2016-08-27 03:27:32 +00:00
|
|
|
/* jslint node: true */
|
|
|
|
'use strict';
|
|
|
|
|
2016-09-05 03:36:26 +00:00
|
|
|
const EnigError = require('./enig_error.js').EnigError;
|
2017-09-09 19:59:23 +00:00
|
|
|
|
|
|
|
const {
|
|
|
|
pad,
|
2018-01-15 19:22:11 +00:00
|
|
|
stylizeString,
|
2017-09-09 19:59:23 +00:00
|
|
|
renderStringLength,
|
|
|
|
renderSubstr,
|
|
|
|
formatByteSize, formatByteSizeAbbr,
|
|
|
|
formatCount, formatCountAbbr,
|
|
|
|
} = require('./string_util.js');
|
2016-08-27 03:27:32 +00:00
|
|
|
|
|
|
|
// deps
|
|
|
|
const _ = require('lodash');
|
|
|
|
|
|
|
|
/*
|
|
|
|
String formatting HEAVILY inspired by David Chambers string-format library
|
|
|
|
and the mini-language branch specifically which was gratiously released
|
|
|
|
under the DO WHAT THE FUCK YOU WANT TO PUBLIC LICENSE.
|
|
|
|
|
|
|
|
We need some extra functionality. Namely, support for RA style pipe codes
|
|
|
|
and ANSI escape sequences.
|
|
|
|
*/
|
|
|
|
|
|
|
|
class ValueError extends EnigError { }
|
|
|
|
class KeyError extends EnigError { }
|
|
|
|
|
|
|
|
const SpecRegExp = {
|
|
|
|
FillAlign : /^(.)?([<>=^])/,
|
|
|
|
Sign : /^[ +-]/,
|
|
|
|
Width : /^\d*/,
|
|
|
|
Precision : /^\d+/,
|
|
|
|
};
|
|
|
|
|
|
|
|
function tokenizeFormatSpec(spec) {
|
|
|
|
const tokens = {
|
|
|
|
fill : '',
|
|
|
|
align : '',
|
|
|
|
sign : '',
|
|
|
|
'#' : false,
|
|
|
|
'0' : false,
|
|
|
|
width : '',
|
|
|
|
',' : false,
|
|
|
|
precision : '',
|
|
|
|
type : '',
|
|
|
|
};
|
|
|
|
|
|
|
|
let index = 0;
|
|
|
|
let match;
|
|
|
|
|
|
|
|
function incIndexByMatch() {
|
|
|
|
index += match[0].length;
|
|
|
|
}
|
|
|
|
|
|
|
|
match = SpecRegExp.FillAlign.exec(spec);
|
|
|
|
if(match) {
|
|
|
|
if(match[1]) {
|
|
|
|
tokens.fill = match[1];
|
|
|
|
}
|
|
|
|
tokens.align = match[2];
|
|
|
|
incIndexByMatch();
|
|
|
|
}
|
|
|
|
|
|
|
|
match = SpecRegExp.Sign.exec(spec.slice(index));
|
|
|
|
if(match) {
|
|
|
|
tokens.sign = match[0];
|
|
|
|
incIndexByMatch();
|
|
|
|
}
|
|
|
|
|
|
|
|
if('#' === spec.charAt(index)) {
|
|
|
|
tokens['#'] = true;
|
|
|
|
++index;
|
|
|
|
}
|
|
|
|
|
|
|
|
if('0' === spec.charAt(index)) {
|
|
|
|
tokens['0'] = true;
|
|
|
|
++index;
|
|
|
|
}
|
|
|
|
|
|
|
|
match = SpecRegExp.Width.exec(spec.slice(index));
|
|
|
|
tokens.width = match[0];
|
|
|
|
incIndexByMatch();
|
|
|
|
|
|
|
|
if(',' === spec.charAt(index)) {
|
|
|
|
tokens[','] = true;
|
|
|
|
++index;
|
|
|
|
}
|
|
|
|
|
|
|
|
if('.' === spec.charAt(index)) {
|
|
|
|
++index;
|
|
|
|
|
|
|
|
match = SpecRegExp.Precision.exec(spec.slice(index));
|
|
|
|
if(!match) {
|
|
|
|
throw new ValueError('Format specifier missing precision');
|
|
|
|
}
|
|
|
|
|
|
|
|
tokens.precision = match[0];
|
|
|
|
incIndexByMatch();
|
|
|
|
}
|
|
|
|
|
|
|
|
if(index < spec.length) {
|
|
|
|
tokens.type = spec.charAt(index);
|
|
|
|
++index;
|
|
|
|
}
|
|
|
|
|
|
|
|
if(index < spec.length) {
|
|
|
|
throw new ValueError('Invalid conversion specification');
|
|
|
|
}
|
|
|
|
|
|
|
|
if(tokens[','] && 's' === tokens.type) {
|
|
|
|
throw new ValueError(`Cannot specify ',' with 's'`); // eslint-disable-line quotes
|
|
|
|
}
|
|
|
|
|
|
|
|
return tokens;
|
|
|
|
}
|
|
|
|
|
|
|
|
function quote(s) {
|
|
|
|
return `"${s.replace(/"/g, '\\"')}"`;
|
|
|
|
}
|
|
|
|
|
|
|
|
function getPadAlign(align) {
|
|
|
|
return {
|
2018-01-15 23:08:35 +00:00
|
|
|
'<' : 'left',
|
|
|
|
'>' : 'right',
|
2016-08-27 03:27:32 +00:00
|
|
|
'^' : 'center',
|
2018-01-15 23:08:35 +00:00
|
|
|
}[align] || '>';
|
2016-08-27 03:27:32 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
function formatString(value, tokens) {
|
|
|
|
const fill = tokens.fill || (tokens['0'] ? '0' : ' ');
|
|
|
|
const align = tokens.align || (tokens['0'] ? '=' : '<');
|
