当前仓库属于暂停状态,部分功能使用受限,详情请查阅 仓库状态说明
1 Star 0 Fork 151

chewel / excelize
暂停

forked from xuri / excelize 
加入 Gitee
与超过 1200万 开发者一起发现、参与优秀开源项目,私有仓库也完全免费 :)
免费加入
克隆/下载
sheet.go 58.18 KB
一键复制 编辑 原始数据 按行查看 历史
1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465666768697071727374757677787980818283848586878889909192939495969798991001011021031041051061071081091101111121131141151161171181191201211221231241251261271281291301311321331341351361371381391401411421431441451461471481491501511521531541551561571581591601611621631641651661671681691701711721731741751761771781791801811821831841851861871881891901911921931941951961971981992002012022032042052062072082092102112122132142152162172182192202212222232242252262272282292302312322332342352362372382392402412422432442452462472482492502512522532542552562572582592602612622632642652662672682692702712722732742752762772782792802812822832842852862872882892902912922932942952962972982993003013023033043053063073083093103113123133143153163173183193203213223233243253263273283293303313323333343353363373383393403413423433443453463473483493503513523533543553563573583593603613623633643653663673683693703713723733743753763773783793803813823833843853863873883893903913923933943953963973983994004014024034044054064074084094104114124134144154164174184194204214224234244254264274284294304314324334344354364374384394404414424434444454464474484494504514524534544554564574584594604614624634644654664674684694704714724734744754764774784794804814824834844854864874884894904914924934944954964974984995005015025035045055065075085095105115125135145155165175185195205215225235245255265275285295305315325335345355365375385395405415425435445455465475485495505515525535545555565575585595605615625635645655665675685695705715725735745755765775785795805815825835845855865875885895905915925935945955965975985996006016026036046056066076086096106116126136146156166176186196206216226236246256266276286296306316326336346356366376386396406416426436446456466476486496506516526536546556566576586596606616626636646656666676686696706716726736746756766776786796806816826836846856866876886896906916926936946956966976986997007017027037047057067077087097107117127137147157167177187197207217227237247257267277287297307317327337347357367377387397407417427437447457467477487497507517527537547557567577587597607617627637647657667677687697707717727737747757767777787797807817827837847857867877887897907917927937947957967977987998008018028038048058068078088098108118128138148158168178188198208218228238248258268278288298308318328338348358368378388398408418428438448458468478488498508518528538548558568578588598608618628638648658668678688698708718728738748758768778788798808818828838848858868878888898908918928938948958968978988999009019029039049059069079089099109119129139149159169179189199209219229239249259269279289299309319329339349359369379389399409419429439449459469479489499509519529539549559569579589599609619629639649659669679689699709719729739749759769779789799809819829839849859869879889899909919929939949959969979989991000100110021003100410051006100710081009101010111012101310141015101610171018101910201021102210231024102510261027102810291030103110321033103410351036103710381039104010411042104310441045104610471048104910501051105210531054105510561057105810591060106110621063106410651066106710681069107010711072107310741075107610771078107910801081108210831084108510861087108810891090109110921093109410951096109710981099110011011102110311041105110611071108110911101111111211131114111511161117111811191120112111221123112411251126112711281129113011311132113311341135113611371138113911401141114211431144114511461147114811491150115111521153115411551156115711581159116011611162116311641165116611671168116911701171117211731174117511761177117811791180118111821183118411851186118711881189119011911192119311941195119611971198119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231123212331234123512361237123812391240124112421243124412451246124712481249125012511252125312541255125612571258125912601261126212631264126512661267126812691270127112721273127412751276127712781279128012811282128312841285128612871288128912901291129212931294129512961297129812991300130113021303130413051306130713081309131013111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336133713381339134013411342134313441345134613471348134913501351135213531354135513561357135813591360136113621363136413651366136713681369137013711372137313741375137613771378137913801381138213831384138513861387138813891390139113921393139413951396139713981399140014011402140314041405140614071408140914101411141214131414141514161417141814191420142114221423142414251426142714281429143014311432143314341435143614371438143914401441144214431444144514461447144814491450145114521453145414551456145714581459146014611462146314641465146614671468146914701471147214731474147514761477147814791480148114821483148414851486148714881489149014911492149314941495149614971498149915001501150215031504150515061507150815091510151115121513151415151516151715181519152015211522152315241525152615271528152915301531153215331534153515361537153815391540154115421543154415451546154715481549155015511552155315541555155615571558155915601561156215631564156515661567156815691570157115721573157415751576157715781579158015811582158315841585158615871588158915901591159215931594159515961597159815991600160116021603160416051606160716081609161016111612161316141615161616171618161916201621162216231624162516261627162816291630163116321633163416351636163716381639164016411642164316441645164616471648164916501651165216531654165516561657165816591660166116621663166416651666166716681669167016711672167316741675167616771678167916801681168216831684168516861687168816891690169116921693169416951696169716981699170017011702170317041705170617071708170917101711171217131714171517161717171817191720172117221723172417251726172717281729173017311732173317341735173617371738173917401741174217431744174517461747174817491750175117521753175417551756175717581759176017611762176317641765176617671768176917701771177217731774177517761777177817791780178117821783178417851786178717881789
// Copyright 2016 - 2021 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 XLSX / XLSM / XLTM files. Supports reading and writing
// spreadsheet documents generated by Microsoft Exce™ 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 1.10 or later.
package excelize
import (
"bytes"
"encoding/json"
"encoding/xml"
"errors"
"fmt"
"io"
"io/ioutil"
"log"
"os"
"path"
"path/filepath"
"reflect"
"regexp"
"strconv"
"strings"
"unicode/utf8"
"github.com/mohae/deepcopy"
)
// NewSheet provides the function to create a new sheet by given a worksheet
// name and returns the index of the sheets in the workbook
// (spreadsheet) after it appended. Note that when creating a new spreadsheet
// file, the default worksheet named `Sheet1` will be created.
func (f *File) NewSheet(name string) int {
// Check if the worksheet already exists
index := f.GetSheetIndex(name)
if index != -1 {
return index
}
f.DeleteSheet(name)
f.SheetCount++
wb := f.workbookReader()
sheetID := 0
for _, v := range wb.Sheets.Sheet {
if v.SheetID > sheetID {
sheetID = v.SheetID
}
}
sheetID++
// Update docProps/app.xml
f.setAppXML()
// Update [Content_Types].xml
f.setContentTypes("/xl/worksheets/sheet"+strconv.Itoa(sheetID)+".xml", ContentTypeSpreadSheetMLWorksheet)
// Create new sheet /xl/worksheets/sheet%d.xml
f.setSheet(sheetID, name)
// Update workbook.xml.rels
rID := f.addRels(f.getWorkbookRelsPath(), SourceRelationshipWorkSheet, fmt.Sprintf("/xl/worksheets/sheet%d.xml", sheetID), "")
// Update workbook.xml
f.setWorkbook(name, sheetID, rID)
return f.GetSheetIndex(name)
}
// contentTypesReader provides a function to get the pointer to the
// [Content_Types].xml structure after deserialization.
func (f *File) contentTypesReader() *xlsxTypes {
var err error
if f.ContentTypes == nil {
f.ContentTypes = new(xlsxTypes)
if err = f.xmlNewDecoder(bytes.NewReader(namespaceStrictToTransitional(f.readXML("[Content_Types].xml")))).
Decode(f.ContentTypes); err != nil && err != io.EOF {
log.Printf("xml decode error: %s", err)
}
}
return f.ContentTypes
}
// contentTypesWriter provides a function to save [Content_Types].xml after
// serialize structure.
func (f *File) contentTypesWriter() {
if f.ContentTypes != nil {
output, _ := xml.Marshal(f.ContentTypes)
f.saveFileList("[Content_Types].xml", output)
}
}
// getWorkbookPath provides a function to get the path of the workbook.xml in
// the spreadsheet.
func (f *File) getWorkbookPath() (path string) {
if rels := f.relsReader("_rels/.rels"); rels != nil {
for _, rel := range rels.Relationships {
if rel.Type == SourceRelationshipOfficeDocument {
path = strings.TrimPrefix(rel.Target, "/")
return
}
}
}
return
}
// getWorkbookRelsPath provides a function to get the path of the workbook.xml.rels
// in the spreadsheet.
func (f *File) getWorkbookRelsPath() (path string) {
wbPath := f.getWorkbookPath()
wbDir := filepath.Dir(wbPath)
if wbDir == "." {
path = "_rels/" + filepath.Base(wbPath) + ".rels"
return
}
path = strings.TrimPrefix(filepath.Dir(wbPath)+"/_rels/"+filepath.Base(wbPath)+".rels", "/")
return
}
// workbookReader provides a function to get the pointer to the workbook.xml
// structure after deserialization.
func (f *File) workbookReader() *xlsxWorkbook {
var err error
if f.WorkBook == nil {
wbPath := f.getWorkbookPath()
f.WorkBook = new(xlsxWorkbook)
if _, ok := f.xmlAttr[wbPath]; !ok {
d := f.xmlNewDecoder(bytes.NewReader(namespaceStrictToTransitional(f.readXML(wbPath))))
f.xmlAttr[wbPath] = append(f.xmlAttr[wbPath], getRootElement(d)...)
f.addNameSpaces(wbPath, SourceRelationship)
}
if err = f.xmlNewDecoder(bytes.NewReader(namespaceStrictToTransitional(f.readXML(wbPath)))).
