2013-12-06 16:49:25 +00:00
|
|
|
# SSF
|
|
|
|
|
|
|
|
SpreadSheet Format (SSF) is a pure-JS library to format data using ECMA-376
|
|
|
|
spreadsheet format codes.
|
|
|
|
|
|
|
|
## Options
|
|
|
|
|
|
|
|
The various API functions take an `opts` argument which control parsing. The
|
|
|
|
default options are described below:
|
|
|
|
|
|
|
|
```js>tmp/opts.js
|
|
|
|
/* Options */
|
|
|
|
var opts_fmt = {};
|
|
|
|
function fixopts(o){for(var y in opts_fmt) if(o[y]===undefined) o[y]=opts_fmt[y];}
|
|
|
|
SSF.opts = opts_fmt;
|
|
|
|
```
|
|
|
|
|
|
|
|
There are two commonly-recognized date code formats:
|
|
|
|
- 1900 mode (where date=0 is 1899-12-31)
|
|
|
|
- 1904 mode (where date=0 is 1904-01-01)
|
|
|
|
|
|
|
|
The difference between the the 1900 and 1904 date modes is 1462 days. Since
|
|
|
|
the 1904 date mode was only default in a few Mac variants of Excel (2011 uses
|
|
|
|
1900 mode), the default is 1900 mode. Consistent with ECMA-376 the name is
|
|
|
|
`date1904`:
|
|
|
|
|
|
|
|
```
|
|
|
|
opts_fmt.date1904 = 0;
|
|
|
|
```
|
|
|
|
|
|
|
|
The default output is a text representation (no effort to capture colors). To
|
|
|
|
control the output, set the `output` variable:
|
|
|
|
|
|
|
|
- `text`: no color (default)
|
|
|
|
- `html`: html output using
|
|
|
|
- `ansi`: ansi color codes (requires `colors` module)
|
|
|
|
|
|
|
|
```
|
|
|
|
opts_fmt.output = "";
|
|
|
|
```
|
|
|
|
|
|
|
|
There are a few places where the specification is ambiguous or where Excel does
|
|
|
|
not follow the spec. They are noted in the document.
|
|
|
|
|
|
|
|
The `mode` option controls compatibility:
|
|
|
|
|
|
|
|
- `ssf`: options that the author believes makes the most sense (default)
|
|
|
|
- `ecma`: compatibility with ECMA-376
|
|
|
|
- `excel`: compatibility with MS-XLSX
|
|
|
|
|
|
|
|
```
|
|
|
|
opts_fmt.mode = "";
|
|
|
|
```
|
|
|
|
|
|
|
|
## Conditional Format Codes
|
|
|
|
|
|
|
|
The specification is a bit unclear here. It initially claims in §18.3.1:
|
|
|
|
|
|
|
|
> Up to four sections of format codes can be specified. The format codes,
|
|
|
|
separated by semicolons, define the formats for positive numbers, negative
|
|
|
|
numbers, zero values, and text, in that order.
|
|
|
|
|
|
|
|
Semicolons can be escaped with the `\` character, so we need to split on those
|
|
|
|
semicolons that aren't prefaced by a slash or within a quoted string:
|
|
|
|
|
|
|
|
```js>tmp/main.js
|
|
|
|
function split_fmt(fmt) {
|
|
|
|
var out = [];
|
|
|
|
var in_str = -1;
|
|
|
|
for(var i = 0, j = 0; i < fmt.length; ++i) {
|
|
|
|
if(in_str != -1) { if(fmt[i] == '"') in_str = -1; continue; }
|
|
|
|
if(fmt[i] == "_" || fmt[i] == "*" || fmt[i] == "\\") { ++i; continue; }
|
|
|
|
if(fmt[i] == '"') { in_str = i; continue; }
|
|
|
|
if(fmt[i] != ";") continue;
|
|
|
|
out.push(fmt.slice(j,i));
|
|
|
|
j = i+1;
|
|
|
|
}
|
|
|
|
out.push(fmt.slice(j));
|
|
|
|
if(in_str !=-1) throw "Format |" + fmt + "| unterminated string at " + in_str;
|
|
|
|
return out;
|
|
|
|
}
|
|
|
|
SSF._split = split_fmt;
|
|
|
|
```
|
|
|
|
|
|
|
|
But it also allows for conditional formatting:
|
|
|
|
|
|
|
|
> To set number formats that are applied only if a number meets a specified
|
|
|
|
> condition, enclose the condition in square brackets. The condition consists
|
|
|
|
> of a comparison operator and a value. Comparison operators include:
|
|
|
|
> `=` Equal to;
|
|
|
|
> `>` Greater than;
|
|
|
|
> `<` Less than;
|
|
|
|
> `>=` Greater than or equal to,
|
|
|
|
> `<=` Less than or equal to,
|
|
|
|
> and `<>` Not equal to.
