 04:58:26 +00:00
|
|
|
|
sheets[props.SheetNames[i]]=parseSheet(getdata(getzipfile(zip, path)));
|
|
|
|
|
sheetRels[props.SheetNames[i]]=parseRels(getdata(getzipfile(zip, relsPath)), path);
|
|
|
|
|
} catch(e) {}
|
|
|
|
|
}
|
2013-03-31 22:56:45 +00:00
|
|
|
|
}
|
2014-01-22 04:58:26 +00:00
|
|
|
|
|
|
|
|
|
if(dir.comments) parseCommentsAddToSheets(zip, dir.comments, sheets, sheetRels);
|
|
|
|
|
|
2013-03-31 22:56:45 +00:00
|
|
|
|
return {
|
|
|
|
|
Directory: dir,
|
|
|
|
|
Workbook: wb,
|
|
|
|
|
Props: props,
|
|
|
|
|
Deps: deps,
|
|
|
|
|
Sheets: sheets,
|
|
|
|
|
SheetNames: props.SheetNames,
|
|
|
|
|
Strings: strs,
|
|
|
|
|
Styles: styles,
|
|
|
|
|
keys: keys,
|
|
|
|
|
files: zip.files
|
|
|
|
|
};
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
var _fs, jszip;
|
2013-04-17 01:32:20 +00:00
|
|
|
|
if(typeof JSZip !== 'undefined') jszip = JSZip;
|
2013-03-31 22:56:45 +00:00
|
|
|
|
if (typeof exports !== 'undefined') {
|
|
|
|
|
if (typeof module !== 'undefined' && module.exports) {
|
|
|
|
|
if(typeof jszip === 'undefined') jszip = require('./jszip').JSZip;
|
|
|
|
|
_fs = require('fs');
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
function readSync(data, options) {
|
|
|
|
|
var zip, d = data;
|
|
|
|
|
var o = options||{};
|
|
|
|
|
switch((o.type||"base64")){
|
|
|
|
|
case "file": d = _fs.readFileSync(data).toString('base64');
|
|
|
|
|
/* falls through */
|
|
|
|
|
case "base64": zip = new jszip(d, { base64:true }); break;
|
|
|
|
|
case "binary": zip = new jszip(d, { base64:false }); break;
|
|
|
|
|
}
|
|
|
|
|
return parseZip(zip);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
function readFileSync(data, options) {
|
|
|
|
|
var o = options||{}; o.type = 'file';
|
|
|
|
|
return readSync(data, o);
|
|
|
|
|
}
|
|
|
|
|
|
2013-10-31 02:53:12 +00:00
|
|
|
|
XLSX.read = readSync;
|
|
|
|
|
XLSX.readFile = readFileSync;
|
|
|
|
|
XLSX.parseZip = parseZip;
|