Decode(f.WorkBook); err != nil && err != io.EOF {
log.Printf("xml decode error: %s", err)
}
}
return f.WorkBook
}
// workBookWriter provides a function to save workbook.xml after serialize
// structure.
func (f *File) workBookWriter() {
if f.WorkBook != nil {
output, _ := xml.Marshal(f.WorkBook)
f.saveFileList(f.getWorkbookPath(), replaceRelationshipsBytes(f.replaceNameSpaceBytes(f.getWorkbookPath(), output)))
}
}
// workSheetWriter provides a function to save xl/worksheets/sheet%d.xml after
// serialize structure.
func (f *File) workSheetWriter() {
var arr []byte
buffer := bytes.NewBuffer(arr)
encoder := xml.NewEncoder(buffer)
for p, sheet := range f.Sheet {
if sheet != nil {
for k, v := range sheet.SheetData.Row {
f.Sheet[p].SheetData.Row[k].C = trimCell(v.C)
}
if sheet.SheetPr != nil || sheet.Drawing != nil || sheet.Hyperlinks != nil || sheet.Picture != nil || sheet.TableParts != nil {
f.addNameSpaces(p, SourceRelationship)
}
// reusing buffer
_ = encoder.Encode(sheet)
f.saveFileList(p, replaceRelationshipsBytes(f.replaceNameSpaceBytes(p, buffer.Bytes())))
ok := f.checked[p]
if ok {
delete(f.Sheet, p)
f.checked[p] = false
}
buffer.Reset()
}
}
}
// trimCell provides a function to trim blank cells which created by fillColumns.
func trimCell(column []xlsxC) []xlsxC {
rowFull := true
for i := range column {
rowFull = column[i].hasValue() && rowFull
}
if rowFull {
return column
}
col := make([]xlsxC, len(column))
i := 0
for _, c := range column {
if c.hasValue() {
col[i] = c
i++
}
}
return col[0:i]
}
// setContentTypes provides a function to read and update property of contents
// type of the spreadsheet.
func (f *File) setContentTypes(partName, contentType string) {
content := f.contentTypesReader()
content.Overrides = append(content.Overrides, xlsxOverride{
PartName: partName,
ContentType: contentType,
})
}
// setSheet provides a function to update sheet property by given index.
func (f *File) setSheet(index int, name string) {
ws := xlsxWorksheet{
Dimension: &xlsxDimension{Ref: "A1"},
SheetViews: &xlsxSheetViews{
SheetView: []xlsxSheetView{{WorkbookViewID: 0}},
},
}
path := "xl/worksheets/sheet" + strconv.Itoa(index) + ".xml"
f.sheetMap[trimSheetName(name)] = path
f.Sheet[path] = &ws
f.xmlAttr[path] = append(f.xmlAttr[path], NameSpaceSpreadSheet)
}
// setWorkbook update workbook property of the spreadsheet. Maximum 31
// characters are allowed in sheet title.
func (f *File) setWorkbook(name string, sheetID, rid int) {
content := f.workbookReader()
content.Sheets.Sheet = append(content.Sheets.Sheet, xlsxSheet{
Name: trimSheetName(name),
SheetID: sheetID,
ID: "rId" + strconv.Itoa(rid),
})
}
// relsWriter provides a function to save relationships after
// serialize structure.
func (f *File) relsWriter() {
for path, rel := range f.Relationships {
if rel != nil {
output, _ := xml.Marshal(rel)
if strings.HasPrefix(path, "xl/worksheets/sheet/rels/sheet") {
output = f.replaceNameSpaceBytes(path, output)
}
f.saveFileList(path, replaceRelationshipsBytes(output))
}
}
}
// setAppXML update docProps/app.xml file of XML.
func (f *File) setAppXML() {
f.saveFileList("docProps/app.xml", []byte(templateDocpropsApp))
}
// replaceRelationshipsBytes; Some tools that read spreadsheet files have very
// strict requirements about the structure of the input XML. This function is
// a horrible hack to fix that after the XML marshalling is completed.
func replaceRelationshipsBytes(content []byte) []byte {
oldXmlns := []byte(`xmlns:relationships="http://schemas.openxmlformats.org/officeDocument/2006/relationships" relationships`)
newXmlns := []byte("r")
return bytesReplace(content, oldXmlns, newXmlns, -1)
}
// SetActiveSheet provides a function to set the default active sheet of the
// workbook by a given index. Note that the active index is different from the
// ID returned by function GetSheetMap(). It should be greater or equal to 0
// and less than the total worksheet numbers.
func (f *File) SetActiveSheet(index int) {
if index < 0 {
index = 0
}
wb := f.workbookReader()
for activeTab := range wb.Sheets.Sheet {
if activeTab == index {
if wb.BookViews == nil {
wb.BookViews = &xlsxBookViews{}
}
if len(wb.BookViews.WorkBookView) > 0 {
wb.BookViews.WorkBookView[0].ActiveTab = activeTab
} else {
wb.BookViews.WorkBookView = append(wb.BookViews.WorkBookView, xlsxWorkBookView{
ActiveTab: activeTab,
})
}
}
}
for idx, name := range f.GetSheetList() {
ws, err := f.workSheetReader(name)
if err != nil {
// Chartsheet or dialogsheet
return
}
if ws.SheetViews == nil {
ws.SheetViews = &xlsxSheetViews{
SheetView: []xlsxSheetView{{WorkbookViewID: 0}},
}
}
if len(ws.SheetViews.SheetView) > 0 {
ws.SheetViews.SheetView[0].TabSelected = false
}
if index == idx {
if len(ws.SheetViews.SheetView) > 0 {
ws.SheetViews.SheetView[0].TabSelected = true
} else {
ws.SheetViews.SheetView = append(ws.SheetViews.SheetView, xlsxSheetView{
TabSelected: true,
})
}
}
}
}
// GetActiveSheetIndex provides a function to get active sheet index of the
// spreadsheet. If not found the active sheet will be return integer 0.
func (f *File) GetActiveSheetIndex() (index int) {
var sheetID = f.getActiveSheetID()
wb := f.workbookReader()
if wb != nil {
for idx, sheet := range wb.Sheets.Sheet {
if sheet.SheetID == sheetID {
index = idx
}
}
}
return
}
// getActiveSheetID provides a function to get active sheet ID of the
// spreadsheet. If not found the active sheet will be return integer 0.
func (f *File) getActiveSheetID() int {
wb := f.workbookReader()
if wb != nil {
if wb.BookViews != nil && len(wb.BookViews.WorkBookView) > 0 {
activeTab := wb.BookViews.WorkBookView[0].ActiveTab
if len(wb.Sheets.Sheet) > activeTab && wb.Sheets.Sheet[activeTab].SheetID != 0 {
return wb.Sheets.Sheet[activeTab].SheetID
}
}
if len(wb.Sheets.Sheet) >= 1 {
return wb.Sheets.Sheet[0].SheetID
}
}
return 0
}
// SetSheetName provides a function to set the worksheet name by given old and
// new worksheet names. Maximum 31 characters are allowed in sheet title and
// this function only changes the name of the sheet and will not update the
// sheet name in the formula or reference associated with the cell. So there
// may be problem formula error or reference missing.
func (f *File) SetSheetName(oldName, newName string) {
oldName = trimSheetName(oldName)
newName = trimSheetName(newName)
if newName == oldName {
return
}
content := f.workbookReader()
for k, v := range content.Sheets.Sheet {
if v.Name == oldName {
content.Sheets.Sheet[k].Name = newName
f.sheetMap[newName] = f.sheetMap[oldName]
delete(f.sheetMap, oldName)
}
}
}
// getSheetNameByID provides a function to get worksheet name of the
// spreadsheet by given worksheet ID. If given sheet ID is invalid, will
// return an empty string.