|
|
|
|
|
|
|
|
One problem is that Excel doesn't support three conditionals. For example:
|
|
|
|
|
|
|
|
```>
|
|
|
|
[Red][<-25]General;[Blue][>25]General;[Green][<>0]General;[Yellow]General
|
|
|
|
```
|
|
|
|
|
|
|
|
One would expect that the format code would color all numbers that are `< -25`
|
|
|
|
in red, all numbers `> 25` in blue, nonzero numbers between `-25` and `25` in
|
|
|
|
green, and color `0` and text in yellow. Excel doesn't do that.
|
|
|
|
|
|
|
|
The two-conditional case works in an "expected" way if you interpret the third
|
|
|
|
clause as the case for numbers that don't fit the first two:
|
|
|
|
|
|
|
|
```>
|
|
|
|
[Red][<-25]General;[Blue][>25]General;[Green]General;[Yellow]General
|
|
|
|
```
|
|
|
|
|
|
|
|
will render values below `-25` as Red, above `25` as Blue, Green for other
|
|
|
|
numbers, and Yellow for text.
|
|
|
|
|
|
|
|
Only the text case is allowed to have the `@` text sigil. Excel interprets it
|
|
|
|
as the last format.
|
|
|
|
|
|
|
|
|
|
|
|
## General Number Format
|
|
|
|
|
|
|
|
The 'general' format for spreadsheets (identified by format code 0) is highly
|
|
|
|
context-sensitive and the implementation tries to follow the format to the best
|
|
|
|
of its abilities given the knowledge.
|
|
|
|
|
|
|
|
```js>tmp/general.js
|
|
|
|
var general_fmt = function(v) {
|
|
|
|
```
|
|
|
|
|
|
|
|
Booleans are serialized in upper case:
|
|
|
|
|
|
|
|
```
|
|
|
|
if(typeof v === 'boolean') return v ? "TRUE" : "FALSE";
|
|
|
|
```
|
|
|
|
|
2013-12-14 03:28:57 +00:00
|
|
|
For numbers, try to display up to 11 digits of the number:
|
2013-12-06 16:49:25 +00:00
|
|
|
|
2013-12-14 03:28:57 +00:00
|
|
|
```
|
|
|
|
if(typeof v === 'number') {
|
|
|
|
return v.toString().substr(0,11);
|
|
|
|
}
|
|
|
|
```
|
|
|
|
|
|
|
|
For strings, just return the text as-is:
|
|
|
|
|
|
|
|
```
|
|
|
|
if(typeof v === 'string') return v;
|
|
|
|
```
|
2013-12-06 16:49:25 +00:00
|
|
|
|
|
|
|
```
|
2013-12-14 03:28:57 +00:00
|
|
|
throw "unsupport value in General format: " + v;
|
2013-12-06 16:49:25 +00:00
|
|
|
};
|
|
|
|
SSF._general = general_fmt;
|
|
|
|
```
|
|
|
|
|
|
|
|
## Implied Number Formats
|
|
|
|
|
|
|
|
These are the commonly-used formats that have a special implied code.
|
|
|
|
None of the international formats are included here.
|
|
|
|
|
|
|
|
```js>tmp/consts.js
|
|
|
|
var table_fmt = {
|
|
|
|
1: '0',
|
|
|
|
2: '0.00',
|
|
|
|
3: '#,##0',
|
|
|
|
4: '#,##0.00',
|
|
|
|
9: '0%',
|
|
|
|
10: '0.00%',
|
|
|
|
11: '0.00E+00',
|
|
|
|
12: '# ?/?',
|
|
|
|
13: '# ??/??',
|
|
|
|
```
|
|
|
|
|
|
|
|
Now Excel and other formats treat code 14 as `mm/dd/yy` (with slashes). Given
|
|
|
|
that the spec gives no internationalization considerations, erring on the side
|
|
|
|
of the applications makes sense here:
|
|
|
|
|
|
|
|
```
|
|
|
|
14: 'mm/dd/yy',
|
|
|
|
15: 'd-mmm-yy',
|
|
|
|
16: 'd-mmm',
|
|
|
|
17: 'mmm-yy',
|
|
|
|
18: 'h:mm AM/PM',
|
|
|
|
19: 'h:mm:ss AM/PM',
|
|
|
|
20: 'h:mm',
|
|
|
|
21: 'h:mm:ss',
|
|
|
|
22: 'm/d/yy h:mm',
|
|
|
|
37: '#,##0 ;(#,##0)',
|
|
|
|
38: '#,##0 ;[Red](#,##0)',
|
|
|
|
39: '#,##0.00;(#,##0.00)',
|
|
|
|
40: '#,##0.00;[Red](#,##0.00)',
|
|
|
|
45: 'mm:ss',
|
|
|
|
46: '[h]:mm:ss',
|
|
|
|
47: 'mmss.0',
|
|
|
|
48: '##0.0E+0',
|
|
|
|
49: '@'