2016-09-05 03:36:26 +00:00
|
|
|
const precision = Number(tokens.precision || renderStringLength(value) + 1);
|
2016-08-27 03:27:32 +00:00
|
|
|
|
|
|
|
if('' !== tokens.type && 's' !== tokens.type) {
|
|
|
|
throw new ValueError(`Unknown format code "${tokens.type}" for String object`);
|
|
|
|
}
|
|
|
|
|
|
|
|
if(tokens[',']) {
|
|
|
|
throw new ValueError(`Cannot specify ',' with 's'`); // eslint-disable-line quotes
|
|
|
|
}
|
|
|
|
|
|
|
|
if(tokens.sign) {
|
|
|
|
throw new ValueError('Sign not allowed in string format specifier');
|
|
|
|
}
|
|
|
|
|
|
|
|
if(tokens['#']) {
|
|
|
|
throw new ValueError('Alternate form (#) not allowed in string format specifier');
|
|
|
|
}
|
|
|
|
|
|
|
|
if('=' === align) {
|
|
|
|
throw new ValueError('"=" alignment not allowed in string format specifier');
|
|
|
|
}
|
|
|
|
|
2016-09-05 03:36:26 +00:00
|
|
|
return pad(renderSubstr(value, 0, precision), Number(tokens.width), fill, getPadAlign(align));
|
2016-08-27 03:27:32 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
const FormatNumRegExp = {
|
|
|
|
UpperType : /[A-Z]/,
|
|
|
|
ExponentRep : /e[+-](?=\d$)/,
|
|
|
|
};
|
|
|
|
|
|
|
|
function formatNumberHelper(n, precision, type) {
|
|
|
|
if(FormatNumRegExp.UpperType.test(type)) {
|
|
|
|
return formatNumberHelper(n, precision, type.toLowerCase()).toUpperCase();
|
|
|
|
}
|
|
|
|
|
|
|
|
switch(type) {
|
|
|
|
case 'c' : return String.fromCharCode(n);
|
|
|
|
case 'd' : return n.toString(10);
|
|
|
|
case 'b' : return n.toString(2);
|
|
|
|
case 'o' : return n.toString(8);
|
|
|
|
case 'x' : return n.toString(16);
|
2018-01-15 19:22:11 +00:00
|
|
|
case 'e' : return n.toExponential(precision).replace(FormatNumRegExp.ExponentRep, '$&0');
|
|
|
|
case 'f' : return n.toFixed(precision);
|
2016-09-05 03:36:26 +00:00
|
|
|
case 'g' :
|
|
|
|
// we don't want useless trailing zeros. parseFloat -> back to string fixes this for us
|
|
|
|
return parseFloat(n.toPrecision(precision || 1)).toString();
|
|
|
|
|
2016-08-27 03:27:32 +00:00
|
|
|
case '%' : return formatNumberHelper(n * 100, precision, 'f') + '%';
|
|
|
|
case '' : return formatNumberHelper(n, precision, 'd');
|
2018-01-15 19:22:11 +00:00
|
|
|
|
2016-08-27 03:27:32 +00:00
|
|
|
default :
|
|
|
|
throw new ValueError(`Unknown format code "${type}" for object of type 'float'`);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
function formatNumber(value, tokens) {
|
|
|
|
const fill = tokens.fill || (tokens['0'] ? '0' : ' ');
|
|
|
|
const align = tokens.align || (tokens['0'] ? '=' : '>');
|
|
|
|
const width = Number(tokens.width);
|
|
|
|
const type = tokens.type || (tokens.precision ? 'g' : '');
|
|
|
|
|
|
|
|
if( [ 'c', 'd', 'b', 'o', 'x', 'X' ].indexOf(type) > -1) {
|
|
|
|
if(0 !== value % 1) {
|
|
|
|
throw new ValueError(`Cannot format non-integer with format specifier "${type}"`);
|
|
|
|
}
|
|
|
|
|
|
|
|
if('' !== tokens.sign && 'c' !== type) {
|
|
|
|
throw new ValueError(`Sign not allowed with integer format specifier 'c'`); // eslint-disable-line quotes
|
|
|
|
}
|
|
|
|
|
|
|
|
if(tokens[','] && 'd' !== type) {
|
|
|
|
throw new ValueError(`Cannot specify ',' with '${type}'`);
|
|
|
|
}
|
|
|
|
|
|
|
|
if('' !== tokens.precision) {
|
|
|
|
throw new ValueError('Precision not allowed in integer format specifier');
|
2018-01-15 19:22:11 +00:00
|
|
|
}
|
2016-08-27 03:27:32 +00:00
|
|
|
} else if( [ 'e', 'E', 'f', 'F', 'g', 'G', '%' ].indexOf(type) > - 1) {
|
|
|
|
if(tokens['#']) {
|
|
|
|
throw new ValueError('Alternate form (#) not allowed in float format specifier');
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
const s = formatNumberHelper(Math.abs(value), Number(tokens.precision || 6), type);
|
2018-01-15 19:22:11 +00:00
|
|
|
const sign = value < 0 || 1 / value < 0 ?
