955 lines
38 KiB
Go
955 lines
38 KiB
Go
// Copyright 2016 - 2022 The excelize Authors. All rights reserved. Use of
|
||
// this source code is governed by a BSD-style license that can be found in
|
||
// the LICENSE file.
|
||
//
|
||
// Package excelize providing a set of functions that allow you to write to and
|
||
// read from XLAM / XLSM / XLSX / XLTM / XLTX files. Supports reading and
|
||
// writing spreadsheet documents generated by Microsoft Excel™ 2007 and later.
|
||
// Supports complex components by high compatibility, and provided streaming
|
||
// API for generating or reading data from a worksheet with huge amounts of
|
||
// data. This library needs Go version 1.15 or later.
|
||
|
||
package excelize
|
||
|
||
import (
|
||
"fmt"
|
||
"math"
|
||
"strconv"
|
||
"strings"
|
||
"time"
|
||
|
||
"github.com/xuri/nfp"
|
||
)
|
||
|
||
// languageInfo defined the required fields of localization support for number format.
|
||
type languageInfo struct {
|
||
apFmt string
|
||
tags []string
|
||
localMonth func(t time.Time, abbr int) string
|
||
}
|
||
|
||
// numberFormat directly maps the number format parser runtime required
|
||
// fields.
|
||
type numberFormat struct {
|
||
section []nfp.Section
|
||
t time.Time
|
||
sectionIdx int
|
||
date1904, isNumeric, hours, seconds bool
|
||
number float64
|
||
ap, localCode, result, value, valueSectionType string
|
||
}
|
||
|
||
var (
|
||
// supportedTokenTypes list the supported number format token types currently.
|
||
supportedTokenTypes = []string{
|
||
nfp.TokenSubTypeLanguageInfo,
|
||
nfp.TokenTypeColor,
|
||
nfp.TokenTypeCurrencyLanguage,
|
||
nfp.TokenTypeDateTimes,
|
||
nfp.TokenTypeElapsedDateTimes,
|
||
nfp.TokenTypeGeneral,
|
||
nfp.TokenTypeLiteral,
|
||
nfp.TokenTypeTextPlaceHolder,
|
||
nfp.TokenTypeZeroPlaceHolder,
|
||
}
|
||
// supportedLanguageInfo directly maps the supported language ID and tags.
|
||
supportedLanguageInfo = map[string]languageInfo{
|
||
"36": {tags: []string{"af"}, localMonth: localMonthsNameAfrikaans, apFmt: apFmtAfrikaans},
|
||
"445": {tags: []string{"bn-IN"}, localMonth: localMonthsNameBangla, apFmt: nfp.AmPm[0]},
|
||
"4": {tags: []string{"zh-Hans"}, localMonth: localMonthsNameChinese1, apFmt: nfp.AmPm[2]},
|
||
"7804": {tags: []string{"zh"}, localMonth: localMonthsNameChinese1, apFmt: nfp.AmPm[2]},
|
||
"804": {tags: []string{"zh-CN"}, localMonth: localMonthsNameChinese1, apFmt: nfp.AmPm[2]},
|
||
"1004": {tags: []string{"zh-SG"}, localMonth: localMonthsNameChinese2, apFmt: nfp.AmPm[2]},
|
||
"7C04": {tags: []string{"zh-Hant"}, localMonth: localMonthsNameChinese3, apFmt: nfp.AmPm[2]},
|
||
"C04": {tags: []string{"zh-HK"}, localMonth: localMonthsNameChinese2, apFmt: nfp.AmPm[2]},
|
||
"1404": {tags: []string{"zh-MO"}, localMonth: localMonthsNameChinese3, apFmt: nfp.AmPm[2]},
|
||
"404": {tags: []string{"zh-TW"}, localMonth: localMonthsNameChinese3, apFmt: nfp.AmPm[2]},
|
||
"9": {tags: []string{"en"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"1000": {tags: []string{
|
||
"aa", "aa-DJ", "aa-ER", "aa-ER", "aa-NA", "agq", "agq-CM", "ak", "ak-GH", "sq-ML",
|
||
"gsw-LI", "gsw-CH", "ar-TD", "ar-KM", "ar-DJ", "ar-ER", "ar-IL", "ar-MR", "ar-PS",
|
||
"ar-SO", "ar-SS", "ar-SD", "ar-001", "ast", "ast-ES", "asa", "asa-TZ", "ksf", "ksf-CM",
|
||
"bm", "bm-Latn-ML", "bas", "bas-CM", "bem", "bem-ZM", "bez", "bez-TZ", "byn", "byn-ER",
|
||
"brx", "brx-IN", "ca-AD", "ca-FR", "ca-IT", "ceb", "ceb-Latn", "ceb-Latn-PH", "tzm-Latn-MA",
|
||
"ccp", "ccp-Cakm", "ccp-Cakm-BD", "ccp-Cakm-IN", "ce-RU", "cgg", "cgg-UG", "cu-RU", "swc",
|
||
"swc-CD", "kw", "ke-GB", "da-GL", "dua", "dua-CM", "nl-AW", "nl-BQ", "nl-CW", "nl-SX",
|
||
"nl-SR", "dz", "ebu", "ebu-KE", "en-AS", "en-AI", "en-AG", "en-AT", "en-BS", "en-BB",
|
||
"en-BE", "en-BM", "en-BW", "en-IO", "en-VG", "en-BI", "en-CM", "en-KY", "en-CX", "en-CC",
|
||
"en-CK", "en-CY", "en-DK", "en-DM", "en-ER", "en-150", "en-FK", "en-FI", "en-FJ", "en-GM",
|
||
"en-DE", "en-GH", "en-GI", "en-GD", "en-GU", "en-GG", "en-GY", "en-IM", "en-IL", "en-JE",
|
||
"en-KE", "en-KI", "en-LS", "en-LR", "en-MO", "en-MG", "en-MW", "en-MT", "en-MH", "en-MU",
|
||
"en-FM", "en-MS", "en-NA", "en-NR", "en-NL", "en-NG", "en-NU", "en-NF", "en-MP", "en-PK",
|
||
"en-PW", "en-PG", "en-PN", "en-PR", "en-RW", "en-KN", "en-LC", "en-VC", "en-WS", "en-SC",
|
||
"en-SL", "en-SX", "en-SI", "en-SB", "en-SS", "en-SH", "en-SD", "en-SZ", "en-SE", "en-CH",
|
||
"en-TZ", "en-TK", "en-TO", "en-TC", "en-TV", "en-UG", "en-UM", "en-VI", "en-VU", "en-001",
|
||
"en-ZM", "eo", "eo-001", "ee", "ee-GH", "ee-TG", "ewo", "ewo-CM", "fo-DK", "fr-DZ",
|
||
"fr-BJ", "fr-BF", "fr-BI", "fr-CF", "fr-TD", "fr-KM", "fr-CG", "fr-DJ", "fr-GQ", "fr-GF",
|
||
"fr-PF", "fr-GA", "fr-GP", "fr-GN", "fr-MG", "fr-MQ", "fr-MR", "fr-MU", "fr-YT", "fr-NC",