func (f *File) getSheetNameByID(ID int) string {
wb := f.workbookReader()
if wb == nil || ID < 1 {
return ""
}
for _, sheet := range wb.Sheets.Sheet {
if ID == sheet.SheetID {
return sheet.Name
}
}
return ""
}
// GetSheetName provides a function to get the sheet name of the workbook by
// the given sheet index. If the given sheet index is invalid, it will return
// an empty string.
func (f *File) GetSheetName(index int) (name string) {
for idx, sheet := range f.GetSheetList() {
if idx == index {
name = sheet
}
}
return
}
// getSheetID provides a function to get worksheet ID of the spreadsheet by
// given sheet name. If given worksheet name is invalid, will return an
// integer type value -1.
func (f *File) getSheetID(name string) int {
var ID = -1
for sheetID, sheet := range f.GetSheetMap() {
if sheet == trimSheetName(name) {
ID = sheetID
}
}
return ID
}
// GetSheetIndex provides a function to get a sheet index of the workbook by
// the given sheet name. If the given sheet name is invalid or sheet doesn't
// exist, it will return an integer type value -1.
func (f *File) GetSheetIndex(name string) int {
var idx = -1
for index, sheet := range f.GetSheetList() {
if sheet == trimSheetName(name) {
idx = index
}
}
return idx
}
// GetSheetMap provides a function to get worksheets, chart sheets, dialog
// sheets ID and name map of the workbook. For example:
//
// f, err := excelize.OpenFile("Book1.xlsx")
// if err != nil {
// return
// }
// for index, name := range f.GetSheetMap() {
// fmt.Println(index, name)
// }
//
func (f *File) GetSheetMap() map[int]string {
wb := f.workbookReader()
sheetMap := map[int]string{}
if wb != nil {
for _, sheet := range wb.Sheets.Sheet {
sheetMap[sheet.SheetID] = sheet.Name
}
}
return sheetMap
}
// GetSheetList provides a function to get worksheets, chart sheets, and
// dialog sheets name list of the workbook.
func (f *File) GetSheetList() (list []string) {
wb := f.workbookReader()
if wb != nil {
for _, sheet := range wb.Sheets.Sheet {
list = append(list, sheet.Name)
}
}
return
}
// getSheetMap provides a function to get worksheet name and XML file path map
// of the spreadsheet.
func (f *File) getSheetMap() map[string]string {
maps := map[string]string{}
for _, v := range f.workbookReader().Sheets.Sheet {
for _, rel := range f.relsReader(f.getWorkbookRelsPath()).Relationships {
if rel.ID == v.ID {
// Construct a target XML as xl/worksheets/sheet%d by split
// path, compatible with different types of relative paths in
// workbook.xml.rels, for example: worksheets/sheet%d.xml
// and /xl/worksheets/sheet%d.xml
path := filepath.ToSlash(strings.TrimPrefix(
strings.Replace(filepath.Clean(fmt.Sprintf("%s/%s", filepath.Dir(f.getWorkbookPath()), rel.Target)), "\\", "/", -1), "/"))
if strings.HasPrefix(rel.Target, "/") {
path = filepath.ToSlash(strings.TrimPrefix(strings.Replace(filepath.Clean(rel.Target), "\\", "/", -1), "/"))
}
if _, ok := f.XLSX[path]; ok {
maps[v.Name] = path
}
}
}
}
return maps
}
// SetSheetBackground provides a function to set background picture by given
// worksheet name and file path.
func (f *File) SetSheetBackground(sheet, picture string) error {
var err error
// Check picture exists first.
if _, err = os.Stat(picture); os.IsNotExist(err) {
return err
}
ext, ok := supportImageTypes[path.Ext(picture)]
if !ok {
return errors.New("unsupported image extension")
}
file, _ := ioutil.ReadFile(picture)
name := f.addMedia(file, ext)
sheetRels := "xl/worksheets/_rels/" + strings.TrimPrefix(f.sheetMap[trimSheetName(sheet)], "xl/worksheets/") + ".rels"
rID := f.addRels(sheetRels, SourceRelationshipImage, strings.Replace(name, "xl", "..", 1), "")
f.addSheetPicture(sheet, rID)
f.addSheetNameSpace(sheet, SourceRelationship)
f.setContentTypePartImageExtensions()
return err
}
// DeleteSheet provides a function to delete worksheet in a workbook by given
// worksheet name. Use this method with caution, which will affect changes in
// references such as formulas, charts, and so on. If there is any referenced
// value of the deleted worksheet, it will cause a file error when you open it.
// This function will be invalid when only the one worksheet is left.
func (f *File) DeleteSheet(name string) {
if f.SheetCount == 1 || f.GetSheetIndex(name) == -1 {
return
}
sheetName := trimSheetName(name)
wb := f.workbookReader()
wbRels := f.relsReader(f.getWorkbookRelsPath())
activeSheetName := f.GetSheetName(f.GetActiveSheetIndex())
deleteSheetID := f.getSheetID(name)
// Delete and adjust defined names
if wb.DefinedNames != nil {
for idx := 0; idx < len(wb.DefinedNames.DefinedName); idx++ {
dn := wb.DefinedNames.DefinedName[idx]
if dn.LocalSheetID != nil {
sheetID := *dn.LocalSheetID + 1
if sheetID == deleteSheetID {
wb.DefinedNames.DefinedName = append(wb.DefinedNames.DefinedName[:idx], wb.DefinedNames.DefinedName[idx+1:]...)
idx--
} else if sheetID > deleteSheetID {
wb.DefinedNames.DefinedName[idx].LocalSheetID = intPtr(*dn.LocalSheetID - 1)
}
}
}
}
for idx, sheet := range wb.Sheets.Sheet {
if sheet.Name == sheetName {
wb.Sheets.Sheet = append(wb.Sheets.Sheet[:idx], wb.Sheets.Sheet[idx+1:]...)
var sheetXML, rels string
if wbRels != nil {
for _, rel := range wbRels.Relationships {
if rel.ID == sheet.ID {
sheetXML = fmt.Sprintf("xl/%s", rel.Target)
pathInfo := strings.Split(rel.Target, "/")
if len(pathInfo) == 2 {
rels = fmt.Sprintf("xl/%s/_rels/%s.rels", pathInfo[0], pathInfo[1])
}
}
}
}
target := f.deleteSheetFromWorkbookRels(sheet.ID)
f.deleteSheetFromContentTypes(target)
f.deleteCalcChain(sheet.SheetID, "")
delete(f.sheetMap, sheetName)
delete(f.XLSX, sheetXML)
delete(f.XLSX, rels)
delete(f.Relationships, rels)
delete(f.Sheet, sheetXML)
delete(f.xmlAttr, sheetXML)
f.SheetCount--
}
}
f.SetActiveSheet(f.GetSheetIndex(activeSheetName))
}
// deleteSheetFromWorkbookRels provides a function to remove worksheet
// relationships by given relationships ID in the file workbook.xml.rels.
func (f *File) deleteSheetFromWorkbookRels(rID string) string {
content := f.relsReader(f.getWorkbookRelsPath())
for k, v := range content.Relationships {
if v.ID == rID {
content.Relationships = append(content.Relationships[:k], content.Relationships[k+1:]...)
return v.Target
}
}
return ""
}
// deleteSheetFromContentTypes provides a function to remove worksheet
// relationships by given target name in the file [Content_Types].xml.
func (f *File) deleteSheetFromContentTypes(target string) {
content := f.contentTypesReader()
for k, v := range content.Overrides {
if v.PartName == "/xl/"+target {
content.Overrides = append(content.Overrides[:k], content.Overrides[k+1:]...)
}
}
}
// CopySheet provides a function to duplicate a worksheet by gave source and
// target worksheet index. Note that currently doesn't support duplicate
// workbooks that contain tables, charts or pictures. For Example:
//
// // Sheet1 already exists...