|
2016-08-27 03:27:32 +00:00
|
|
|
'-' :
|
|
|
|
'-' === tokens.sign ? '' : tokens.sign;
|
|
|
|
|
|
|
|
const prefix = tokens['#'] && ( [ 'b', 'o', 'x', 'X' ].indexOf(type) > -1 ) ? '0' + type : '';
|
|
|
|
|
|
|
|
if(tokens[',']) {
|
|
|
|
const match = /^(\d*)(.*)$/.exec(s);
|
2018-01-15 19:22:11 +00:00
|
|
|
const separated = match[1].replace(/.(?=(...)+$)/g, '$&,') + match[2];
|
2016-08-27 03:27:32 +00:00
|
|
|
|
|
|
|
if('=' !== align) {
|
|
|
|
return pad(sign + separated, width, fill, getPadAlign(align));
|
|
|
|
}
|
|
|
|
|
|
|
|
if('0' === fill) {
|
|
|
|
const shortfall = Math.max(0, width - sign.length - separated.length);
|
|
|
|
const digits = /^\d*/.exec(separated)[0].length;
|
|
|
|
let padding = '';
|
|
|
|
// :TODO: do this differntly...
|
|
|
|
for(let n = 0; n < shortfall; n++) {
|
|
|
|
padding = ((digits + n) % 4 === 3 ? ',' : '0') + padding;
|
|
|
|
}
|
|
|
|
|
|
|
|
return sign + (/^,/.test(padding) ? '0' : '') + padding + separated;
|
|
|
|
}
|
|
|
|
|
|
|
|
return sign + pad(separated, width - sign.length, fill, getPadAlign('>'));
|
|
|
|
}
|
|
|
|
|
|
|
|
if(0 === width) {
|
|
|
|
return sign + prefix + s;
|
2018-01-15 19:22:11 +00:00
|
|
|
}
|
2016-08-27 03:27:32 +00:00
|
|
|
|
|
|
|
if('=' === align) {
|
|
|
|
return sign + prefix + pad(s, width - sign.length - prefix.length, fill, getPadAlign('>'));
|
|
|
|
}
|
|
|
|
|
|
|
|
return pad(sign + prefix + s, width, fill, getPadAlign(align));
|
|
|
|
}
|
|
|
|
|
|
|
|
const transformers = {
|
|
|
|
// String standard
|
|
|
|
toUpperCase : String.prototype.toUpperCase,
|
|
|
|
toLowerCase : String.prototype.toLowerCase,
|
|
|
|
|
|
|
|
// some super l33b BBS styles!!
|
|
|
|
styleUpper : (s) => stylizeString(s, 'upper'),
|
|
|
|
styleLower : (s) => stylizeString(s, 'lower'),
|
|
|
|
styleTitle : (s) => stylizeString(s, 'title'),
|
|
|
|
styleFirstLower : (s) => stylizeString(s, 'first lower'),
|
|
|
|
styleSmallVowels : (s) => stylizeString(s, 'small vowels'),
|
|
|
|
styleBigVowels : (s) => stylizeString(s, 'big vowels'),
|
|
|
|
styleSmallI : (s) => stylizeString(s, 'small i'),
|
|
|
|
styleMixed : (s) => stylizeString(s, 'mixed'),
|
|
|
|
styleL33t : (s) => stylizeString(s, 'l33t'),
|
2016-09-29 03:54:25 +00:00
|
|
|
|
2017-09-22 03:23:30 +00:00
|
|
|
// :TODO:
|
2018-01-15 19:22:11 +00:00
|
|
|
// toMegs(), toKilobytes(), ...