|
||
"fr-NE", "fr-RW", "fr-BL", "fr-MF", "fr-PM", "fr-SC", "fr-SY", "fr-TG", "fr-TN", "fr-VU",
|
||
"fr-WF", "fur", "fur-IT", "ff-Latn-BF", "ff-CM", "ff-Latn-CM", "ff-Latn-GM", "ff-Latn-GH",
|
||
"ff-GN", "ff-Latn-GN", "ff-Latn-GW", "ff-Latn-LR", "ff-MR", "ff-Latn-MR", "ff-Latn-NE",
|
||
"ff-Latn-SL", "lg", "lg-UG", "de-BE", "de-IT", "el-CY", "guz", "guz-KE", "ha-Latn-GH",
|
||
"ha-Latn-NG", "ia-FR", "ia-001", "it-SM", "it-VA", "jv", "jv-Latn", "jv-Latn-ID", "dyo",
|
||
"dyo-SN", "kea", "kea-CV", "kab", "kab-DZ", "kkj", "kkj-CM", "kln", "kln-KE", "kam",
|
||
"kam-KE", "ks-Arab-IN", "ki", "ki-KE", "sw-TZ", "sw-UG", "ko-KP", "khq", "khq-ML", "ses",
|
||
"ses-ML", "nmg", "nmq-CM", "ku-Arab-IR", "lkt", "lkt-US", "lag", "lag-TZ", "ln", "ln-AO",
|
||
"ln-CF", "ln-CD", "nds", "nds-DE", "nds-NL", "lu", "lu-CD", "luo", "luo", "luo-KE", "luy",
|
||
"luy-KE", "jmc", "jmc-TZ", "mgh", "mgh-MZ", "kde", "kde-TZ", "mg", "mg-MG", "gv", "gv-IM",
|
||
"mas", "mas-KE", "mas-TZ", "mas-IR", "mer", "mer-KE", "mgo", "mgo-CM", "mfe", "mfe-MU",
|
||
"mua", "mua-CM", "nqo", "nqo-GN", "nqa", "naq-NA", "nnh", "nnh-CM", "jgo", "jgo-CM",
|
||
"lrc-IQ", "lrc-IR", "nd", "nd-ZW", "nb-SJ", "nus", "nus-SD", "nus-SS", "nyn", "nyn-UG",
|
||
"om-KE", "os", "os-GE", "os-RU", "ps-PK", "fa-AF", "pt-AO", "pt-CV", "pt-GQ", "pt-GW",
|
||
"pt-LU", "pt-MO", "pt-MZ", "pt-ST", "pt-CH", "pt-TL", "prg-001", "ksh", "ksh-DE", "rof",
|
||
"rof-TZ", "rn", "rn-BI", "ru-BY", "ru-KZ", "ru-KG", "ru-UA", "rwk", "rwk-TZ", "ssy",
|
||
"ssy-ER", "saq", "saq-KE", "sg", "sq-CF", "sbp", "sbp-TZ", "seh", "seh-MZ", "ksb", "ksb-TZ",
|
||
"sn", "sn-Latn", "sn-Latn-ZW", "xog", "xog-UG", "so-DJ", "so-ET", "so-KE", "nr", "nr-ZA",
|
||
"st-LS", "es-BZ", "es-BR", "es-PH", "zgh", "zgh-Tfng-MA", "zgh-Tfng", "ss", "ss-ZA",
|
||
"ss-SZ", "sv-AX", "shi", "shi-Tfng", "shi-Tfng-MA", "shi-Latn", "shi-Latn-MA", "dav",
|
||
"dav-KE", "ta-MY", "ta-SG", "twq", "twq-NE", "teo", "teo-KE", "teo-UG", "bo-IN", "tig",
|
||
"tig-ER", "to", "to-TO", "tr-CY", "uz-Arab", "us-Arab-AF", "vai", "vai-Vaii",
|
||
"vai-Vaii-LR", "vai-Latn-LR", "vai-Latn", "vo", "vo-001", "vun", "vun-TZ", "wae",
|
||
"wae-CH", "wal", "wae-ET", "yav", "yav-CM", "yo-BJ", "dje", "dje-NE",
|
||
}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"C09": {tags: []string{"en-AU"}, localMonth: localMonthsNameEnglish, apFmt: strings.ToLower(nfp.AmPm[0])},
|
||
"2829": {tags: []string{"en-BZ"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"1009": {tags: []string{"en-CA"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"2409": {tags: []string{"en-029"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"3C09": {tags: []string{"en-HK"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"4009": {tags: []string{"en-IN"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"1809": {tags: []string{"en-IE"}, localMonth: localMonthsNameEnglish, apFmt: strings.ToLower(nfp.AmPm[0])},
|
||
"2009": {tags: []string{"en-JM"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"4409": {tags: []string{"en-MY"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"1409": {tags: []string{"en-NZ"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"3409": {tags: []string{"en-PH"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"4809": {tags: []string{"en-SG"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"1C09": {tags: []string{"en-ZA"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"2C09": {tags: []string{"en-TT"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"4C09": {tags: []string{"en-AE"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"809": {tags: []string{"en-GB"}, localMonth: localMonthsNameEnglish, apFmt: strings.ToLower(nfp.AmPm[0])},
|
||
"409": {tags: []string{"en-US"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"3009": {tags: []string{"en-ZW"}, localMonth: localMonthsNameEnglish, apFmt: nfp.AmPm[0]},
|
||
"C": {tags: []string{"fr"}, localMonth: localMonthsNameFrench, apFmt: nfp.AmPm[0]},
|
||
"7": {tags: []string{"de"}, localMonth: localMonthsNameGerman, apFmt: nfp.AmPm[0]},
|
||
"C07": {tags: []string{"de-AT"}, localMonth: localMonthsNameAustria, apFmt: nfp.AmPm[0]},
|
||
"407": {tags: []string{"de-DE"}, localMonth: localMonthsNameGerman, apFmt: nfp.AmPm[0]},
|
||
"3C": {tags: []string{"ga"}, localMonth: localMonthsNameIrish, apFmt: apFmtIrish},
|
||
"83C": {tags: []string{"ga-IE"}, localMonth: localMonthsNameIrish, apFmt: apFmtIrish},
|
||
"10": {tags: []string{"it"}, localMonth: localMonthsNameItalian, apFmt: nfp.AmPm[0]},
|
||
"11": {tags: []string{"ja"}, localMonth: localMonthsNameChinese3, apFmt: apFmtJapanese},
|
||
"411": {tags: []string{"ja-JP"}, localMonth: localMonthsNameChinese3, apFmt: apFmtJapanese},
|
||
"12": {tags: []string{"ko"}, localMonth: localMonthsNameKorean, apFmt: apFmtKorean},