// index := f.NewSheet("Sheet2")
// err := f.CopySheet(1, index)
// return err
//
func (f *File) CopySheet(from, to int) error {
if from < 0 || to < 0 || from == to || f.GetSheetName(from) == "" || f.GetSheetName(to) == "" {
return errors.New("invalid worksheet index")
}
return f.copySheet(from, to)
}
// copySheet provides a function to duplicate a worksheet by gave source and
// target worksheet name.
func (f *File) copySheet(from, to int) error {
fromSheet := f.GetSheetName(from)
sheet, err := f.workSheetReader(fromSheet)
if err != nil {
return err
}
worksheet := deepcopy.Copy(sheet).(*xlsxWorksheet)
toSheetID := strconv.Itoa(f.getSheetID(f.GetSheetName(to)))
path := "xl/worksheets/sheet" + toSheetID + ".xml"
if len(worksheet.SheetViews.SheetView) > 0 {
worksheet.SheetViews.SheetView[0].TabSelected = false
}
worksheet.Drawing = nil
worksheet.TableParts = nil
worksheet.PageSetUp = nil
f.Sheet[path] = worksheet
toRels := "xl/worksheets/_rels/sheet" + toSheetID + ".xml.rels"
fromRels := "xl/worksheets/_rels/sheet" + strconv.Itoa(f.getSheetID(fromSheet)) + ".xml.rels"
_, ok := f.XLSX[fromRels]
if ok {
f.XLSX[toRels] = f.XLSX[fromRels]
}
fromSheetXMLPath := f.sheetMap[trimSheetName(fromSheet)]
fromSheetAttr := f.xmlAttr[fromSheetXMLPath]
f.xmlAttr[path] = fromSheetAttr
return err
}
// SetSheetVisible provides a function to set worksheet visible by given worksheet
// name. A workbook must contain at least one visible worksheet. If the given
// worksheet has been activated, this setting will be invalidated. Sheet state
// values as defined by https://docs.microsoft.com/en-us/dotnet/api/documentformat.openxml.spreadsheet.sheetstatevalues
//
// visible
// hidden
// veryHidden
//
// For example, hide Sheet1:
//
// err := f.SetSheetVisible("Sheet1", false)
//
func (f *File) SetSheetVisible(name string, visible bool) error {
name = trimSheetName(name)
content := f.workbookReader()
if visible {
for k, v := range content.Sheets.Sheet {
if v.Name == name {
content.Sheets.Sheet[k].State = ""
}
}
return nil
}
count := 0
for _, v := range content.Sheets.Sheet {
if v.State != "hidden" {
count++
}
}
for k, v := range content.Sheets.Sheet {
xlsx, err := f.workSheetReader(v.Name)
if err != nil {
return err
}
tabSelected := false
if len(xlsx.SheetViews.SheetView) > 0 {
tabSelected = xlsx.SheetViews.SheetView[0].TabSelected
}
if v.Name == name && count > 1 && !tabSelected {
content.Sheets.Sheet[k].State = "hidden"
}
}
return nil
}
// parseFormatPanesSet provides a function to parse the panes settings.
func parseFormatPanesSet(formatSet string) (*formatPanes, error) {
format := formatPanes{}
err := json.Unmarshal([]byte(formatSet), &format)
return &format, err
}
// SetPanes provides a function to create and remove freeze panes and split panes
// by given worksheet name and panes format set.
//
// activePane defines the pane that is active. The possible values for this
// attribute are defined in the following table:
//
// Enumeration Value | Description
// --------------------------------+-------------------------------------------------------------
// bottomLeft (Bottom Left Pane) | Bottom left pane, when both vertical and horizontal
// | splits are applied.
// |
// | This value is also used when only a horizontal split has
// | been applied, dividing the pane into upper and lower
// | regions. In that case, this value specifies the bottom
// | pane.
// |
// bottomRight (Bottom Right Pane) | Bottom right pane, when both vertical and horizontal
// | splits are applied.
// |
// topLeft (Top Left Pane) | Top left pane, when both vertical and horizontal splits
// | are applied.
// |
// | This value is also used when only a horizontal split has
// | been applied, dividing the pane into upper and lower
// | regions. In that case, this value specifies the top pane.
// |
// | This value is also used when only a vertical split has
// | been applied, dividing the pane into right and left
// | regions. In that case, this value specifies the left pane
// |
// topRight (Top Right Pane) | Top right pane, when both vertical and horizontal
// | splits are applied.
// |
// | This value is also used when only a vertical split has
// | been applied, dividing the pane into right and left
// | regions. In that case, this value specifies the right
// | pane.
//
// Pane state type is restricted to the values supported currently listed in the following table:
//
// Enumeration Value | Description
// --------------------------------+-------------------------------------------------------------
// frozen (Frozen) | Panes are frozen, but were not split being frozen. In
// | this state, when the panes are unfrozen again, a single
// | pane results, with no split.
// |
// | In this state, the split bars are not adjustable.
// |
// split (Split) | Panes are split, but not frozen. In this state, the split
// | bars are adjustable by the user.
//
// x_split (Horizontal Split Position): Horizontal position of the split, in
// 1/20th of a point; 0 (zero) if none. If the pane is frozen, this value
// indicates the number of columns visible in the top pane.
//
// y_split (Vertical Split Position): Vertical position of the split, in 1/20th
// of a point; 0 (zero) if none. If the pane is frozen, this value indicates the
// number of rows visible in the left pane. The possible values for this
// attribute are defined by the W3C XML Schema double datatype.
//
// top_left_cell: Location of the top left visible cell in the bottom right pane
// (when in Left-To-Right mode).
//
// sqref (Sequence of References): Range of the selection. Can be non-contiguous
// set of ranges.
//
// An example of how to freeze column A in the Sheet1 and set the active cell on
// Sheet1!K16:
//
// f.SetPanes("Sheet1", `{"freeze":true,"split":false,"x_split":1,"y_split":0,"top_left_cell":"B1","active_pane":"topRight","panes":[{"sqref":"K16","active_cell":"K16","pane":"topRight"}]}`)
//
// An example of how to freeze rows 1 to 9 in the Sheet1 and set the active cell
// ranges on Sheet1!A11:XFD11:
//
// f.SetPanes("Sheet1", `{"freeze":true,"split":false,"x_split":0,"y_split":9,"top_left_cell":"A34","active_pane":"bottomLeft","panes":[{"sqref":"A11:XFD11","active_cell":"A11","pane":"bottomLeft"}]}`)
//
// An example of how to create split panes in the Sheet1 and set the active cell
// on Sheet1!J60:
//
// f.SetPanes("Sheet1", `{"freeze":false,"split":true,"x_split":3270,"y_split":1800,"top_left_cell":"N57","active_pane":"bottomLeft","panes":[{"sqref":"I36","active_cell":"I36"},{"sqref":"G33","active_cell":"G33","pane":"topRight"},{"sqref":"J60","active_cell":"J60","pane":"bottomLeft"},{"sqref":"O60","active_cell":"O60","pane":"bottomRight"}]}`)
//
// An example of how to unfreeze and remove all panes on Sheet1:
//
// f.SetPanes("Sheet1", `{"freeze":false,"split":false}`)
//
func (f *File) SetPanes(sheet, panes string) error {
fs, _ := parseFormatPanesSet(panes)
ws, err := f.workSheetReader(sheet)
if err != nil {
return err
}
p := &xlsxPane{
ActivePane: fs.ActivePane,
TopLeftCell: fs.TopLeftCell,
XSplit: float64(fs.XSplit),
YSplit: float64(fs.YSplit),
}
if fs.Freeze {
p.State = "frozen"
}
ws.SheetViews.SheetView[len(ws.SheetViews.SheetView)-1].Pane = p
if !(fs.Freeze) && !(fs.Split) {
if len(ws.SheetViews.SheetView) > 0 {
ws.SheetViews.SheetView[len(ws.SheetViews.SheetView)-1].Pane = nil
}
}
s := []*xlsxSelection{}
for _, p := range fs.Panes {
s = append(s, &xlsxSelection{
ActiveCell: p.ActiveCell,
Pane: p.Pane,
SQRef: p.SQRef,
})
}
ws.SheetViews.SheetView[len(ws.SheetViews.SheetView)-1].Selection = s
return err
}
// GetSheetVisible provides a function to get worksheet visible by given worksheet
// name. For example, get visible state of Sheet1:
//
// f.GetSheetVisible("Sheet1")
//
func (f *File) GetSheetVisible(name string) bool {
content := f.workbookReader()
visible := false
for k, v := range content.Sheets.Sheet {
if v.Name == trimSheetName(name) {
if content.Sheets.Sheet[k].State == "" || content.Sheets.Sheet[k].State == "visible" {
visible = true
}
}
}
return visible
}
// SearchSheet provides a function to get coordinates by given worksheet name,
// cell value, and regular expression. The function doesn't support searching
// on the calculated result, formatted numbers and conditional lookup
// currently. If it is a merged cell, it will return the coordinates of the
// upper left corner of the merged area.