|
|
|
|
};
|
|
|
|
```
|
|
|
|
|
|
|
|
These test cases were manually generated in Excel 2011 [value, code, result]:
|
|
|
|
|
|
|
|
```json>test/implied.json
|
|
|
|
[
|
|
|
|
[12345.6789, 0, "12345.6789"],
|
|
|
|
[12345.6789, 1, "12346"],
|
|
|
|
[12345.6789, 2, "12345.68"],
|
|
|
|
[12345.6789, 3, "12,346"],
|
|
|
|
[12345.6789, 4, "12,345.68"],
|
|
|
|
[12345.6789, 9, "1234568%"],
|
|
|
|
[12345.6789, 10, "1234567.89%"],
|
|
|
|
[12345.6789, 11, "1.23E+04"],
|
|
|
|
[12345.6789, 12, "12345 2/3"],
|
|
|
|
[12345.6789, 13, "12345 55/81"],
|
|
|
|
[12345.6789, 14, "10/18/33"],
|
|
|
|
[12345.6789, 15, "18-Oct-33"],
|
|
|
|
[12345.6789, 16, "18-Oct"],
|
|
|
|
[12345.6789, 17, "Oct-33"],
|
|
|
|
[12345.6789, 18, "4:17 PM"],
|
|
|
|
[12345.6789, 19, "4:17:37 PM"],
|
|
|
|
[12345.6789, 20, "16:17"],
|
|
|
|
[12345.6789, 21, "16:17:37"],
|
|
|
|
[12345.6789, 22, "10/18/33 16:17"],
|
|
|
|
[12345.6789, 37, "12,346"],
|
|
|
|
[12345.6789, 38, "12,346"],
|
|
|
|
[12345.6789, 39, "12,345.68"],
|
|
|
|
[12345.6789, 40, "12,345.68"],
|
|
|
|
[12345.6789, 45, "17:37"],
|
|
|
|
[12345.6789, 46, "296296:17:37"],
|
|
|
|
[12345.6789, 47, "1737.0"],
|
|
|
|
[12345.6789, 48, "12.3E+3"],
|
|
|
|
[12345.6789, 49, "12345.6789"]
|
|
|
|
]
|
|
|
|
```
|
|
|
|
|
|
|
|
## Dates and Time
|
|
|
|
|
|
|
|
The code `ddd` displays short day-of-week and `dddd` shows long day-of-week:
|
|
|
|
|
|
|
|
```js>tmp/consts.js
|
|
|
|
var days = [
|
|
|
|
['Sun', 'Sunday'],
|
|
|
|
['Mon', 'Monday'],
|
|
|
|
['Tue', 'Tuesday'],
|
|
|
|
['Wed', 'Wednesday'],
|
|
|
|
['Thu', 'Thursday'],
|
|
|
|
['Fri', 'Friday'],
|
|
|
|
['Sat', 'Saturday']
|
|
|
|
];
|
|
|
|
|
|
|
|
```
|
|
|
|
|
|
|
|
`mmm` shows short month, `mmmm` shows long month, and `mmmmm` shows one char:
|
|
|
|
|
|
|
|
```
|
|
|
|
var months = [
|
|
|
|
['J', 'Jan', 'January'],
|
|
|
|
['F', 'Feb', 'February'],
|
|
|
|
['M', 'Mar', 'March'],
|
|
|
|
['A', 'Apr', 'April'],
|
|
|
|
['M', 'May', 'May'],
|
|
|
|
['J', 'Jun', 'June'],
|
|
|
|
['J', 'Jul', 'July'],
|
|
|
|
['A', 'Aug', 'August'],
|
|
|
|
['S', 'Sep', 'September'],
|
|
|
|
['O', 'Oct', 'October'],
|
|
|
|
['N', 'Nov', 'November'],
|
|
|
|
['D', 'Dec', 'December']
|
|
|
|
];
|
|
|
|
```
|
|
|
|
|
|
|
|
## Parsing Date and Time Codes
|
|
|
|
|
|
|
|
Most spreadsheet formats store dates and times as floating point numbers (where
|
|
|
|
the integer part is a day code based on a format and the fractional part is the
|
|
|
|
portion of a 24 hour day).
|
|
|
|
|
|
|
|
|
|
|
|
```js>tmp/date.js
|
|
|
|
var parse_date_code = function parse_date_code(v,opts) {
|
|
|
|
var date = Math.floor(v), time = Math.round(86400 * (v - date)), dow=0;
|
|
|
|
var dout=[], out={D:date, T:time}; fixopts(opts = (opts||{}));
|
|
|
|
```
|
|
|
|
|
|
|
|
Excel help actually recommends treating the 1904 date codes as 1900 date codes
|
|
|
|
shifted by 1462 days.
|
|
|
|
|
|
|
|
```
|
|
|
|
if(opts.date1904) date += 1462;
|
|
|
|
```
|
|
|
|
|
|
|
|
Due to a bug in Lotus 1-2-3 which was propagated by Excel and other variants,
|
|
|
|
the year 1900 is recognized as a leap year. JS has no way of representing that
|
|
|
|
abomination as a `Date`, so the easiest way is to store the data as a tuple.
|
|
|
|
|
|
|
|
February 29, 1900 (date `60`) is recognized as a Wednesday.