|
||
"412": {tags: []string{"ko-KR"}, localMonth: localMonthsNameKorean, apFmt: apFmtKorean},
|
||
"7C50": {tags: []string{"mn-Mong"}, localMonth: localMonthsNameTraditionalMongolian, apFmt: nfp.AmPm[0]},
|
||
"850": {tags: []string{"mn-Mong-CN"}, localMonth: localMonthsNameTraditionalMongolian, apFmt: nfp.AmPm[0]},
|
||
"C50": {tags: []string{"mn-Mong-MN"}, localMonth: localMonthsNameTraditionalMongolian, apFmt: nfp.AmPm[0]},
|
||
"19": {tags: []string{"ru"}, localMonth: localMonthsNameRussian, apFmt: nfp.AmPm[0]},
|
||
"819": {tags: []string{"ru-MD"}, localMonth: localMonthsNameRussian, apFmt: nfp.AmPm[0]},
|
||
"419": {tags: []string{"ru-RU"}, localMonth: localMonthsNameRussian, apFmt: nfp.AmPm[0]},
|
||
"A": {tags: []string{"es"}, localMonth: localMonthsNameSpanish, apFmt: apFmtSpanish},
|
||
"2C0A": {tags: []string{"es-AR"}, localMonth: localMonthsNameSpanish, apFmt: apFmtSpanish},
|
||
"200A": {tags: []string{"es-VE"}, localMonth: localMonthsNameSpanish, apFmt: apFmtSpanish},
|
||
"400A": {tags: []string{"es-BO"}, localMonth: localMonthsNameSpanish, apFmt: apFmtSpanish},
|
||
"340A": {tags: []string{"es-CL"}, localMonth: localMonthsNameSpanish, apFmt: apFmtSpanish},
|
||
"240A": {tags: []string{"es-CO"}, localMonth: localMonthsNameSpanish, apFmt: apFmtSpanish},
|
||
"140A": {tags: []string{"es-CR"}, localMonth: localMonthsNameSpanish, apFmt: apFmtSpanish},
|
||
"5C0A": {tags: []string{"es-CU"}, localMonth: localMonthsNameSpanish, apFmt: apFmtCuba},
|
||
"1C0A": {tags: []string{"es-DO"}, localMonth: localMonthsNameSpanish, apFmt: apFmtSpanish},
|
||
"300A": {tags: []string{"es-EC"}, localMonth: localMonthsNameSpanish, apFmt: apFmtSpanish},
|
||
"440A": {tags: []string{"es-SV"}, localMonth: localMonthsNameSpanish, apFmt: apFmtSpanish},
|
||
"1E": {tags: []string{"th"}, localMonth: localMonthsNameThai, apFmt: nfp.AmPm[0]},
|
||
"41E": {tags: []string{"th-TH"}, localMonth: localMonthsNameThai, apFmt: nfp.AmPm[0]},
|
||
"51": {tags: []string{"bo"}, localMonth: localMonthsNameTibetan, apFmt: apFmtTibetan},
|
||
"451": {tags: []string{"bo-CN"}, localMonth: localMonthsNameTibetan, apFmt: apFmtTibetan},
|
||
"1F": {tags: []string{"tr"}, localMonth: localMonthsNameTurkish, apFmt: apFmtTurkish},
|
||
"41F": {tags: []string{"tr-TR"}, localMonth: localMonthsNameTurkish, apFmt: apFmtTurkish},
|
||
"52": {tags: []string{"cy"}, localMonth: localMonthsNameWelsh, apFmt: apFmtWelsh},
|
||
"452": {tags: []string{"cy-GB"}, localMonth: localMonthsNameWelsh, apFmt: apFmtWelsh},
|
||
"2A": {tags: []string{"vi"}, localMonth: localMonthsNameVietnamese, apFmt: apFmtVietnamese},
|
||
"42A": {tags: []string{"vi-VN"}, localMonth: localMonthsNameVietnamese, apFmt: apFmtVietnamese},
|
||
"88": {tags: []string{"wo"}, localMonth: localMonthsNameWolof, apFmt: apFmtWolof},
|
||
"488": {tags: []string{"wo-SN"}, localMonth: localMonthsNameWolof, apFmt: apFmtWolof},
|
||
"34": {tags: []string{"xh"}, localMonth: localMonthsNameXhosa, apFmt: nfp.AmPm[0]},
|
||
"434": {tags: []string{"xh-ZA"}, localMonth: localMonthsNameXhosa, apFmt: nfp.AmPm[0]},
|
||
"78": {tags: []string{"ii"}, localMonth: localMonthsNameYi, apFmt: apFmtYi},
|
||
"478": {tags: []string{"ii-CN"}, localMonth: localMonthsNameYi, apFmt: apFmtYi},
|
||
"35": {tags: []string{"zu"}, localMonth: localMonthsNameZulu, apFmt: nfp.AmPm[0]},
|
||
"435": {tags: []string{"zu-ZA"}, localMonth: localMonthsNameZulu, apFmt: nfp.AmPm[0]},
|
||
}
|
||
// monthNamesBangla list the month names in the Bangla.
|
||
monthNamesBangla = []string{
|
||
"\u099C\u09BE\u09A8\u09C1\u09AF\u09BC\u09BE\u09B0\u09C0",
|
||
"\u09AB\u09C7\u09AC\u09CD\u09B0\u09C1\u09AF\u09BC\u09BE\u09B0\u09C0",
|
||
"\u09AE\u09BE\u09B0\u09CD\u099A",
|
||
"\u098F\u09AA\u09CD\u09B0\u09BF\u09B2",
|
||
"\u09AE\u09C7",
|
||
"\u099C\u09C1\u09A8",
|
||
"\u099C\u09C1\u09B2\u09BE\u0987",
|
||
"\u0986\u0997\u09B8\u09CD\u099F",
|
||
"\u09B8\u09C7\u09AA\u09CD\u099F\u09C7\u09AE\u09CD\u09AC\u09B0",
|
||
"\u0985\u0995\u09CD\u099F\u09CB\u09AC\u09B0",
|
||
"\u09A8\u09AD\u09C7\u09AE\u09CD\u09AC\u09B0",
|
||
"\u09A1\u09BF\u09B8\u09C7\u09AE\u09CD\u09AC\u09B0",
|
||
}
|
||
// monthNamesAfrikaans list the month names in the Afrikaans.
|
||
monthNamesAfrikaans = []string{"Januarie", "Februarie", "Maart", "April", "Mei", "Junie", "Julie", "Augustus", "September", "Oktober", "November", "Desember"}
|
||
// monthNamesChinese list the month names in the Chinese.
|
||
monthNamesChinese = []string{"一", "二", "三", "四", "五", "六", "七", "八", "九", "十", "十一", "十二"}
|
||
// monthNamesFrench list the month names in the French.
|
||
monthNamesFrench = []string{"janvier", "février", "mars", "avril", "mai", "juin", "juillet", "août", "septembre", "octobre", "novembre", "décembre"}
|
||
// monthNamesGerman list the month names in the German.