//
// An example of search the coordinates of the value of "100" on Sheet1:
//
// result, err := f.SearchSheet("Sheet1", "100")
//
// An example of search the coordinates where the numerical value in the range
// of "0-9" of Sheet1 is described:
//
// result, err := f.SearchSheet("Sheet1", "[0-9]", true)
//
func (f *File) SearchSheet(sheet, value string, reg ...bool) ([]string, error) {
var (
regSearch bool
result []string
)
for _, r := range reg {
regSearch = r
}
name, ok := f.sheetMap[trimSheetName(sheet)]
if !ok {
return result, ErrSheetNotExist{sheet}
}
if f.Sheet[name] != nil {
// flush data
output, _ := xml.Marshal(f.Sheet[name])
f.saveFileList(name, f.replaceNameSpaceBytes(name, output))
}
return f.searchSheet(name, value, regSearch)
}
// searchSheet provides a function to get coordinates by given worksheet name,
// cell value, and regular expression.
func (f *File) searchSheet(name, value string, regSearch bool) (result []string, err error) {
var (
cellName, inElement string
cellCol, row int
d *xlsxSST
)
d = f.sharedStringsReader()
decoder := f.xmlNewDecoder(bytes.NewReader(f.readXML(name)))
for {
var token xml.Token
token, err = decoder.Token()
if err != nil || token == nil {
if err == io.EOF {
err = nil
}
break
}
switch xmlElement := token.(type) {
case xml.StartElement:
inElement = xmlElement.Name.Local
if inElement == "row" {
row, err = attrValToInt("r", xmlElement.Attr)
if err != nil {
return
}
}
if inElement == "c" {
colCell := xlsxC{}
_ = decoder.DecodeElement(&colCell, &xmlElement)
val, _ := colCell.getValueFrom(f, d)
if regSearch {
regex := regexp.MustCompile(value)
if !regex.MatchString(val) {
continue
}
} else {
if val != value {
continue
}
}
cellCol, _, err = CellNameToCoordinates(colCell.R)
if err != nil {
return result, err
}
cellName, err = CoordinatesToCellName(cellCol, row)
if err != nil {
return result, err
}
result = append(result, cellName)
}
default:
}
}
return
}
// attrValToInt provides a function to convert the local names to an integer
// by given XML attributes and specified names.
func attrValToInt(name string, attrs []xml.Attr) (val int, err error) {
for _, attr := range attrs {
if attr.Name.Local == name {
val, err = strconv.Atoi(attr.Value)
if err != nil {
return
}
}
}
return
}
// SetHeaderFooter provides a function to set headers and footers by given
// worksheet name and the control characters.
//
// Headers and footers are specified using the following settings fields:
//
// Fields | Description
// ------------------+-----------------------------------------------------------
// AlignWithMargins | Align header footer margins with page margins
// DifferentFirst | Different first-page header and footer indicator
// DifferentOddEven | Different odd and even page headers and footers indicator
// ScaleWithDoc | Scale header and footer with document scaling
// OddFooter | Odd Page Footer
// OddHeader | Odd Header
// EvenFooter | Even Page Footer
// EvenHeader | Even Page Header
// FirstFooter | First Page Footer
// FirstHeader | First Page Header
//
// The following formatting codes can be used in 6 string type fields:
// OddHeader, OddFooter, EvenHeader, EvenFooter, FirstFooter, FirstHeader
//
// Formatting Code | Description
// ------------------------+-------------------------------------------------------------------------
// && | The character "&"
// |
// &font-size | Size of the text font, where font-size is a decimal font size in points
// |
// &"font name,font type" | A text font-name string, font name, and a text font-type string,
// | font type
// |
// &"-,Regular" | Regular text format. Toggles bold and italic modes to off
// |
// &A | Current worksheet's tab name
// |
// &B or &"-,Bold" | Bold text format, from off to on, or vice versa. The default mode is off
// |
// &D | Current date
// |
// &C | Center section
// |
// &E | Double-underline text format
// |
// &F | Current workbook's file name
// |
// &G | Drawing object as background
// |
// &H | Shadow text format
// |
// &I or &"-,Italic" | Italic text format
// |
// &K | Text font color
// |
// | An RGB Color is specified as RRGGBB
// |
// | A Theme Color is specified as TTSNNN where TT is the theme color Id,
// | S is either "+" or "-" of the tint/shade value, and NNN is the
// | tint/shade value
// |
// &L | Left section
// |
// &N | Total number of pages
// |
// &O | Outline text format
// |
// &P[[+|-]n] | Without the optional suffix, the current page number in decimal
// |
// &R | Right section
// |
// &S | Strikethrough text format
// |
// &T | Current time
// |
// &U | Single-underline text format. If double-underline mode is on, the next
// | occurrence in a section specifier toggles double-underline mode to off;
// | otherwise, it toggles single-underline mode, from off to on, or vice
// | versa. The default mode is off
// |
// &X | Superscript text format
// |
// &Y | Subscript text format
// |
// &Z | Current workbook's file path
//
// For example:
//
// err := f.SetHeaderFooter("Sheet1", &excelize.FormatHeaderFooter{
// DifferentFirst: true,
// DifferentOddEven: true,
// OddHeader: "&R&P",
// OddFooter: "&C&F",
// EvenHeader: "&L&P",
// EvenFooter: "&L&D&R&T",
// FirstHeader: `&CCenter &"-,Bold"Bold&"-,Regular"HeaderU+000A&D`,
// })
//
// This example shows:
//
// - The first page has its own header and footer
//
// - Odd and even-numbered pages have different headers and footers
//
// - Current page number in the right section of odd-page headers
//
// - Current workbook's file name in the center section of odd-page footers
//
// - Current page number in the left section of even-page headers
//
// - Current date in the left section and the current time in the right section
// of even-page footers
//
// - The text "Center Bold Header" on the first line of the center section of
// the first page, and the date on the second line of the center section of
// that same page
//
// - No footer on the first page
//
func (f *File) SetHeaderFooter(sheet string, settings *FormatHeaderFooter) error {
ws, err := f.workSheetReader(sheet)
if err != nil {
return err
}
if settings == nil {
ws.HeaderFooter = nil
return err
}
v := reflect.ValueOf(*settings)
// Check 6 string type fields: OddHeader, OddFooter, EvenHeader, EvenFooter,
// FirstFooter, FirstHeader
for i := 4; i < v.NumField()-1; i++ {
if v.Field(i).Len() >= 255 {
return fmt.Errorf("field %s must be less than 255 characters", v.Type().Field(i).Name)
}
}
ws.HeaderFooter = &xlsxHeaderFooter{
AlignWithMargins: settings.AlignWithMargins,
DifferentFirst: settings.DifferentFirst,
DifferentOddEven: settings.DifferentOddEven,
ScaleWithDoc: settings.ScaleWithDoc,
OddHeader: settings.OddHeader,
OddFooter: settings.OddFooter,
EvenHeader: settings.EvenHeader,
EvenFooter: settings.EvenFooter,
FirstFooter: settings.FirstFooter,
FirstHeader: settings.FirstHeader,
}
return err
}
// ProtectSheet provides a function to prevent other users from accidentally
// or deliberately changing, moving, or deleting data in a worksheet. For
// example, protect Sheet1 with protection settings:
//
// err := f.ProtectSheet("Sheet1", &excelize.FormatSheetProtection{
// Password: "password",
// EditScenarios: false,
// })
//
func (f *File) ProtectSheet(sheet string, settings *FormatSheetProtection) error {
ws, err := f.workSheetReader(sheet)
if err != nil {
return err
}
if settings == nil {
settings = &FormatSheetProtection{
EditObjects: true,
EditScenarios: true,
SelectLockedCells: true,
}
}
ws.SheetProtection = &xlsxSheetProtection{
AutoFilter: settings.AutoFilter,
DeleteColumns: settings.DeleteColumns,
DeleteRows: settings.DeleteRows,
FormatCells: settings.FormatCells,
FormatColumns: settings.FormatColumns,
FormatRows: settings.FormatRows,
InsertColumns: settings.InsertColumns,
InsertHyperlinks: settings.InsertHyperlinks,
InsertRows: settings.InsertRows,
Objects: settings.EditObjects,
PivotTables: settings.PivotTables,
Scenarios: settings.EditScenarios,
SelectLockedCells: settings.SelectLockedCells,
SelectUnlockedCells: settings.SelectUnlockedCells,
Sheet: true,
Sort: settings.Sort,
}
if settings.Password != "" {
ws.SheetProtection.Password = genSheetPasswd(settings.Password)
}
return err
}
// UnprotectSheet provides a function to unprotect an Excel worksheet.