|
|
|
|
|
|
|
|
```
|
|
|
|
if(date === 60) {dout = [1900,2,29]; dow=3;}
|
|
|
|
```
|
|
|
|
|
|
|
|
For the other dates, using the JS date mechanism suffices.
|
|
|
|
|
|
|
|
```
|
|
|
|
else {
|
|
|
|
if(date > 60) --date;
|
|
|
|
/* 1 = Jan 1 1900 */
|
|
|
|
var d = new Date(1900,0,1);
|
|
|
|
d.setDate(d.getDate() + date - 1);
|
|
|
|
dout = [d.getFullYear(), d.getMonth()+1,d.getDate()];
|
|
|
|
dow = d.getDay();
|
|
|
|
```
|
|
|
|
|
|
|
|
Note that Excel opted to keep the day-of-week metric consistent with the extra
|
|
|
|
day. In practice, that means the days before the fake leap day are off. For
|
|
|
|
example, date code `55` is "Friday, February 24, 1900" when in fact it was a
|
|
|
|
Saturday. The "right" thing to do is to keep the DOW consistent and just break
|
|
|
|
the fact that there are two Wednesdays in that "week".
|
|
|
|
|
|
|
|
```
|
|
|
|
if(opts.mode === 'excel' && date < 60) dow = (dow + 6) % 7;
|
|
|
|
}
|
|
|
|
```
|
|
|
|
|
|
|
|
Because JS dates cannot represent the bad leap day, this returns an object:
|
|
|
|
|
|
|
|
```
|
|
|
|
out.y = dout[0]; out.m = dout[1]; out.d = dout[2];
|
|
|
|
out.S = time % 60; time = Math.floor(time / 60);
|
|
|
|
out.M = time % 60; time = Math.floor(time / 60);
|
|
|
|
out.H = time;
|
|
|
|
out.q = dow;
|
|
|
|
return out;
|
|
|
|
};
|
|
|
|
SSF.parse_date_code = parse_date_code;
|
|
|
|
```
|
|
|
|
|
2013-12-14 03:28:57 +00:00
|
|
|
## Evaluating Number Formats
|
|
|
|
|
|
|
|
```js>tmp/number.js
|
|
|
|
String.prototype.reverse = function() { return this.split("").reverse().join(""); }
|
|
|
|
var commaify = function(s) { return s.reverse().replace(/.../g,"$&,").reverse(); };
|
|
|
|
var write_num = function(type, fmt, val) {
|
|
|
|
```
|
|
|
|
|
|
|
|
Percentage values should be physically shifted:
|
|
|
|
|
|
|
|
```js>tmp/number.js
|
|
|
|
var mul = 0;
|
|
|
|
fmt = fmt.replace(/%/g,function(x) { mul++; return ""; });
|
|
|
|
if(mul !== 0) return write_num(type, fmt, val * Math.pow(10,2*mul)) + fill("%",mul);
|
|
|
|
```
|
|
|
|
|
|
|
|
For exponents, get the exponent and mantissa and format them separately:
|
|
|
|
|
|
|
|
```
|
|
|
|
if(fmt.indexOf("E") > -1) {
|
|
|
|
var o = val.toExponential(fmt.indexOf("E") - fmt.indexOf(".") - 1);
|
|
|
|
if(fmt.match(/E\+00$/) && o.match(/e[+-][0-9]$/)) o = o.substr(0,o.length-1) + "0" + o[o.length-1];
|
|
|
|
if(fmt.match(/E\-/) && o.match(/e\+/)) o = o.replace(/e\+/,"e");
|
|
|
|
return o.replace("e","E");
|
|
|
|
}
|
|
|
|
```
|
|
|
|
|
|
|
|
The default cases are hard-coded. TODO: actually parse them
|
|
|
|
|
|
|
|
```js>tmp/number.js
|
|
|
|
switch(fmt) {
|
|
|
|
case "0": return Math.round(val);
|
|
|
|
case "0.00": return Math.round(val*100)/100;
|
|
|
|
case "#,##0": return commaify(String(Math.round(val)));
|
|
|
|
case "#,##0.00": return commaify(String(Math.floor(val))) + "." + Math.round((val-Math.floor(val))*100);
|
|
|
|
default:
|
|
|
|
}
|
|
|
|
console.log(type, fmt, val);
|
|
|
|
return "0";
|
|
|
|
};
|
|
|
|
```
|
|
|
|
|
2013-12-06 16:49:25 +00:00
|
|
|
## Evaluating Format Strings
|
|
|
|
|
|
|
|
```js>tmp/main.js
|
|
|
|
function eval_fmt(fmt, v, opts) {
|
|
|
|
var out = [], o = "", i = 0, c = "", lst='t', q = {}, dt;
|
|
|
|
fixopts(opts = (opts || {}));
|
|
|
|
var hr='H';
|
|
|
|
/* Tokenize */
|
|
|
|
while(i < fmt.length) {
|
|
|
|
switch((c = fmt[i])) {
|
|
|
|
```
|
|
|
|
|
|
|
|
Text between double-quotes are treated literally, and individual characters are
|
|
|
|
literal if they are preceded by a slash:
|
|
|
|
|
|
|
|
```
|
|
|
|
case '"': /* Literal text */
|
|
|
|
for(o="";fmt[++i] !== '"';) o += fmt[(fmt[i] === '\\' ? ++i : i)];
|
|
|
|
out.push({t:'t', v:o}); break;
|
|
|
|
case '\\': out.push({t:'t', v:fmt[++i]}); ++i; break;
|
|
|
|
```
|
|
|
|
|
|
|
|
The '@' symbol refers to the original text. The ECMA spec is not complete, but
|
|
|
|
Excel does not allow for '@' and non-literal text to appear in the same format.