|
||
monthNamesGerman = []string{"Januar", "Februar", "März", "April", "Mai", "Juni", "Juli", "August", "September", "Oktober", "November", "Dezember"}
|
||
// monthNamesAustria list the month names in the Austria.
|
||
monthNamesAustria = []string{"Jänner", "Februar", "März", "April", "Mai", "Juni", "Juli", "August", "September", "Oktober", "November", "Dezember"}
|
||
// monthNamesIrish list the month names in the Irish.
|
||
monthNamesIrish = []string{"Eanáir", "Feabhra", "Márta", "Aibreán", "Bealtaine", "Meitheamh", "Iúil", "Lúnasa", "Meán Fómhair", "Deireadh Fómhair", "Samhain", "Nollaig"}
|
||
// monthNamesItalian list the month names in the Italian.
|
||
monthNamesItalian = []string{"gennaio", "febbraio", "marzo", "aprile", "maggio", "giugno", "luglio", "agosto", "settembre", "ottobre", "novembre", "dicembre"}
|
||
// monthNamesRussian list the month names in the Russian.
|
||
monthNamesRussian = []string{"январь", "февраль", "март", "апрель", "май", "июнь", "июль", "август", "сентябрь", "октябрь", "ноябрь", "декабрь"}
|
||
// monthNamesSpanish list the month names in the Spanish.
|
||
monthNamesSpanish = []string{"enero", "febrero", "marzo", "abril", "mayo", "junio", "julio", "agosto", "septiembre", "octubre", "noviembre", "diciembre"}
|
||
// monthNamesThai list the month names in the Thai.
|
||
monthNamesThai = []string{
|
||
"\u0e21\u0e01\u0e23\u0e32\u0e04\u0e21",
|
||
"\u0e01\u0e38\u0e21\u0e20\u0e32\u0e1e\u0e31\u0e19\u0e18\u0e4c",
|
||
"\u0e21\u0e35\u0e19\u0e32\u0e04\u0e21",
|
||
"\u0e40\u0e21\u0e29\u0e32\u0e22\u0e19",
|
||
"\u0e1e\u0e24\u0e29\u0e20\u0e32\u0e04\u0e21",
|
||
"\u0e21\u0e34\u0e16\u0e38\u0e19\u0e32\u0e22\u0e19",
|
||
"\u0e01\u0e23\u0e01\u0e0e\u0e32\u0e04\u0e21",
|
||
"\u0e2a\u0e34\u0e07\u0e2b\u0e32\u0e04\u0e21",
|
||
"\u0e01\u0e31\u0e19\u0e22\u0e32\u0e22\u0e19",
|
||
"\u0e15\u0e38\u0e25\u0e32\u0e04\u0e21",
|
||
"\u0e1e\u0e24\u0e28\u0e08\u0e34\u0e01\u0e32\u0e22\u0e19",
|
||
"\u0e18\u0e31\u0e19\u0e27\u0e32\u0e04\u0e21",
|
||
}
|
||
// monthNamesTibetan list the month names in the Tibetan.
|
||
monthNamesTibetan = []string{
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f51\u0f44\u0f0b\u0f54\u0f7c\u0f0b",
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f42\u0f49\u0f72\u0f66\u0f0b\u0f54\u0f0b",
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f42\u0f66\u0f74\u0f58\u0f0b\u0f54\u0f0b",
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f56\u0f5e\u0f72\u0f0b\u0f54\u0f0b",
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f63\u0f94\u0f0b\u0f54\u0f0b",
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f51\u0fb2\u0f74\u0f42\u0f0b\u0f54\u0f0b",
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f56\u0f51\u0f74\u0f53\u0f0b\u0f54\u0f0b",
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f56\u0f62\u0f92\u0fb1\u0f51\u0f0b\u0f54\u0f0b",
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f51\u0f42\u0f74\u0f0b\u0f54\u0f0b",
|
||
"\u0f66\u0fa4\u0fb1\u0f72\u0f0b\u0f5f\u0fb3\u0f0b\u0f56\u0f45\u0f74\u0f0b\u0f54\u0f0d",
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f56\u0f45\u0f74\u0f0b\u0f42\u0f45\u0f72\u0f42\u0f0b\u0f54\u0f0b",
|
||
"\u0f5f\u0fb3\u0f0b\u0f56\u0f0b\u0f56\u0f45\u0f74\u0f0b\u0f42\u0f49\u0f72\u0f66\u0f0b\u0f54\u0f0b",
|
||
}
|
||
// monthNamesTurkish list the month names in the Turkish.
|
||
monthNamesTurkish = []string{"Ocak", "Şubat", "Mart", "Nisan", "Mayıs", "Haziran", "Temmuz", "Ağustos", "Eylül", "Ekim", "Kasım", "Aralık"}
|
||
// monthNamesWelsh list the month names in the Welsh.
|
||
monthNamesWelsh = []string{"Ionawr", "Chwefror", "Mawrth", "Ebrill", "Mai", "Mehefin", "Gorffennaf", "Awst", "Medi", "Hydref", "Tachwedd", "Rhagfyr"}
|
||
// monthNamesWolof list the month names in the Wolof.
|
||
monthNamesWolof = []string{"Samwiye", "Fewriye", "Maars", "Awril", "Me", "Suwe", "Sullet", "Ut", "Septàmbar", "Oktoobar", "Noowàmbar", "Desàmbar"}
|
||
// monthNamesXhosa list the month names in the Xhosa.
|
||
monthNamesXhosa = []string{"Januwari", "Febuwari", "Matshi", "Aprili", "Meyi", "Juni", "Julayi", "Agasti", "Septemba", "Oktobha", "Novemba", "Disemba"}
|
||
// monthNamesYi list the month names in the Yi.
|
||
monthNamesYi = []string{"\ua2cd", "\ua44d", "\ua315", "\ua1d6", "\ua26c", "\ua0d8", "\ua3c3", "\ua246", "\ua22c", "\ua2b0", "\ua2b0\ua2aa", "\ua2b0\ua44b"}
|
||
// monthNamesZulu list the month names in the Zulu.
|
||
monthNamesZulu = []string{"Januwari", "Febhuwari", "Mashi", "Ephreli", "Meyi", "Juni", "Julayi", "Agasti", "Septemba", "Okthoba", "Novemba", "Disemba"}
|
||
// apFmtAfrikaans defined the AM/PM name in the Afrikaans.
|
||
apFmtAfrikaans = "vm./nm."
|
||
// apFmtCuba defined the AM/PM name in the Cuba.
|
||
apFmtCuba = "a.m./p.m."
|
||
// apFmtIrish defined the AM/PM name in the Irish.
|
||
apFmtIrish = "r.n./i.n."
|
||
// apFmtJapanese defined the AM/PM name in the Japanese.
|
||
apFmtJapanese = "午前/午後"
|
||
// apFmtKorean defined the AM/PM name in the Korean.
|
||
apFmtKorean = "오전/오후"
|
||
// apFmtSpanish defined the AM/PM name in the Spanish.
|
||
apFmtSpanish = "a. m./p. m."