func (f *File) UnprotectSheet(sheet string) error {
ws, err := f.workSheetReader(sheet)
if err != nil {
return err
}
ws.SheetProtection = nil
return err
}
// trimSheetName provides a function to trim invaild characters by given worksheet
// name.
func trimSheetName(name string) string {
if strings.ContainsAny(name, ":\\/?*[]") || utf8.RuneCountInString(name) > 31 {
r := make([]rune, 0, 31)
for _, v := range name {
switch v {
case 58, 92, 47, 63, 42, 91, 93: // replace :\/?*[]
continue
default:
r = append(r, v)
}
if len(r) == 31 {
break
}
}
name = string(r)
}
return name
}
// PageLayoutOption is an option of a page layout of a worksheet. See
// SetPageLayout().
type PageLayoutOption interface {
setPageLayout(layout *xlsxPageSetUp)
}
// PageLayoutOptionPtr is a writable PageLayoutOption. See GetPageLayout().
type PageLayoutOptionPtr interface {
PageLayoutOption
getPageLayout(layout *xlsxPageSetUp)
}
type (
// BlackAndWhite specified print black and white.
BlackAndWhite bool
// FirstPageNumber specified the first printed page number. If no value is
// specified, then 'automatic' is assumed.
FirstPageNumber uint
// PageLayoutOrientation defines the orientation of page layout for a
// worksheet.
PageLayoutOrientation string
// PageLayoutPaperSize defines the paper size of the worksheet.
PageLayoutPaperSize int
// FitToHeight specified the number of vertical pages to fit on.
FitToHeight int
// FitToWidth specified the number of horizontal pages to fit on.
FitToWidth int
// PageLayoutScale defines the print scaling. This attribute is restricted
// to values ranging from 10 (10%) to 400 (400%). This setting is
// overridden when fitToWidth and/or fitToHeight are in use.
PageLayoutScale uint
)
const (
// OrientationPortrait indicates page layout orientation id portrait.
OrientationPortrait = "portrait"
// OrientationLandscape indicates page layout orientation id landscape.
OrientationLandscape = "landscape"
)
// setPageLayout provides a method to set the print black and white for the
// worksheet.
func (p BlackAndWhite) setPageLayout(ps *xlsxPageSetUp) {
ps.BlackAndWhite = bool(p)
}
// getPageLayout provides a method to get the print black and white for the
// worksheet.
func (p *BlackAndWhite) getPageLayout(ps *xlsxPageSetUp) {
if ps == nil {
*p = false
return
}
*p = BlackAndWhite(ps.BlackAndWhite)
}
// setPageLayout provides a method to set the first printed page number for
// the worksheet.
func (p FirstPageNumber) setPageLayout(ps *xlsxPageSetUp) {
if 0 < uint(p) {
ps.FirstPageNumber = uint(p)
ps.UseFirstPageNumber = true
}
}
// getPageLayout provides a method to get the first printed page number for
// the worksheet.
func (p *FirstPageNumber) getPageLayout(ps *xlsxPageSetUp) {
if ps == nil || ps.FirstPageNumber == 0 || !ps.UseFirstPageNumber {
*p = 1
return
}
*p = FirstPageNumber(ps.FirstPageNumber)
}
// setPageLayout provides a method to set the orientation for the worksheet.
func (o PageLayoutOrientation) setPageLayout(ps *xlsxPageSetUp) {
ps.Orientation = string(o)
}
// getPageLayout provides a method to get the orientation for the worksheet.
func (o *PageLayoutOrientation) getPageLayout(ps *xlsxPageSetUp) {
// Excel default: portrait
if ps == nil || ps.Orientation == "" {
*o = OrientationPortrait
return
}
*o = PageLayoutOrientation(ps.Orientation)
}
// setPageLayout provides a method to set the paper size for the worksheet.
func (p PageLayoutPaperSize) setPageLayout(ps *xlsxPageSetUp) {
ps.PaperSize = int(p)
}
// getPageLayout provides a method to get the paper size for the worksheet.
func (p *PageLayoutPaperSize) getPageLayout(ps *xlsxPageSetUp) {
// Excel default: 1
if ps == nil || ps.PaperSize == 0 {
*p = 1
return
}
*p = PageLayoutPaperSize(ps.PaperSize)
}
// setPageLayout provides a method to set the fit to height for the worksheet.
func (p FitToHeight) setPageLayout(ps *xlsxPageSetUp) {
if int(p) > 0 {
ps.FitToHeight = int(p)
}
}
// getPageLayout provides a method to get the fit to height for the worksheet.
func (p *FitToHeight) getPageLayout(ps *xlsxPageSetUp) {
if ps == nil || ps.FitToHeight == 0 {
*p = 1
return
}
*p = FitToHeight(ps.FitToHeight)
}
// setPageLayout provides a method to set the fit to width for the worksheet.
func (p FitToWidth) setPageLayout(ps *xlsxPageSetUp) {
if int(p) > 0 {
ps.FitToWidth = int(p)
}
}
// getPageLayout provides a method to get the fit to width for the worksheet.
func (p *FitToWidth) getPageLayout(ps *xlsxPageSetUp) {
if ps == nil || ps.FitToWidth == 0 {
*p = 1
return
}
*p = FitToWidth(ps.FitToWidth)
}
// setPageLayout provides a method to set the scale for the worksheet.
func (p PageLayoutScale) setPageLayout(ps *xlsxPageSetUp) {
if 10 <= uint(p) && uint(p) <= 400 {
ps.Scale = uint(p)
}
}
// getPageLayout provides a method to get the scale for the worksheet.
func (p *PageLayoutScale) getPageLayout(ps *xlsxPageSetUp) {
if ps == nil || ps.Scale < 10 || ps.Scale > 400 {
*p = 100
return
}
*p = PageLayoutScale(ps.Scale)
}
// SetPageLayout provides a function to sets worksheet page layout.
//
// Available options:
//
// BlackAndWhite(bool)
// FirstPageNumber(uint)
// PageLayoutOrientation(string)
// PageLayoutPaperSize(int)
// FitToHeight(int)
// FitToWidth(int)
// PageLayoutScale(uint)
//
// The following shows the paper size sorted by excelize index number:
//
// Index | Paper Size
// -------+-----------------------------------------------
// 1 | Letter paper (8.5 in. by 11 in.)
// 2 | Letter small paper (8.5 in. by 11 in.)
// 3 | Tabloid paper (11 in. by 17 in.)
// 4 | Ledger paper (17 in. by 11 in.)
// 5 | Legal paper (8.5 in. by 14 in.)
// 6 | Statement paper (5.5 in. by 8.5 in.)
// 7 | Executive paper (7.25 in. by 10.5 in.)
// 8 | A3 paper (297 mm by 420 mm)
// 9 | A4 paper (210 mm by 297 mm)
// 10 | A4 small paper (210 mm by 297 mm)
// 11 | A5 paper (148 mm by 210 mm)
// 12 | B4 paper (250 mm by 353 mm)
// 13 | B5 paper (176 mm by 250 mm)
// 14 | Folio paper (8.5 in. by 13 in.)
// 15 | Quarto paper (215 mm by 275 mm)
// 16 | Standard paper (10 in. by 14 in.)
// 17 | Standard paper (11 in. by 17 in.)
// 18 | Note paper (8.5 in. by 11 in.)
// 19 | #9 envelope (3.875 in. by 8.875 in.)
// 20 | #10 envelope (4.125 in. by 9.5 in.)
// 21 | #11 envelope (4.5 in. by 10.375 in.)
// 22 | #12 envelope (4.75 in. by 11 in.)
// 23 | #14 envelope (5 in. by 11.5 in.)
// 24 | C paper (17 in. by 22 in.)
// 25 | D paper (22 in. by 34 in.)
// 26 | E paper (34 in. by 44 in.)
// 27 | DL envelope (110 mm by 220 mm)
// 28 | C5 envelope (162 mm by 229 mm)
// 29 | C3 envelope (324 mm by 458 mm)
// 30 | C4 envelope (229 mm by 324 mm)
// 31 | C6 envelope (114 mm by 162 mm)
// 32 | C65 envelope (114 mm by 229 mm)
// 33 | B4 envelope (250 mm by 353 mm)
// 34 | B5 envelope (176 mm by 250 mm)
// 35 | B6 envelope (176 mm by 125 mm)
// 36 | Italy envelope (110 mm by 230 mm)
// 37 | Monarch envelope (3.875 in. by 7.5 in.).