|
|
|
|
It seems as if they only support one mode. (clearly this is a TODO for excel
|
|
|
|
mode but I'm not convinced that's the right approach)
|
|
|
|
|
|
|
|
```
|
|
|
|
case '@': /* Text Placeholder */
|
|
|
|
out.push({t:'T', v:v}); ++i; break;
|
|
|
|
```
|
|
|
|
|
|
|
|
The date codes `m,d,y,h,s` are standard. There are some special formats like
|
|
|
|
`e` (era year) that have different behaviors in Japanese/Chinese locales.
|
|
|
|
|
|
|
|
```
|
|
|
|
/* Dates */
|
|
|
|
case 'm': case 'd': case 'y': case 'h': case 's': case 'e':
|
|
|
|
if(!dt) dt = parse_date_code(v, opts);
|
|
|
|
o = fmt[i]; while(fmt[++i] === c) o+=c;
|
2013-12-14 03:28:57 +00:00
|
|
|
```
|
|
|
|
|
|
|
|
For the special case of s.00, the suffix should be swallowed with the s:
|
|
|
|
|
|
|
|
```
|
|
|
|
if(c === 's' && fmt[i] === '.' && fmt[i+1] === '0') { o+='.'; while(fmt[++i] === '0') o+= '0'; }
|
|
|
|
```
|
|
|
|
|
|
|
|
Only the forward corrections are made here. The reverse corrections are made later:
|
|
|
|
|
|
|
|
```
|
2013-12-06 16:49:25 +00:00
|
|
|
if(c === 'm' && lst.toLowerCase() === 'h') c = 'M'; /* m = minute */
|
|
|
|
if(c === 'h') c = hr;
|
|
|
|
q={t:c, v:o}; out.push(q); lst = c; break;
|
|
|
|
```
|
|
|
|
|
|
|
|
The (poorly documented) rule regarding `A/P` and `AM/PM` is that if they show up
|
|
|
|
in the format then _all_ instances of `h` are considered 12-hour and not 24-hour
|
|
|
|
format (even in cases like `hh AM/PM hh hh hh`).
|
|
|
|
|
|
|
|
However, the undocumented `H` and `HH` do appear to reset the `AM/PM` indicator.
|
|
|
|
It is not implemented at the moment because I am not 100% sure of the rules with
|
|
|
|
the HH/hh jazz. TODO: investigate this further.
|
|
|
|
|
|
|
|
```
|
|
|
|
case 'A':
|
|
|
|
if(!dt) dt = parse_date_code(v, opts);
|
|
|
|
q={t:c,v:"A"};
|
|
|
|
if(fmt.substr(i, 3) === "A/P") {q.v = dt.H >= 12 ? "P" : "A"; q.t = 'T'; hr='h';i+=3;}
|
|
|
|
else if(fmt.substr(i,5) === "AM/PM") { q.v = dt.H >= 12 ? "PM" : "AM"; q.t = 'T'; i+=5; hr='h'; }
|
|
|
|
else q.t = "t";
|
|
|
|
out.push(q); lst = c; break;
|
2013-12-14 03:28:57 +00:00
|
|
|
```
|
|
|
|
|
|
|
|
Conditional and color blocks should be handled at one point (TODO). For now,
|
|
|
|
only the absolute time `[h]` is captured (using the pseudo-type `Z`):
|
|
|
|
|
|
|
|
```
|
2013-12-06 16:49:25 +00:00
|
|
|
case '[': /* TODO: Fix this -- ignore all conditionals and formatting */
|
2013-12-14 03:28:57 +00:00
|
|
|
o = c;
|
|
|
|
while(fmt[i++] !== ']') o += fmt[i];
|
|
|
|
if(o == "[h]") out.push({t:'Z', v:o});
|
|
|
|
break;
|
|
|
|
```
|
|
|
|
|
|
|
|
Number blocks (following the general pattern `[0#][0#.,E+-%]*`) are grouped together:
|
|
|
|
|
|
|
|
```
|
|
|
|
/* Numbers */
|
|
|
|
case '0': case '#':
|
|
|
|
var nn = ""; while("0#.,E+-%".indexOf(c=fmt[i++]) > -1) nn += c;
|
|
|
|
out.push({t:'n', v:nn}); break;
|
|
|
|
|
|
|
|
```
|
|
|
|
|
|
|
|
The fraction question mark characters present their own challenges. For example, the
|
|
|
|
number 123.456 under format `|??| / |???| |???| foo` is `|15432| / |125| | | foo`:
|
|
|
|
|
|
|
|
```
|
|
|
|
case '?':
|
|
|
|
o = fmt[i]; while(fmt[++i] === c) o+=c;
|
|
|
|
q={t:c, v:o}; out.push(q); lst = c; break;
|
|
|
|
|
2013-12-06 16:49:25 +00:00
|
|
|
default:
|
|
|
|