|
||
// apFmtTibetan defined the AM/PM name in the Tibetan.
|
||
apFmtTibetan = "\u0f66\u0f94\u0f0b\u0f51\u0fb2\u0f7c\u0f0b/\u0f55\u0fb1\u0f72\u0f0b\u0f51\u0fb2\u0f7c\u0f0b"
|
||
// apFmtTurkish defined the AM/PM name in the Turkish.
|
||
apFmtTurkish = "\u00F6\u00F6/\u00F6\u0053"
|
||
// apFmtVietnamese defined the AM/PM name in the Vietnamese.
|
||
apFmtVietnamese = "SA/CH"
|
||
// apFmtWolof defined the AM/PM name in the Wolof.
|
||
apFmtWolof = "Sub/Ngo"
|
||
// apFmtYi defined the AM/PM name in the Yi.
|
||
apFmtYi = "\ua3b8\ua111/\ua06f\ua2d2"
|
||
// apFmtWelsh defined the AM/PM name in the Welsh.
|
||
apFmtWelsh = "yb/yh"
|
||
)
|
||
|
||
// prepareNumberic split the number into two before and after parts by a
|
||
// decimal point.
|
||
func (nf *numberFormat) prepareNumberic(value string) {
|
||
if nf.isNumeric, _ = isNumeric(value); !nf.isNumeric {
|
||
return
|
||
}
|
||
}
|
||
|
||
// format provides a function to return a string parse by number format
|
||
// expression. If the given number format is not supported, this will return
|
||
// the original cell value.
|
||
func format(value, numFmt string, date1904 bool) string {
|
||
p := nfp.NumberFormatParser()
|
||
nf := numberFormat{section: p.Parse(numFmt), value: value, date1904: date1904}
|
||
nf.number, nf.valueSectionType = nf.getValueSectionType(value)
|
||
nf.prepareNumberic(value)
|
||
for i, section := range nf.section {
|
||
nf.sectionIdx = i
|
||
if section.Type != nf.valueSectionType {
|
||
continue
|
||
}
|
||
if nf.isNumeric {
|
||
switch section.Type {
|
||
case nfp.TokenSectionPositive:
|
||
return nf.positiveHandler()
|
||
case nfp.TokenSectionNegative:
|
||
return nf.negativeHandler()
|
||
default:
|
||
return nf.zeroHandler()
|
||
}
|
||
}
|
||
return nf.textHandler()
|
||
}
|
||
return value
|
||
}
|
||
|
||
// positiveHandler will be handling positive selection for a number format
|
||
// expression.
|
||
func (nf *numberFormat) positiveHandler() (result string) {
|
||
nf.t, nf.hours, nf.seconds = timeFromExcelTime(nf.number, nf.date1904), false, false
|
||
for i, token := range nf.section[nf.sectionIdx].Items {
|
||
if inStrSlice(supportedTokenTypes, token.TType, true) == -1 || token.TType == nfp.TokenTypeGeneral {
|
||
result = nf.value
|
||
return
|
||
}
|
||
if token.TType == nfp.TokenTypeCurrencyLanguage {
|
||
if err := nf.currencyLanguageHandler(i, token); err != nil {
|
||
result = nf.value
|
||
return
|
||
}
|
||
}
|
||
if token.TType == nfp.TokenTypeDateTimes {
|
||
nf.dateTimesHandler(i, token)
|
||
}
|
||
if token.TType == nfp.TokenTypeElapsedDateTimes {
|
||
nf.elapsedDateTimesHandler(token)
|
||
}
|
||
if token.TType == nfp.TokenTypeLiteral {
|
||
nf.result += token.TValue
|
||
continue
|
||
}
|
||
if token.TType == nfp.TokenTypeZeroPlaceHolder && token.TValue == strings.Repeat("0", len(token.TValue)) {
|
||
if isNum, precision := isNumeric(nf.value); isNum {
|
||
if nf.number < 1 {
|
||
nf.result += "0"
|
||
continue
|
||
}
|
||
if precision > 15 {
|
||
nf.result += roundPrecision(nf.value, 15)
|
||
} else {
|
||
nf.result += fmt.Sprintf("%.f", nf.number)
|
||
}
|
||
continue
|
||
}
|
||
}
|
||
}
|
||
result = nf.result
|
||
return
|
||
}
|
||
|
||
// currencyLanguageHandler will be handling currency and language types tokens for a number
|
||
// format expression.
|
||
func (nf *numberFormat) currencyLanguageHandler(i int, token nfp.Token) (err error) {
|
||
for _, part := range token.Parts {
|
||
if inStrSlice(supportedTokenTypes, part.Token.TType, true) == -1 {
|
||
err = ErrUnsupportedNumberFormat
|
||
return
|
||
}
|
||
if _, ok := supportedLanguageInfo[strings.ToUpper(part.Token.TValue)]; !ok {
|
||
err = ErrUnsupportedNumberFormat
|
||
return
|
||
}
|
||
nf.localCode = strings.ToUpper(part.Token.TValue)
|
||
}
|
||
return
|
||
}
|
||
|
||
// localAmPm return AM/PM name by supported language ID.
|
||
func (nf *numberFormat) localAmPm(ap string) string {
|
||
if languageInfo, ok := supportedLanguageInfo[nf.localCode]; ok {
|
||
return languageInfo.apFmt
|
||
}
|
||
return ap
|
||
}
|
||
|
||
// localMonthsNameEnglish returns the English name of the month.
|
||
func localMonthsNameEnglish(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
return t.Month().String()[:3]
|
||
}
|
||
if abbr == 4 {
|
||
return t.Month().String()
|
||
}
|
||
return t.Month().String()[:1]
|
||
}
|
||
|
||
// localMonthsNameAfrikaans returns the Afrikaans name of the month.
|
||
func localMonthsNameAfrikaans(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
month := monthNamesAfrikaans[int(t.Month())-1]
|
||
if len([]rune(month)) <= 3 {
|
||
return month
|
||
}
|
||
return string([]rune(month)[:3]) + "."
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesAfrikaans[int(t.Month())-1]
|
||
}
|
||
return monthNamesAfrikaans[int(t.Month())-1][:1]
|
||
}
|
||
|
||
// localMonthsNameAustria returns the Austria name of the month.
|
||
func localMonthsNameAustria(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
return string([]rune(monthNamesAustria[int(t.Month())-1])[:3])
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesAustria[int(t.Month())-1]
|
||
}
|
||
return monthNamesAustria[int(t.Month())-1][:1]
|
||
}
|
||
|
||
// localMonthsNameBangla returns the German name of the month.
|
||
func localMonthsNameBangla(t time.Time, abbr int) string {
|
||
if abbr == 3 || abbr == 4 {
|
||
return monthNamesBangla[int(t.Month())-1]
|
||
}
|
||
return string([]rune(monthNamesBangla[int(t.Month())-1])[:1])
|
||
}
|
||
|
||
// localMonthsNameFrench returns the French name of the month.
|
||
func localMonthsNameFrench(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
month := monthNamesFrench[int(t.Month())-1]
|
||
if len([]rune(month)) <= 4 {
|
||
return month
|
||
}
|
||
return string([]rune(month)[:4]) + "."
