Bläddra i källkod

Implement consistent row addressing by Excel row number starting with 1 (#350)

* Implement consistent row addressing by Excel row number starting with 1

1. Added second versions for all row manipulation methods with zero-based row addressing.
2. Fixed methods documentation to explicitly describe which row addressing used in method.
3. Added WARNING to README.md.
4. Cosmetic change: All row test moved to file `rows_test.go`.

* TravisCI: go1.12 added to tests matrix

* BACKWARD INCOMPARTIBLE: Use only Excel numbering logic from 1 row

* README updated
Veniamin Albaev 6 år sedan
förälder
incheckning
12c1e2481e
5 ändrade filer med 573 tillägg och 471 borttagningar
  1. 1 0
      .travis.yml
  2. 5 0
      README.md
  3. 0 424
      excelize_test.go
  4. 60 47
      rows.go
  5. 507 0
      rows_test.go

+ 1 - 0
.travis.yml

@@ -8,6 +8,7 @@ go:
   - 1.9.x
   - 1.10.x
   - 1.11.x
+  - 1.12.x
 
 os:
   - linux

+ 5 - 0
README.md

@@ -15,6 +15,11 @@
 
 Excelize is a library written in pure Go and providing a set of functions that allow you to write to and read from XLSX files. Support reads and writes XLSX file generated by Microsoft Excel™ 2007 and later. Support save file without losing original charts of XLSX. This library needs Go version 1.8 or later. The full API docs can be seen using go's built-in documentation tool, or online at [godoc.org](https://godoc.org/github.com/360EntSecGroup-Skylar/excelize) and [docs reference](https://xuri.me/excelize/).
 
+**WARNING!**
+
+From version 1.5 all row manipulation methods uses Excel row numbering starting with `1` instead of zero-based numbering
+which take place in some methods in eraler versions.   
+
 ## Basic Usage
 
 ### Installation

+ 0 - 424
excelize_test.go

@@ -287,19 +287,6 @@ func TestColWidth(t *testing.T) {
 	convertRowHeightToPixels(0)
 }
 
-func TestRowHeight(t *testing.T) {
-	xlsx := NewFile()
-	xlsx.SetRowHeight("Sheet1", 1, 50)
-	xlsx.SetRowHeight("Sheet1", 4, 90)
-	t.Log(xlsx.GetRowHeight("Sheet1", 1))
-	t.Log(xlsx.GetRowHeight("Sheet1", 0))
-	err := xlsx.SaveAs(filepath.Join("test", "TestRowHeight.xlsx"))
-	if !assert.NoError(t, err) {
-		t.FailNow()
-	}
-	convertColWidthToPixels(0)
-}
-
 func TestSetCellHyperLink(t *testing.T) {
 	xlsx, err := OpenFile(filepath.Join("test", "Book1.xlsx"))
 	if err != nil {
@@ -1112,335 +1099,6 @@ func TestRemoveCol(t *testing.T) {
 	assert.NoError(t, xlsx.SaveAs(filepath.Join("test", "TestRemoveCol.xlsx")))
 }
 