if("$-+/():!^&'~{}<>= ".indexOf(c) === -1)
|
|
|
|
throw 'unrecognized character ' + fmt[i] + ' in ' + fmt;
|
|
|
|
out.push({t:'t', v:c}); ++i; break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
/* walk backwards */
|
|
|
|
for(i=out.length-1, lst='t'; i >= 0; --i) {
|
|
|
|
switch(out[i].t) {
|
|
|
|
case 'h': case 'H': out[i].t = hr; lst='h'; break;
|
|
|
|
case 'd': case 'y': case 's': case 'M': case 'e': lst=out[i].t; break;
|
|
|
|
case 'm': if(lst === 's') out[i].t = 'M'; break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* replace fields */
|
|
|
|
for(i=0; i < out.length; ++i) {
|
|
|
|
switch(out[i].t) {
|
|
|
|
case 't': case 'T': break;
|
2013-12-14 03:28:57 +00:00
|
|
|
case 'd': case 'm': case 'y': case 'h': case 'H': case 'M': case 's': case 'A': case 'e': case 'Z':
|
2013-12-06 16:49:25 +00:00
|
|
|
out[i].v = write_date(out[i].t, out[i].v, dt);
|
|
|
|
out[i].t = 't'; break;
|
2013-12-14 03:28:57 +00:00
|
|
|
case 'n':
|
|
|
|
out[i].v = write_num(out[i].t, out[i].v, v);
|
|
|
|
out[i].t = 't'; break;
|
2013-12-06 16:49:25 +00:00
|
|
|
default: throw "unrecognized type " + out[i].t;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return out.map(function(x){return x.v;}).join("");
|
|
|
|
}
|
|
|
|
SSF._eval = eval_fmt;
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
|
|
There is some overloading of the `m` character. According to the spec:
|
|
|
|
|
|
|
|
> If "m" or "mm" code is used immediately after the "h" or "hh" code (for
|
|
|
|
hours) or immediately before the "ss" code (for seconds), the application shall
|
|
|
|
display minutes instead of the month.
|
|
|
|
|
|
|
|
|
|
|
|
```js>tmp/date.js
|
|
|
|
var write_date = function(type, fmt, val) {
|
|
|
|
switch(type) {
|
|
|
|
case 'y': switch(fmt) { /* year */
|
|
|
|
case 'y': case 'yy': return pad(val.y % 100,2);
|
|
|
|
default: return val.y;
|
|
|
|
} break;
|
|
|
|
case 'm': switch(fmt) { /* month */
|
|
|
|
case 'm': return val.m;
|
|
|
|
case 'mm': return pad(val.m,2);
|
|
|
|
case 'mmm': return months[val.m-1][1];
|
|
|
|
case 'mmmm': return months[val.m-1][2];
|
|
|
|
case 'mmmmm': return months[val.m-1][0];
|
|
|
|
default: throw 'bad month format: ' + fmt;
|
|
|
|
} break;
|
|
|
|
case 'd': switch(fmt) { /* day */
|
|
|
|
case 'd': return val.d;
|
|
|
|
case 'dd': return pad(val.d,2);
|
|
|
|
case 'ddd': return days[val.q][0];
|
|
|
|
case 'dddd': return days[val.q][1];
|
|
|
|
default: throw 'bad day format: ' + fmt;
|
|
|
|
} break;
|
|
|
|
case 'h': switch(fmt) { /* 12-hour */
|
|
|
|
case 'h': return 1+(val.H+11)%12;
|
|
|
|
case 'hh': return pad(1+(val.H+11)%12, 2);
|
|
|
|
default: throw 'bad hour format: ' + fmt;
|
|
|
|
} break;
|
|
|
|
case 'H': switch(fmt) { /* 24-hour */
|
|
|
|
case 'h': return val.H;
|
|
|
|
case 'hh': return pad(val.H, 2);
|
|
|
|
default: throw 'bad hour format: ' + fmt;
|
|
|
|
} break;
|
|
|
|
case 'M': switch(fmt) { /* minutes */
|
|
|
|
case 'm': return val.M;
|
|
|
|
case 'mm': return pad(val.M, 2);
|
|
|
|
default: throw 'bad minute format: ' + fmt;
|
|
|
|
} break;
|
|
|
|
case 's': switch(fmt) { /* seconds */
|
|
|
|
case 's': return val.S;
|
|
|
|
case 'ss': return pad(val.S, 2);
|
2013-12-14 03:28:57 +00:00
|
|
|
case 'ss.0': console.log(val);
|
2013-12-06 16:49:25 +00:00
|
|
|