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesFrench[int(t.Month())-1]
|
||
}
|
||
return monthNamesFrench[int(t.Month())-1][:1]
|
||
}
|
||
|
||
// localMonthsNameIrish returns the Irish name of the month.
|
||
func localMonthsNameIrish(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
switch int(t.Month()) {
|
||
case 1, 4, 8:
|
||
return string([]rune(monthNamesIrish[int(t.Month())-1])[:3])
|
||
case 2, 3, 6:
|
||
return string([]rune(monthNamesIrish[int(t.Month())-1])[:5])
|
||
case 9, 10:
|
||
return string([]rune(monthNamesIrish[int(t.Month())-1])[:1]) + "Fómh"
|
||
default:
|
||
return string([]rune(monthNamesIrish[int(t.Month())-1])[:4])
|
||
}
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesIrish[int(t.Month())-1]
|
||
}
|
||
return string([]rune(monthNamesIrish[int(t.Month())-1])[:1])
|
||
}
|
||
|
||
// localMonthsNameItalian returns the Italian name of the month.
|
||
func localMonthsNameItalian(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
return monthNamesItalian[int(t.Month())-1][:3]
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesItalian[int(t.Month())-1]
|
||
}
|
||
return monthNamesItalian[int(t.Month())-1][:1]
|
||
}
|
||
|
||
// localMonthsNameGerman returns the German name of the month.
|
||
func localMonthsNameGerman(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
return string([]rune(monthNamesGerman[int(t.Month())-1])[:3])
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesGerman[int(t.Month())-1]
|
||
}
|
||
return string([]rune(monthNamesGerman[int(t.Month())-1])[:1])
|
||
}
|
||
|
||
// localMonthsNameChinese1 returns the Chinese name of the month.
|
||
func localMonthsNameChinese1(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
return strconv.Itoa(int(t.Month())) + "月"
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesChinese[int(t.Month())-1] + "月"
|
||
}
|
||
return monthNamesChinese[int(t.Month())-1]
|
||
}
|
||
|
||
// localMonthsNameChinese2 returns the Chinese name of the month.
|
||
func localMonthsNameChinese2(t time.Time, abbr int) string {
|
||
if abbr == 3 || abbr == 4 {
|
||
return monthNamesChinese[int(t.Month())-1] + "月"
|
||
}
|
||
return monthNamesChinese[int(t.Month())-1]
|
||
}
|
||
|
||
// localMonthsNameChinese3 returns the Chinese name of the month.
|
||
func localMonthsNameChinese3(t time.Time, abbr int) string {
|
||
if abbr == 3 || abbr == 4 {
|
||
return strconv.Itoa(int(t.Month())) + "月"
|
||
}
|
||
return strconv.Itoa(int(t.Month()))
|
||
}
|
||
|
||
// localMonthsNameKorean returns the Korean name of the month.
|
||
func localMonthsNameKorean(t time.Time, abbr int) string {
|
||
if abbr == 3 || abbr == 4 {
|
||
return strconv.Itoa(int(t.Month())) + "월"
|
||
}
|
||
return strconv.Itoa(int(t.Month()))
|
||
}
|
||
|
||
// localMonthsNameTraditionalMongolian returns the Traditional Mongolian name of the month.
|
||
func localMonthsNameTraditionalMongolian(t time.Time, abbr int) string {
|
||
if abbr == 5 {
|
||
return "M"
|
||
}
|
||
return fmt.Sprintf("M%02d", int(t.Month()))
|
||
}
|
||
|
||
// localMonthsNameRussian returns the Russian name of the month.
|
||
func localMonthsNameRussian(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
month := monthNamesRussian[int(t.Month())-1]
|
||
if len([]rune(month)) <= 4 {
|
||
return month
|
||
}
|
||
return string([]rune(month)[:3]) + "."
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesRussian[int(t.Month())-1]
|
||
}
|
||
return string([]rune(monthNamesRussian[int(t.Month())-1])[:1])
|
||
}
|
||
|
||
// localMonthsNameSpanish returns the Spanish name of the month.
|
||
func localMonthsNameSpanish(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
return monthNamesSpanish[int(t.Month())-1][:3]
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesSpanish[int(t.Month())-1]
|
||
}
|
||
return monthNamesSpanish[int(t.Month())-1][:1]
|
||
}
|
||
|
||
// localMonthsNameThai returns the Thai name of the month.
|
||
func localMonthsNameThai(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
r := []rune(monthNamesThai[int(t.Month())-1])
|
||
return string(r[:1]) + "." + string(r[len(r)-2:len(r)-1]) + "."
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesThai[int(t.Month())-1]
|
||
}
|
||
return string([]rune(monthNamesThai[int(t.Month())-1])[:1])
|
||
}
|
||
|
||
// localMonthsNameTibetan returns the Tibetan name of the month.
|
||
func localMonthsNameTibetan(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
return "\u0f5f\u0fb3\u0f0b" + []string{"\u0f21", "\u0f22", "\u0f23", "\u0f24", "\u0f25", "\u0f26", "\u0f27", "\u0f28", "\u0f29", "\u0f21\u0f20", "\u0f21\u0f21", "\u0f21\u0f22"}[int(t.Month())-1]
|
||
}
|
||
if abbr == 5 {
|
||
if t.Month() == 10 {
|
||
return "\u0f66"
|
||
}
|
||
return "\u0f5f"
|
||
}
|
||
return monthNamesTibetan[int(t.Month())-1]
|
||
}
|
||
|
||
// localMonthsNameTurkish returns the Turkish name of the month.
|
||
func localMonthsNameTurkish(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
return string([]rune(monthNamesTurkish[int(t.Month())-1])[:3])
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesTurkish[int(t.Month())-1]
|
||
}
|
||
return string([]rune(monthNamesTurkish[int(t.Month())-1])[:1])
|
||
}
|
||
|
||
// localMonthsNameWelsh returns the Welsh name of the month.
|
||
func localMonthsNameWelsh(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
switch int(t.Month()) {
|
||
case 2, 7:
|
||
return string([]rune(monthNamesWelsh[int(t.Month())-1])[:5])
|
||
case 8, 9, 11, 12:
|
||
return string([]rune(monthNamesWelsh[int(t.Month())-1])[:4])
|
||
default:
|
||
return string([]rune(monthNamesWelsh[int(t.Month())-1])[:3])
|
||
}
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesWelsh[int(t.Month())-1]
|
||
}
|
||
return string([]rune(monthNamesWelsh[int(t.Month())-1])[:1])
|
||
}
|
||
|
||
// localMonthsNameVietnamese returns the Vietnamese name of the month.