-func TestInsertRow(t *testing.T) {
-	xlsx := NewFile()
-	for j := 1; j <= 10; j++ {
-		for i := 0; i <= 10; i++ {
-			axis := ToAlphaString(i) + strconv.Itoa(j)
-			xlsx.SetCellStr("Sheet1", axis, axis)
-		}
-	}
-	xlsx.SetCellHyperLink("Sheet1", "A5", "https://github.com/360EntSecGroup-Skylar/excelize", "External")
-	xlsx.InsertRow("Sheet1", -1)
-	xlsx.InsertRow("Sheet1", 4)
-
-	assert.NoError(t, xlsx.SaveAs(filepath.Join("test", "TestInsertRow.xlsx")))
-}
-
-// Testing internal sructure state after insert operations.
-// It is important for insert workflow to be constant to avoid side effect with functions related to internal structure.
-func TestInsertRowInEmptyFile(t *testing.T) {
-	xlsx := NewFile()
-	sheet1 := xlsx.GetSheetName(1)
-	r := xlsx.workSheetReader(sheet1)
-	xlsx.InsertRow(sheet1, 0)
-	assert.Len(t, r.SheetData.Row, 0)
-	xlsx.InsertRow(sheet1, 1)
-	assert.Len(t, r.SheetData.Row, 0)
-	xlsx.InsertRow(sheet1, 99)
-	assert.Len(t, r.SheetData.Row, 0)
-	assert.NoError(t, xlsx.SaveAs(filepath.Join("test", "TestInsertRowInEmptyFile.xlsx")))
-}
-
-func TestDuplicateRow(t *testing.T) {
-	const sheet = "Sheet1"
-	outFile := filepath.Join("test", "TestDuplicateRow.%s.xlsx")
-
-	cells := map[string]string{
-		"A1": "A1 Value",
-		"A2": "A2 Value",
-		"A3": "A3 Value",
-		"B1": "B1 Value",
-		"B2": "B2 Value",
-		"B3": "B3 Value",
-	}
-
-	newFileWithDefaults := func() *File {
-		f := NewFile()
-		for cell, val := range cells {
-			f.SetCellStr(sheet, cell, val)
-
-		}
-		return f
-	}
-
-	t.Run("FromSingleRow", func(t *testing.T) {
-		xlsx := NewFile()
-		xlsx.SetCellStr(sheet, "A1", cells["A1"])
-		xlsx.SetCellStr(sheet, "B1", cells["B1"])
-
-		xlsx.DuplicateRow(sheet, 1)
-		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.FromSingleRow_1"))) {
-			t.FailNow()
-		}
-		expect := map[string]string{
-			"A1": cells["A1"], "B1": cells["B1"],
-			"A2": cells["A1"], "B2": cells["B1"],
-		}
-		for cell, val := range expect {
-			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
-				t.FailNow()
-			}
-		}
-
-		xlsx.DuplicateRow(sheet, 2)
-		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.FromSingleRow_2"))) {
-			t.FailNow()
-		}
-		expect = map[string]string{
-			"A1": cells["A1"], "B1": cells["B1"],
-			"A2": cells["A1"], "B2": cells["B1"],
-			"A3": cells["A1"], "B3": cells["B1"],
-		}
-		for cell, val := range expect {
-			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
-				t.FailNow()
-			}
-		}
-	})
-
-	t.Run("UpdateDuplicatedRows", func(t *testing.T) {
-		xlsx := NewFile()
-		xlsx.SetCellStr(sheet, "A1", cells["A1"])
-		xlsx.SetCellStr(sheet, "B1", cells["B1"])
-
-		xlsx.DuplicateRow(sheet, 1)
-
-		xlsx.SetCellStr(sheet, "A2", cells["A2"])
-		xlsx.SetCellStr(sheet, "B2", cells["B2"])
-
-		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.UpdateDuplicatedRows"))) {
-			t.FailNow()
-		}
-		expect := map[string]string{
-			"A1": cells["A1"], "B1": cells["B1"],
-			"A2": cells["A2"], "B2": cells["B2"],
-		}
-		for cell, val := range expect {
-			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
-				t.FailNow()
-			}
-		}
-	})
-
-	t.Run("FirstOfMultipleRows", func(t *testing.T) {
-		xlsx := newFileWithDefaults()
-
-		xlsx.DuplicateRow(sheet, 1)
-
-		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.FirstOfMultipleRows"))) {
-			t.FailNow()
-		}
-		expect := map[string]string{
-			"A1": cells["A1"], "B1": cells["B1"],
-			"A2": cells["A1"], "B2": cells["B1"],
-			"A3": cells["A2"], "B3": cells["B2"],
-			"A4": cells["A3"], "B4": cells["B3"],
-		}
-		for cell, val := range expect {
-			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
-				t.FailNow()
-			}
-		}
-	})
-
-	t.Run("ZeroWithNoRows", func(t *testing.T) {
-		xlsx := NewFile()
-
-		xlsx.DuplicateRow(sheet, 0)
-
-		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.ZeroWithNoRows"))) {
-			t.FailNow()
-		}
-		assert.Equal(t, "", xlsx.GetCellValue(sheet, "A1"))
-		assert.Equal(t, "", xlsx.GetCellValue(sheet, "B1"))
-		assert.Equal(t, "", xlsx.GetCellValue(sheet, "A2"))
-		assert.Equal(t, "", xlsx.GetCellValue(sheet, "B2"))
-		expect := map[string]string{
-			"A1": "", "B1": "",
-			"A2": "", "B2": "",
-		}
-		for cell, val := range expect {
-			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
-				t.FailNow()
-			}
-		}
-	})
-
-	t.Run("MiddleRowOfEmptyFile", func(t *testing.T) {
-		xlsx := NewFile()
-
-		xlsx.DuplicateRow(sheet, 99)
-
-		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.MiddleRowOfEmptyFile"))) {