default: throw 'bad second format: ' + fmt;
|
|
|
|
} break;
|
|
|
|
```
|
|
|
|
|
2013-12-14 03:28:57 +00:00
|
|
|
The `Z` type refers to absolute time measures:
|
|
|
|
|
|
|
|
```
|
|
|
|
case 'Z': switch(fmt) {
|
|
|
|
case '[h]': return val.D*24+val.H;
|
|
|
|
default: throw 'bad abstime format: ' + fmt;
|
|
|
|
} break;
|
|
|
|
```
|
|
|
|
|
2013-12-06 16:49:25 +00:00
|
|
|
The `e` format behavior in excel diverges from the spec. It claims that `ee`
|
|
|
|
should be a two-digit year, but `ee` in excel is actually the four-digit year:
|
|
|
|
|
|
|
|
```
|
|
|
|
/* TODO: handle the ECMA spec format ee -> yy */
|
|
|
|
case 'e': { return val.y; } break;
|
|
|
|
case 'A': return (val.h>=12 ? 'P' : 'A') + fmt.substr(1);
|
|
|
|
default: throw 'bad format type ' + type + ' in ' + fmt;
|
|
|
|
}
|
|
|
|
};
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
```js>tmp/main.js
|
|
|
|
function choose_fmt(fmt, v) {
|
|
|
|
if(typeof fmt === "string") fmt = split_fmt(fmt);
|
|
|
|
if(typeof v !== "number") return fmt[3];
|
|
|
|
return v > 0 ? fmt[0] : v < 0 ? fmt[1] : fmt[2];
|
|
|
|
}
|
|
|
|
|
|
|
|
var format = function format(fmt,v,o) {
|
|
|
|
fixopts(o = (o||{}));
|
|
|
|
if(fmt === 0) return general_fmt(v, o);
|
|
|
|
if(typeof fmt === 'number') fmt = table_fmt[fmt];
|
|
|
|
var f = choose_fmt(fmt, v, o);
|
|
|
|
return eval_fmt(f, v, o);
|
|
|
|
};
|
|
|
|
|
|
|
|
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
```js>tmp/main.js
|
|
|
|
|
|
|
|
SSF._choose = choose_fmt;
|
|
|
|
SSF._table = table_fmt;
|
|
|
|
SSF.load = function(fmt, idx) { table_fmt[idx] = fmt; };
|
|
|
|
SSF.format = format;
|
|
|
|
```
|
|
|
|
|
|
|
|
## JS Boilerplate
|
|
|
|
|
|
|
|
```js>tmp/00_header.js
|
2013-12-14 03:28:57 +00:00
|
|
|
/* ssf.js (C) 2013 SheetJS -- http://sheetjs.com */
|
2013-12-06 16:49:25 +00:00
|
|
|
var SSF;
|
|
|
|
(function(SSF){
|
|
|
|
String.prototype.reverse=function(){return this.split("").reverse().join("");};
|
|
|
|
var _strrev = function(x) { return String(x).reverse(); };
|
|
|
|
function fill(c,l) { return new Array(l+1).join(c); }
|
|
|
|
function pad(v,d){var t=String(v);return t.length>=d?t:(fill(0,d-t.length)+t);}
|
|
|
|
```
|
|
|
|
|
|
|
|
```js>tmp/zz_footer_n.js
|
|
|
|
})(typeof exports !== 'undefined' ? exports : SSF);
|
|
|
|
```
|
|
|
|
|
|
|
|
```js>tmp/zz_footer.js
|
|
|
|
})(SSF);
|
|
|
|
```
|
|
|
|
|
|
|
|
## .vocrc and post-commands
|
|
|
|
|
|
|
|
```bash>tmp/post.sh
|
|
|
|
#!/bin/bash
|
|
|
|
npm install
|
2013-12-14 03:28:57 +00:00
|
|
|
cat tmp/{00_header,opts,consts,general,date,number,main,zz_footer_n}.js > ssf_node.js
|
|
|
|
cat tmp/{00_header,opts,consts,general,date,number,main,zz_footer}.js > ssf.js
|
2013-12-06 16:49:25 +00:00
|
|
|
```
|
|
|
|
|
|
|
|
```json>.vocrc
|
|
|
|
{
|
|
|
|
"post": "bash tmp/post.sh"
|
|
|
|
}
|
|
|
|
```
|
|
|
|
|
|
|
|
```>.gitignore
|
|
|
|
.gitignore
|
|
|
|
tmp/
|
|
|
|
node_modules/
|
|
|
|
.vocrc
|
|
|
|
```
|
|
|
|
|
2013-12-14 03:28:57 +00:00
|
|
|
```make>Makefile
|
|
|
|
.PHONY: test ssf
|
|
|
|
ssf: ssf.md
|
|
|
|
voc ssf.md
|
|
|
|
|
|
|
|
test:
|
|
|
|
npm test
|
|
|
|
```
|
|
|
|
|
2013-12-06 16:49:25 +00:00
|
|
|
```json>package.json
|
|
|
|
{
|
|
|
|
"name": "ssf",
|
2013-12-14 03:28:57 +00:00
|
|
|
"version": "0.2.1",
|