|
||
func localMonthsNameVietnamese(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
return "Thg " + strconv.Itoa(int(t.Month()))
|
||
}
|
||
if abbr == 5 {
|
||
return "T " + strconv.Itoa(int(t.Month()))
|
||
}
|
||
return "Tháng " + strconv.Itoa(int(t.Month()))
|
||
}
|
||
|
||
// localMonthsNameWolof returns the Wolof name of the month.
|
||
func localMonthsNameWolof(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
switch int(t.Month()) {
|
||
case 3, 6:
|
||
return string([]rune(monthNamesWolof[int(t.Month())-1])[:3])
|
||
case 5, 8:
|
||
return string([]rune(monthNamesWolof[int(t.Month())-1])[:2])
|
||
case 9:
|
||
return string([]rune(monthNamesWolof[int(t.Month())-1])[:4]) + "."
|
||
case 11:
|
||
return "Now."
|
||
default:
|
||
return string([]rune(monthNamesWolof[int(t.Month())-1])[:3]) + "."
|
||
}
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesWolof[int(t.Month())-1]
|
||
}
|
||
return string([]rune(monthNamesWolof[int(t.Month())-1])[:1])
|
||
}
|
||
|
||
// localMonthsNameXhosa returns the Xhosa name of the month.
|
||
func localMonthsNameXhosa(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
switch int(t.Month()) {
|
||
case 4:
|
||
return "uEpr."
|
||
case 8:
|
||
return "u" + string([]rune(monthNamesXhosa[int(t.Month())-1])[:2]) + "."
|
||
default:
|
||
return "u" + string([]rune(monthNamesXhosa[int(t.Month())-1])[:3]) + "."
|
||
}
|
||
}
|
||
if abbr == 4 {
|
||
return "u" + monthNamesXhosa[int(t.Month())-1]
|
||
}
|
||
return "u"
|
||
}
|
||
|
||
// localMonthsNameYi returns the Yi name of the month.
|
||
func localMonthsNameYi(t time.Time, abbr int) string {
|
||
if abbr == 3 || abbr == 4 {
|
||
return string(monthNamesYi[int(t.Month())-1]) + "\ua1aa"
|
||
}
|
||
return string([]rune(monthNamesYi[int(t.Month())-1])[:1])
|
||
}
|
||
|
||
// localMonthsNameZulu returns the Zulu name of the month.
|
||
func localMonthsNameZulu(t time.Time, abbr int) string {
|
||
if abbr == 3 {
|
||
if int(t.Month()) == 8 {
|
||
return string([]rune(monthNamesZulu[int(t.Month())-1])[:4])
|
||
}
|
||
return string([]rune(monthNamesZulu[int(t.Month())-1])[:3])
|
||
}
|
||
if abbr == 4 {
|
||
return monthNamesZulu[int(t.Month())-1]
|
||
}
|
||
return string([]rune(monthNamesZulu[int(t.Month())-1])[:1])
|
||
}
|
||
|
||
// localMonthName return months name by supported language ID.
|
||
func (nf *numberFormat) localMonthsName(abbr int) string {
|
||
if languageInfo, ok := supportedLanguageInfo[nf.localCode]; ok {
|
||
return languageInfo.localMonth(nf.t, abbr)
|
||
}
|
||
return localMonthsNameEnglish(nf.t, abbr)
|
||
}
|
||
|
||
// dateTimesHandler will be handling date and times types tokens for a number
|
||
// format expression.
|
||
func (nf *numberFormat) dateTimesHandler(i int, token nfp.Token) {
|
||
if idx := inStrSlice(nfp.AmPm, strings.ToUpper(token.TValue), false); idx != -1 {
|
||
if nf.ap == "" {
|
||
nextHours := nf.hoursNext(i)
|
||
aps := strings.Split(nf.localAmPm(token.TValue), "/")
|
||
nf.ap = aps[0]
|
||
if nextHours > 12 {
|
||
nf.ap = aps[1]
|
||
}
|
||
}
|
||
nf.result += nf.ap
|
||
return
|
||
}
|
||
if strings.Contains(strings.ToUpper(token.TValue), "M") {
|
||
l := len(token.TValue)
|
||
if l == 1 && !nf.hours && !nf.secondsNext(i) {
|
||
nf.result += strconv.Itoa(int(nf.t.Month()))
|
||
return
|
||
}
|
||
if l == 2 && !nf.hours && !nf.secondsNext(i) {
|
||
nf.result += fmt.Sprintf("%02d", int(nf.t.Month()))
|
||
return
|
||
}
|
||
if l == 3 {
|
||
nf.result += nf.localMonthsName(3)
|
||
return
|
||
}
|
||
if l == 4 || l > 5 {
|
||
nf.result += nf.localMonthsName(4)
|
||
return
|
||
}
|
||
if l == 5 {
|
||
nf.result += nf.localMonthsName(5)
|
||
return
|
||
}
|
||
}
|
||
nf.yearsHandler(i, token)
|
||
nf.daysHandler(i, token)
|
||
nf.hoursHandler(i, token)
|
||
nf.minutesHandler(token)
|
||
nf.secondsHandler(token)
|
||
}
|
||
|
||
// yearsHandler will be handling years in the date and times types tokens for a
|
||
// number format expression.
|
||
func (nf *numberFormat) yearsHandler(i int, token nfp.Token) {
|
||
years := strings.Contains(strings.ToUpper(token.TValue), "Y")
|
||
if years && len(token.TValue) <= 2 {
|
||
nf.result += strconv.Itoa(nf.t.Year())[2:]
|
||
return
|
||
}
|
||
if years && len(token.TValue) > 2 {
|
||
nf.result += strconv.Itoa(nf.t.Year())
|
||
return
|
||
}
|
||
}
|
||
|
||
// daysHandler will be handling days in the date and times types tokens for a
|
||
// number format expression.
|
||
func (nf *numberFormat) daysHandler(i int, token nfp.Token) {
|
||
if strings.Contains(strings.ToUpper(token.TValue), "D") {
|
||
switch len(token.TValue) {
|
||
case 1:
|
||
nf.result += strconv.Itoa(nf.t.Day())
|
||
return
|
||
case 2:
|
||
nf.result += fmt.Sprintf("%02d", nf.t.Day())
|
||
return
|
||
case 3:
|
||
nf.result += nf.t.Weekday().String()[:3]
|
||
return
|
||
default:
|
||
nf.result += nf.t.Weekday().String()
|
||
return
|
||
}
|
||
}
|
||
}
|
||
|
||
// hoursHandler will be handling hours in the date and times types tokens for a
|
||
// number format expression.
|
||
func (nf *numberFormat) hoursHandler(i int, token nfp.Token) {
|
||
nf.hours = strings.Contains(strings.ToUpper(token.TValue), "H")
|
||
if nf.hours {
|
||
h := nf.t.Hour()
|
||
ap, ok := nf.apNext(i)
|
||
if ok {
|
||
nf.ap = ap[0]
|
||
if h > 12 {
|
||
h -= 12
|
||
nf.ap = ap[1]
|
||
}
|
||
}
|
||
if nf.ap != "" && nf.hoursNext(i) == -1 && h > 12 {
|
||
h -= 12
|
||
}
|
||
switch len(token.TValue) {
|
||
case 1:
|
||
nf.result += strconv.Itoa(h)
|
||
return
|
||
default:
|
||
nf.result += fmt.Sprintf("%02d", h)
|
||
return
|
||
}
|
||
}
|
||
}
|
||
|
||
// minutesHandler will be handling minutes in the date and times types tokens
|
||
// for a number format expression.