// 38 | 6 3/4 envelope (3.625 in. by 6.5 in.)
// 39 | US standard fanfold (14.875 in. by 11 in.)
// 40 | German standard fanfold (8.5 in. by 12 in.)
// 41 | German legal fanfold (8.5 in. by 13 in.)
// 42 | ISO B4 (250 mm by 353 mm)
// 43 | Japanese postcard (100 mm by 148 mm)
// 44 | Standard paper (9 in. by 11 in.)
// 45 | Standard paper (10 in. by 11 in.)
// 46 | Standard paper (15 in. by 11 in.)
// 47 | Invite envelope (220 mm by 220 mm)
// 50 | Letter extra paper (9.275 in. by 12 in.)
// 51 | Legal extra paper (9.275 in. by 15 in.)
// 52 | Tabloid extra paper (11.69 in. by 18 in.)
// 53 | A4 extra paper (236 mm by 322 mm)
// 54 | Letter transverse paper (8.275 in. by 11 in.)
// 55 | A4 transverse paper (210 mm by 297 mm)
// 56 | Letter extra transverse paper (9.275 in. by 12 in.)
// 57 | SuperA/SuperA/A4 paper (227 mm by 356 mm)
// 58 | SuperB/SuperB/A3 paper (305 mm by 487 mm)
// 59 | Letter plus paper (8.5 in. by 12.69 in.)
// 60 | A4 plus paper (210 mm by 330 mm)
// 61 | A5 transverse paper (148 mm by 210 mm)
// 62 | JIS B5 transverse paper (182 mm by 257 mm)
// 63 | A3 extra paper (322 mm by 445 mm)
// 64 | A5 extra paper (174 mm by 235 mm)
// 65 | ISO B5 extra paper (201 mm by 276 mm)
// 66 | A2 paper (420 mm by 594 mm)
// 67 | A3 transverse paper (297 mm by 420 mm)
// 68 | A3 extra transverse paper (322 mm by 445 mm)
// 69 | Japanese Double Postcard (200 mm x 148 mm)
// 70 | A6 (105 mm x 148 mm)
// 71 | Japanese Envelope Kaku #2
// 72 | Japanese Envelope Kaku #3
// 73 | Japanese Envelope Chou #3
// 74 | Japanese Envelope Chou #4
// 75 | Letter Rotated (11in x 8 1/2 11 in)
// 76 | A3 Rotated (420 mm x 297 mm)
// 77 | A4 Rotated (297 mm x 210 mm)
// 78 | A5 Rotated (210 mm x 148 mm)
// 79 | B4 (JIS) Rotated (364 mm x 257 mm)
// 80 | B5 (JIS) Rotated (257 mm x 182 mm)
// 81 | Japanese Postcard Rotated (148 mm x 100 mm)
// 82 | Double Japanese Postcard Rotated (148 mm x 200 mm)
// 83 | A6 Rotated (148 mm x 105 mm)
// 84 | Japanese Envelope Kaku #2 Rotated
// 85 | Japanese Envelope Kaku #3 Rotated
// 86 | Japanese Envelope Chou #3 Rotated
// 87 | Japanese Envelope Chou #4 Rotated
// 88 | B6 (JIS) (128 mm x 182 mm)
// 89 | B6 (JIS) Rotated (182 mm x 128 mm)
// 90 | (12 in x 11 in)
// 91 | Japanese Envelope You #4
// 92 | Japanese Envelope You #4 Rotated
// 93 | PRC 16K (146 mm x 215 mm)
// 94 | PRC 32K (97 mm x 151 mm)
// 95 | PRC 32K(Big) (97 mm x 151 mm)
// 96 | PRC Envelope #1 (102 mm x 165 mm)
// 97 | PRC Envelope #2 (102 mm x 176 mm)
// 98 | PRC Envelope #3 (125 mm x 176 mm)
// 99 | PRC Envelope #4 (110 mm x 208 mm)
// 100 | PRC Envelope #5 (110 mm x 220 mm)
// 101 | PRC Envelope #6 (120 mm x 230 mm)
// 102 | PRC Envelope #7 (160 mm x 230 mm)
// 103 | PRC Envelope #8 (120 mm x 309 mm)
// 104 | PRC Envelope #9 (229 mm x 324 mm)
// 105 | PRC Envelope #10 (324 mm x 458 mm)
// 106 | PRC 16K Rotated
// 107 | PRC 32K Rotated
// 108 | PRC 32K(Big) Rotated
// 109 | PRC Envelope #1 Rotated (165 mm x 102 mm)
// 110 | PRC Envelope #2 Rotated (176 mm x 102 mm)
// 111 | PRC Envelope #3 Rotated (176 mm x 125 mm)
// 112 | PRC Envelope #4 Rotated (208 mm x 110 mm)
// 113 | PRC Envelope #5 Rotated (220 mm x 110 mm)
// 114 | PRC Envelope #6 Rotated (230 mm x 120 mm)
// 115 | PRC Envelope #7 Rotated (230 mm x 160 mm)
// 116 | PRC Envelope #8 Rotated (309 mm x 120 mm)
// 117 | PRC Envelope #9 Rotated (324 mm x 229 mm)
// 118 | PRC Envelope #10 Rotated (458 mm x 324 mm)
//
func (f *File) SetPageLayout(sheet string, opts ...PageLayoutOption) error {
s, err := f.workSheetReader(sheet)
if err != nil {
return err
}
ps := s.PageSetUp
if ps == nil {
ps = new(xlsxPageSetUp)
s.PageSetUp = ps
}
for _, opt := range opts {
opt.setPageLayout(ps)
}
return err
}
// GetPageLayout provides a function to gets worksheet page layout.
//
// Available options:
// PageLayoutOrientation(string)
// PageLayoutPaperSize(int)
// FitToHeight(int)
// FitToWidth(int)
func (f *File) GetPageLayout(sheet string, opts ...PageLayoutOptionPtr) error {
s, err := f.workSheetReader(sheet)
if err != nil {
return err
}
ps := s.PageSetUp
for _, opt := range opts {
opt.getPageLayout(ps)
}
return err
}
// SetDefinedName provides a function to set the defined names of the workbook
// or worksheet. If not specified scope, the default scope is workbook.
// For example:
//
// f.SetDefinedName(&excelize.DefinedName{
// Name: "Amount",
// RefersTo: "Sheet1!$A$2:$D$5",
// Comment: "defined name comment",
// Scope: "Sheet2",
// })
//
func (f *File) SetDefinedName(definedName *DefinedName) error {
wb := f.workbookReader()
d := xlsxDefinedName{
Name: definedName.Name,
Comment: definedName.Comment,
Data: definedName.RefersTo,
}
if definedName.Scope != "" {
if sheetID := f.getSheetID(definedName.Scope); sheetID != 0 {
sheetID--
d.LocalSheetID = &sheetID
}
}
if wb.DefinedNames != nil {
for _, dn := range wb.DefinedNames.DefinedName {
var scope string
if dn.LocalSheetID != nil {
scope = f.getSheetNameByID(*dn.LocalSheetID + 1)
}
if scope == definedName.Scope && dn.Name == definedName.Name {
return errors.New("the same name already exists on the scope")
}
}
wb.DefinedNames.DefinedName = append(wb.DefinedNames.DefinedName, d)
return nil
}
wb.DefinedNames = &xlsxDefinedNames{
DefinedName: []xlsxDefinedName{d},
}
return nil
}
// DeleteDefinedName provides a function to delete the defined names of the
// workbook or worksheet. If not specified scope, the default scope is
// workbook. For example:
//
// f.DeleteDefinedName(&excelize.DefinedName{
// Name: "Amount",
// Scope: "Sheet2",
// })
//
func (f *File) DeleteDefinedName(definedName *DefinedName) error {
wb := f.workbookReader()
if wb.DefinedNames != nil {
for idx, dn := range wb.DefinedNames.DefinedName {
var scope string
if dn.LocalSheetID != nil {
scope = f.getSheetNameByID(*dn.LocalSheetID + 1)
}
if scope == definedName.Scope && dn.Name == definedName.Name {
wb.DefinedNames.DefinedName = append(wb.DefinedNames.DefinedName[:idx], wb.DefinedNames.DefinedName[idx+1:]...)