|
|
|
|
// toList(), toCommaList(),
|
|
|
|
|
2016-09-29 03:54:25 +00:00
|
|
|
sizeWithAbbr : (n) => formatByteSize(n, true, 2),
|
|
|
|
sizeWithoutAbbr : (n) => formatByteSize(n, false, 2),
|
|
|
|
sizeAbbr : (n) => formatByteSizeAbbr(n),
|
2017-09-09 19:59:23 +00:00
|
|
|
countWithAbbr : (n) => formatCount(n, true, 0),
|
|
|
|
countWithoutAbbr : (n) => formatCount(n, false, 0),
|
|
|
|
countAbbr : (n) => formatCountAbbr(n),
|
2016-08-27 03:27:32 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
function transformValue(transformerName, value) {
|
|
|
|
if(transformerName in transformers) {
|
|
|
|
const transformer = transformers[transformerName];
|
|
|
|
value = transformer.apply(value, [ value ] );
|
|
|
|
}
|
|
|
|
|
|
|
|
return value;
|
|
|
|
}
|
|
|
|
|
|
|
|
// :TODO: Use explicit set of chars for paths & function/transforms such that } is allowed as fill/etc.
|
2018-01-15 19:22:11 +00:00
|
|
|
const REGEXP_BASIC_FORMAT = /{([^.!:}]+(?:\.[^.!:}]+)*)(?:!([^:}]+))?(?::([^}]+))?}/g;
|
2016-08-27 03:27:32 +00:00
|
|
|
|
|
|
|
function getValue(obj, path) {
|
|
|
|
const value = _.get(obj, path);
|
2016-09-05 03:36:26 +00:00
|
|
|
if(!_.isUndefined(value)) {
|
2016-08-27 03:27:32 +00:00
|
|
|
return _.isFunction(value) ? value() : value;
|
|
|
|
}
|
2018-01-15 19:22:11 +00:00
|
|
|
|
2016-08-27 03:27:32 +00:00
|
|
|
throw new KeyError(quote(path));
|
|
|
|
}
|
|
|
|
|
|
|
|
module.exports = function format(fmt, obj) {
|
|
|
|
|
|
|
|
const re = REGEXP_BASIC_FORMAT;
|
2016-09-07 05:20:11 +00:00
|
|
|
re.lastIndex = 0; // reset from prev
|
2016-09-05 03:36:26 +00:00
|
|
|
|
2016-08-27 03:27:32 +00:00
|
|
|
let match;
|
|
|
|
let pos;
|
|
|
|
let out = '';
|
2016-09-05 03:36:26 +00:00
|
|
|
let objPath ;
|
|
|
|
let transformer;
|
|
|
|
let formatSpec;
|
|
|
|
let value;
|
|
|
|
let tokens;
|
|
|
|
|
2016-08-27 03:27:32 +00:00
|
|
|
do {
|
|
|
|
pos = re.lastIndex;
|
|
|
|
match = re.exec(fmt);
|
|
|
|
|
|
|
|
if(match) {
|
|
|
|
if(match.index > pos) {
|
|
|
|
out += fmt.slice(pos, match.index);
|
|
|
|
}
|
|
|
|
|
2016-09-05 03:36:26 +00:00
|
|
|
objPath = match[1];
|
|
|
|
transformer = match[2];
|
|
|
|
formatSpec = match[3];
|
2016-08-27 03:27:32 +00:00
|
|
|
|
2016-09-05 03:36:26 +00:00
|
|
|
value = getValue(obj, objPath);
|
2016-08-27 03:27:32 +00:00
|
|
|
if(transformer) {
|
|
|
|
value = transformValue(transformer, value);
|
|
|
|
}
|
|
|
|
|
2016-09-05 03:36:26 +00:00
|
|
|
tokens = tokenizeFormatSpec(formatSpec || '');
|
2016-08-27 03:27:32 +00:00
|
|
|
|
2016-09-09 04:32:24 +00:00
|
|
|
if(_.isNumber(value)) {
|
2016-08-27 03:27:32 +00:00
|
|
|
out += formatNumber(value, tokens);
|
|
|
|
} else {
|
|
|
|
out += formatString(value, tokens);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
} while(0 !== re.lastIndex);
|
|
|
|
|
|
|
|
// remainder
|
|
|
|
if(pos < fmt.length) {
|
|
|
|
out += fmt.slice(pos);
|
2018-01-15 19:22:11 +00:00
|
|
|
}
|
2016-08-27 03:27:32 +00:00
|
|
|
|
2018-01-15 19:22:11 +00:00
|
|
|
return out;
|
2016-08-27 03:27:32 +00:00
|
|
|
};
|