|
||
func (nf *numberFormat) minutesHandler(token nfp.Token) {
|
||
if strings.Contains(strings.ToUpper(token.TValue), "M") {
|
||
nf.hours = false
|
||
switch len(token.TValue) {
|
||
case 1:
|
||
nf.result += strconv.Itoa(nf.t.Minute())
|
||
return
|
||
default:
|
||
nf.result += fmt.Sprintf("%02d", nf.t.Minute())
|
||
return
|
||
}
|
||
}
|
||
}
|
||
|
||
// secondsHandler will be handling seconds in the date and times types tokens
|
||
// for a number format expression.
|
||
func (nf *numberFormat) secondsHandler(token nfp.Token) {
|
||
nf.seconds = strings.Contains(strings.ToUpper(token.TValue), "S")
|
||
if nf.seconds {
|
||
switch len(token.TValue) {
|
||
case 1:
|
||
nf.result += strconv.Itoa(nf.t.Second())
|
||
return
|
||
default:
|
||
nf.result += fmt.Sprintf("%02d", nf.t.Second())
|
||
return
|
||
}
|
||
}
|
||
}
|
||
|
||
// elapsedDateTimesHandler will be handling elapsed date and times types tokens
|
||
// for a number format expression.
|
||
func (nf *numberFormat) elapsedDateTimesHandler(token nfp.Token) {
|
||
if strings.Contains(strings.ToUpper(token.TValue), "H") {
|
||
nf.result += fmt.Sprintf("%.f", nf.t.Sub(excel1900Epoc).Hours())
|
||
return
|
||
}
|
||
if strings.Contains(strings.ToUpper(token.TValue), "M") {
|
||
nf.result += fmt.Sprintf("%.f", nf.t.Sub(excel1900Epoc).Minutes())
|
||
return
|
||
}
|
||
if strings.Contains(strings.ToUpper(token.TValue), "S") {
|
||
nf.result += fmt.Sprintf("%.f", nf.t.Sub(excel1900Epoc).Seconds())
|
||
return
|
||
}
|
||
}
|
||
|
||
// hoursNext detects if a token of type hours exists after a given tokens list.
|
||
func (nf *numberFormat) hoursNext(i int) int {
|
||
tokens := nf.section[nf.sectionIdx].Items
|
||
for idx := i + 1; idx < len(tokens); idx++ {
|
||
if tokens[idx].TType == nfp.TokenTypeDateTimes {
|
||
if strings.Contains(strings.ToUpper(tokens[idx].TValue), "H") {
|
||
t := timeFromExcelTime(nf.number, false)
|
||
return t.Hour()
|
||
}
|
||
}
|
||
}
|
||
return -1
|
||
}
|
||
|
||
// apNext detects if a token of type AM/PM exists after a given tokens list.
|
||
func (nf *numberFormat) apNext(i int) ([]string, bool) {
|
||
tokens := nf.section[nf.sectionIdx].Items
|
||
for idx := i + 1; idx < len(tokens); idx++ {
|
||
if tokens[idx].TType == nfp.TokenTypeDateTimes {
|
||
if strings.Contains(strings.ToUpper(tokens[idx].TValue), "H") {
|
||
return nil, false
|
||
}
|
||
if i := inStrSlice(nfp.AmPm, tokens[idx].TValue, false); i != -1 {
|
||
return strings.Split(nf.localAmPm(tokens[idx].TValue), "/"), true
|
||
}
|
||
}
|
||
}
|
||
return nil, false
|
||
}
|
||
|
||
// secondsNext detects if a token of type seconds exists after a given tokens
|
||
// list.
|
||
func (nf *numberFormat) secondsNext(i int) bool {
|
||
tokens := nf.section[nf.sectionIdx].Items
|
||
for idx := i + 1; idx < len(tokens); idx++ {
|
||
if tokens[idx].TType == nfp.TokenTypeDateTimes {
|
||
return strings.Contains(strings.ToUpper(tokens[idx].TValue), "S")
|
||
}
|
||
}
|
||
return false
|
||
}
|
||
|
||
// negativeHandler will be handling negative selection for a number format
|
||
// expression.
|
||
func (nf *numberFormat) negativeHandler() (result string) {
|
||
for _, token := range nf.section[nf.sectionIdx].Items {
|
||
if inStrSlice(supportedTokenTypes, token.TType, true) == -1 || token.TType == nfp.TokenTypeGeneral {
|
||
result = nf.value
|
||
return
|
||
}
|
||
if token.TType == nfp.TokenTypeLiteral {
|
||
nf.result += token.TValue
|
||
continue
|
||
}
|
||
if token.TType == nfp.TokenTypeZeroPlaceHolder && token.TValue == strings.Repeat("0", len(token.TValue)) {
|
||
if isNum, precision := isNumeric(nf.value); isNum {
|
||
if math.Abs(nf.number) < 1 {
|
||
nf.result += "0"
|
||
continue
|
||
}
|
||
if precision > 15 {
|
||
nf.result += strings.TrimLeft(roundPrecision(nf.value, 15), "-")
|
||
} else {
|
||
nf.result += fmt.Sprintf("%.f", math.Abs(nf.number))
|
||
}
|
||
continue
|
||
}
|
||
}
|
||
}
|
||
result = nf.result
|
||
return
|
||
}
|
||
|
||
// zeroHandler will be handling zero selection for a number format expression.
|
||
func (nf *numberFormat) zeroHandler() string {
|
||
return nf.value
|
||
}
|
||
|
||
// textHandler will be handling text selection for a number format expression.
|
||
func (nf *numberFormat) textHandler() (result string) {
|
||
for _, token := range nf.section[nf.sectionIdx].Items {
|
||
if token.TType == nfp.TokenTypeLiteral {
|
||
result += token.TValue
|
||
}
|
||
if token.TType == nfp.TokenTypeTextPlaceHolder {
|
||
result += nf.value
|
||
}
|
||
}
|
||
return result
|
||
}
|
||
|
||
// getValueSectionType returns its applicable number format expression section
|
||
// based on the given value.
|
||
func (nf *numberFormat) getValueSectionType(value string) (float64, string) {
|
||
isNum, _ := isNumeric(value)
|
||
if !isNum {
|
||
return 0, nfp.TokenSectionText
|
||
}
|
||
number, _ := strconv.ParseFloat(value, 64)
|
||
if number > 0 {
|
||
return number, nfp.TokenSectionPositive
|
||
}
|
||
if number < 0 {
|
||
return number, nfp.TokenSectionNegative
|
||
}
|
||
return number, nfp.TokenSectionZero
|
||
}
|