2013-12-06 16:49:25 +00:00
|
|
|
"author": "SheetJS",
|
|
|
|
"description": "pure-JS library to format data using ECMA-376 spreadsheet Format Codes",
|
|
|
|
"keywords": [ "format", "sprintf", "spreadsheet" ],
|
|
|
|
"main": "ssf_node.js",
|
|
|
|
"dependencies": {
|
|
|
|
"voc":"",
|
|
|
|
"colors":""
|
|
|
|
},
|
|
|
|
"devDependencies": {
|
|
|
|
"mocha":""
|
|
|
|
},
|
|
|
|
"repository": { "type":"git", "url":"git://github.com/SheetJS/ssf.git" },
|
|
|
|
"scripts": {
|
|
|
|
"test": "mocha -R spec"
|
|
|
|
},
|
|
|
|
"bugs": { "url": "https://github.com/SheetJS/ssf/issues" },
|
|
|
|
"license": "Apache-2.0",
|
|
|
|
"engines": { "node": ">=0.8" }
|
|
|
|
}
|
|
|
|
```
|
|
|
|
|
|
|
|
# Test Driver
|
|
|
|
|
|
|
|
Travis CI is used for node testing:
|
|
|
|
|
|
|
|
```>.travis.yml
|
|
|
|
language: node_js
|
|
|
|
node_js:
|
|
|
|
- "0.10"
|
|
|
|
- "0.8"
|
|
|
|
before_install:
|
|
|
|
- "npm install -g mocha"
|
|
|
|
```
|
|
|
|
|
|
|
|
The mocha test driver tests the implied formats:
|
|
|
|
|
|
|
|
```js>test/implied.js
|
|
|
|
/* vim: set ts=2: */
|
|
|
|
var SSF = require('../');
|
|
|
|
var fs = require('fs'), assert = require('assert');
|
|
|
|
var data = JSON.parse(fs.readFileSync('./test/implied.json','utf8'));
|
2013-12-14 03:28:57 +00:00
|
|
|
var skip = [12, 13, 47, 48];
|
2013-12-06 16:49:25 +00:00
|
|
|
describe('implied formats', function() {
|
|
|
|
data.forEach(function(d) {
|
2013-12-14 03:28:57 +00:00
|
|
|
it(d[1]+" for "+d[0], skip.indexOf(d[1]) > -1 ? null : function(){
|
2013-12-06 16:49:25 +00:00
|
|
|
assert.equal(SSF.format(d[1], d[0], {}), d[2]);
|
|
|
|
});
|
|
|
|
});
|
|
|
|
});
|
|
|
|
```
|
|
|
|
|
|
|
|
The old test driver was manual:
|
|
|
|
|
|
|
|
```js>tmp/test.njs
|
|
|
|
var SSF = require('../ssf_node');
|
|
|
|
var x = 'd\\-mmm\\-yy\\ yyyy\\ dd\\ \\;\\ yy\\ mm\\ dd';
|
|
|
|
var y = 'd\\-mmm\\-yy\\ yyyy\\ dd\\ ;\\ yy\\ mm\\ dd';
|
|
|
|
var z = 'd\\ dd\\ ddd\\ dddd\\ m\\ mm\\ mmm\\ mmmm\\ mmmmm\\ yy\\ yyyy';
|
|
|
|
console.error(SSF.parse_date_code(65.9));
|
|
|
|
console.error(SSF.format(x, 65.9));
|
|
|
|
console.error(SSF.format(y, 65.9));
|
|
|
|
console.error()
|
|
|
|
console.error(SSF.format(z, 55.9));
|
|
|
|
console.error(SSF.format(z, 55.9, {mode:"excel"}));
|
|
|
|
console.error(SSF.format(z, 55.9));
|
|
|
|
console.error()
|
|
|
|
console.error(SSF.format(z, 65.9));
|
|
|
|
console.error(SSF.format(z, 65.9, {mode:"excel"}));
|
|
|
|
console.error(SSF.format(z, 65.9));
|
|
|
|
console.error()
|
|
|
|
console.error(SSF.format(19, 65.9));
|
|
|
|
console.error(SSF.format(20, 65.9));
|
|
|
|
```
|
|
|
|
|
|
|
|
# LICENSE
|
|
|
|
|
|
|
|
```>LICENSE
|
|
|
|
Copyright 2013 SheetJS
|
|
|
|
|
|
|
|
Licensed under the Apache License, Version 2.0 (the "License");
|
|
|
|
you may not use this file except in compliance with the License.
|
|
|
|
You may obtain a copy of the License at
|
|
|
|
|
|
|
|
http://www.apache.org/licenses/LICENSE-2.0
|
|
|
|
|
|
|
|
Unless required by applicable law or agreed to in writing, software
|
|
|
|
distributed under the License is distributed on an "AS IS" BASIS,
|
|
|
|
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
See the License for the specific language governing permissions and
|
|
|
|
limitations under the License.
|
|
|
|
```
|