return nil
}
}
}
return errors.New("no defined name on the scope")
}
// GetDefinedName provides a function to get the defined names of the workbook
// or worksheet.
func (f *File) GetDefinedName() []DefinedName {
var definedNames []DefinedName
wb := f.workbookReader()
if wb.DefinedNames != nil {
for _, dn := range wb.DefinedNames.DefinedName {
definedName := DefinedName{
Name: dn.Name,
Comment: dn.Comment,
RefersTo: dn.Data,
Scope: "Workbook",
}
if dn.LocalSheetID != nil && *dn.LocalSheetID >= 0 {
definedName.Scope = f.getSheetNameByID(*dn.LocalSheetID + 1)
}
definedNames = append(definedNames, definedName)
}
}
return definedNames
}
// GroupSheets provides a function to group worksheets by given worksheets
// name. Group worksheets must contain an active worksheet.
func (f *File) GroupSheets(sheets []string) error {
// check an active worksheet in group worksheets
var inActiveSheet bool
activeSheet := f.GetActiveSheetIndex()
sheetMap := f.GetSheetList()
for idx, sheetName := range sheetMap {
for _, s := range sheets {
if s == sheetName && idx == activeSheet {
inActiveSheet = true
}
}
}
if !inActiveSheet {
return errors.New("group worksheet must contain an active worksheet")
}
// check worksheet exists
wss := []*xlsxWorksheet{}
for _, sheet := range sheets {
worksheet, err := f.workSheetReader(sheet)
if err != nil {
return err
}
wss = append(wss, worksheet)
}
for _, ws := range wss {
sheetViews := ws.SheetViews.SheetView
if len(sheetViews) > 0 {
for idx := range sheetViews {
ws.SheetViews.SheetView[idx].TabSelected = true
}
continue
}
}
return nil
}
// UngroupSheets provides a function to ungroup worksheets.
func (f *File) UngroupSheets() error {
activeSheet := f.GetActiveSheetIndex()
for index, sheet := range f.GetSheetList() {
if activeSheet == index {
continue
}
ws, _ := f.workSheetReader(sheet)
sheetViews := ws.SheetViews.SheetView
if len(sheetViews) > 0 {
for idx := range sheetViews {
ws.SheetViews.SheetView[idx].TabSelected = false
}
}
}
return nil
}
// InsertPageBreak create a page break to determine where the printed page
// ends and where begins the next one by given worksheet name and axis, so the
// content before the page break will be printed on one page and after the
// page break on another.
func (f *File) InsertPageBreak(sheet, cell string) (err error) {
var ws *xlsxWorksheet
var row, col int
var rowBrk, colBrk = -1, -1
if ws, err = f.workSheetReader(sheet); err != nil {
return
}
if col, row, err = CellNameToCoordinates(cell); err != nil {
return
}
col--
row--
if col == row && col == 0 {
return
}
if ws.RowBreaks == nil {
ws.RowBreaks = &xlsxBreaks{}
}
if ws.ColBreaks == nil {
ws.ColBreaks = &xlsxBreaks{}
}
for idx, brk := range ws.RowBreaks.Brk {
if brk.ID == row {
rowBrk = idx
}
}
for idx, brk := range ws.ColBreaks.Brk {
if brk.ID == col {
colBrk = idx
}
}
if row != 0 && rowBrk == -1 {
ws.RowBreaks.Brk = append(ws.RowBreaks.Brk, &xlsxBrk{
ID: row,
Max: 16383,
Man: true,
})
ws.RowBreaks.ManualBreakCount++
}
if col != 0 && colBrk == -1 {
ws.ColBreaks.Brk = append(ws.ColBreaks.Brk, &xlsxBrk{
ID: col,
Max: 1048575,
Man: true,
})
ws.ColBreaks.ManualBreakCount++
}
ws.RowBreaks.Count = len(ws.RowBreaks.Brk)
ws.ColBreaks.Count = len(ws.ColBreaks.Brk)
return
}
// RemovePageBreak remove a page break by given worksheet name and axis.
func (f *File) RemovePageBreak(sheet, cell string) (err error) {
var ws *xlsxWorksheet
var row, col int
if ws, err = f.workSheetReader(sheet); err != nil {
return
}
if col, row, err = CellNameToCoordinates(cell); err != nil {
return
}
col--
row--
if col == row && col == 0 {
return
}
removeBrk := func(ID int, brks []*xlsxBrk) []*xlsxBrk {
for i, brk := range brks {
if brk.ID == ID {
brks = append(brks[:i], brks[i+1:]...)
}
}
return brks
}
if ws.RowBreaks == nil || ws.ColBreaks == nil {
return
}
rowBrks := len(ws.RowBreaks.Brk)
colBrks := len(ws.ColBreaks.Brk)
if rowBrks > 0 && rowBrks == colBrks {
ws.RowBreaks.Brk = removeBrk(row, ws.RowBreaks.Brk)
ws.ColBreaks.Brk = removeBrk(col, ws.ColBreaks.Brk)
ws.RowBreaks.Count = len(ws.RowBreaks.Brk)
ws.ColBreaks.Count = len(ws.ColBreaks.Brk)
ws.RowBreaks.ManualBreakCount--
ws.ColBreaks.ManualBreakCount--
return
}
if rowBrks > 0 && rowBrks > colBrks {
ws.RowBreaks.Brk = removeBrk(row, ws.RowBreaks.Brk)
ws.RowBreaks.Count = len(ws.RowBreaks.Brk)
ws.RowBreaks.ManualBreakCount--
return
}
if colBrks > 0 && colBrks > rowBrks {
ws.ColBreaks.Brk = removeBrk(col, ws.ColBreaks.Brk)
ws.ColBreaks.Count = len(ws.ColBreaks.Brk)
ws.ColBreaks.ManualBreakCount--
}
return
}
// relsReader provides a function to get the pointer to the structure
// after deserialization of xl/worksheets/_rels/sheet%d.xml.rels.
func (f *File) relsReader(path string) *xlsxRelationships {
var err error
if f.Relationships[path] == nil {
_, ok := f.XLSX[path]
if ok {
c := xlsxRelationships{}
if err = f.xmlNewDecoder(bytes.NewReader(namespaceStrictToTransitional(f.readXML(path)))).
Decode(&c); err != nil && err != io.EOF {
log.Printf("xml decode error: %s", err)
}
f.Relationships[path] = &c
}
}
return f.Relationships[path]
}
// fillSheetData ensures there are enough rows, and columns in the chosen
// row to accept data. Missing rows are backfilled and given their row number
// Uses the last populated row as a hint for the size of the next row to add
func prepareSheetXML(ws *xlsxWorksheet, col int, row int) {
rowCount := len(ws.SheetData.Row)
sizeHint := 0
var ht float64
var customHeight bool
if ws.SheetFormatPr != nil && ws.SheetFormatPr.CustomHeight {
ht = ws.SheetFormatPr.DefaultRowHeight
customHeight = true
}
if rowCount > 0 {
sizeHint = len(ws.SheetData.Row[rowCount-1].C)
}
if rowCount < row {
// append missing rows
for rowIdx := rowCount; rowIdx < row; rowIdx++ {
ws.SheetData.Row = append(ws.SheetData.Row, xlsxRow{R: rowIdx + 1, CustomHeight: customHeight, Ht: ht, C: make([]xlsxC, 0, sizeHint)})
}
}
rowData := &ws.SheetData.Row[row-1]
fillColumns(rowData, col, row)
}
// fillColumns fill cells in the column of the row as contiguous.
func fillColumns(rowData *xlsxRow, col, row int) {
cellCount := len(rowData.C)
if cellCount < col {
for colIdx := cellCount; colIdx < col; colIdx++ {
cellName, _ := CoordinatesToCellName(colIdx+1, row)
rowData.C = append(rowData.C, xlsxC{R: cellName})
}
}
}
// makeContiguousColumns make columns in specific rows as contiguous.
func makeContiguousColumns(ws *xlsxWorksheet, fromRow, toRow, colCount int) {
for ; fromRow < toRow; fromRow++ {
rowData := &ws.SheetData.Row[fromRow-1]
fillColumns(rowData, colCount, fromRow)
}
}
马建仓 AI 助手
尝试更多
代码解读
代码找茬
代码优化
Go
1
https://gitee.com/chewel/excelize.git
git@gitee.com:chewel/excelize.git
chewel
excelize
excelize
master

搜索帮助

344bd9b3 5694891 D2dac590 5694891