-			t.FailNow()
-		}
-		expect := map[string]string{
-			"A98":  "",
-			"A99":  "",
-			"A100": "",
-		}
-		for cell, val := range expect {
-			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
-				t.FailNow()
-			}
-		}
-	})
-
-	t.Run("WithLargeOffsetToMiddleOfData", func(t *testing.T) {
-		xlsx := newFileWithDefaults()
-
-		xlsx.DuplicateRowTo(sheet, 1, 3)
-
-		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.WithLargeOffsetToMiddleOfData"))) {
-			t.FailNow()
-		}
-		expect := map[string]string{
-			"A1": cells["A1"], "B1": cells["B1"],
-			"A2": cells["A2"], "B2": cells["B2"],
-			"A3": cells["A1"], "B3": cells["B1"],
-			"A4": cells["A3"], "B4": cells["B3"],
-		}
-		for cell, val := range expect {
-			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
-				t.FailNow()
-			}
-		}
-	})
-
-	t.Run("WithLargeOffsetToEmptyRows", func(t *testing.T) {
-		xlsx := newFileWithDefaults()
-
-		xlsx.DuplicateRowTo(sheet, 1, 7)
-
-		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.WithLargeOffsetToEmptyRows"))) {
-			t.FailNow()
-		}
-		expect := map[string]string{
-			"A1": cells["A1"], "B1": cells["B1"],
-			"A2": cells["A2"], "B2": cells["B2"],
-			"A3": cells["A3"], "B3": cells["B3"],
-			"A7": cells["A1"], "B7": cells["B1"],
-		}
-		for cell, val := range expect {
-			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
-				t.FailNow()
-			}
-		}
-	})
-
-	t.Run("InsertBefore", func(t *testing.T) {
-		xlsx := newFileWithDefaults()
-
-		xlsx.DuplicateRowTo(sheet, 2, 1)
-
-		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.InsertBefore"))) {
-			t.FailNow()
-		}
-
-		expect := map[string]string{
-			"A1": cells["A2"], "B1": cells["B2"],
-			"A2": cells["A1"], "B2": cells["B1"],
-			"A3": cells["A2"], "B3": cells["B2"],
-			"A4": cells["A3"], "B4": cells["B3"],
-		}
-		for cell, val := range expect {
-			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
-				t.FailNow()
-			}
-		}
-	})
-
-	t.Run("InsertBeforeWithLargeOffset", func(t *testing.T) {
-		xlsx := newFileWithDefaults()
-
-		xlsx.DuplicateRowTo(sheet, 3, 1)
-
-		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.InsertBeforeWithLargeOffset"))) {
-			t.FailNow()
-		}
-
-		expect := map[string]string{
-			"A1": cells["A3"], "B1": cells["B3"],
-			"A2": cells["A1"], "B2": cells["B1"],
-			"A3": cells["A2"], "B3": cells["B2"],
-			"A4": cells["A3"], "B4": cells["B3"],
-		}
-		for cell, val := range expect {
-			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell)) {
-				t.FailNow()
-			}
-		}
-	})
-}
-
-func TestDuplicateRowInvalidRownum(t *testing.T) {
-	const sheet = "Sheet1"
-	outFile := filepath.Join("test", "TestDuplicateRowInvalidRownum.%s.xlsx")
-
-	cells := map[string]string{
-		"A1": "A1 Value",
-		"A2": "A2 Value",
-		"A3": "A3 Value",
-		"B1": "B1 Value",
-		"B2": "B2 Value",
-		"B3": "B3 Value",
-	}
-
-	testRows := []int{-2, -1}
-
-	testRowPairs := []struct {
-		row1 int
-		row2 int
-	}{
-		{-1, -1},
-		{-1, 0},
-		{-1, 1},
-		{0, -1},
-		{0, 0},
-		{0, 1},
-		{1, -1},
-		{1, 1},
-		{1, 0},
-	}
-
-	for i, row := range testRows {
-		name := fmt.Sprintf("TestRow_%d", i+1)
-		t.Run(name, func(t *testing.T) {
-			xlsx := NewFile()
-			for col, val := range cells {
-				xlsx.SetCellStr(sheet, col, val)
-			}
-			xlsx.DuplicateRow(sheet, row)
-
-			for col, val := range cells {
-				if !assert.Equal(t, val, xlsx.GetCellValue(sheet, col)) {
-					t.FailNow()
-				}
-			}
-			assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, name)))
-		})
-	}
-
-	for i, pair := range testRowPairs {
-		name := fmt.Sprintf("TestRowPair_%d", i+1)
-		t.Run(name, func(t *testing.T) {
-			xlsx := NewFile()
-			for col, val := range cells {
-				xlsx.SetCellStr(sheet, col, val)
-			}
-			xlsx.DuplicateRowTo(sheet, pair.row1, pair.row2)
-
-			for col, val := range cells {
-				if !assert.Equal(t, val, xlsx.GetCellValue(sheet, col)) {
-					t.FailNow()
-				}
-			}
-			assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, name)))
-		})
-	}
-}
-
 func TestSetPane(t *testing.T) {
 	xlsx := NewFile()
 	xlsx.SetPanes("Sheet1", `{"freeze":false,"split":false}`)
@@ -1455,33 +1113,6 @@ func TestSetPane(t *testing.T) {
 	assert.NoError(t, xlsx.SaveAs(filepath.Join("test", "TestSetPane.xlsx")))
 }
 
-func TestRemoveRow(t *testing.T) {
-	xlsx := NewFile()
-	for j := 1; j <= 10; j++ {
-		for i := 0; i <= 10; i++ {
-			axis := ToAlphaString(i) + strconv.Itoa(j)
-			xlsx.SetCellStr("Sheet1", axis, axis)
-		}
-	}
-	xlsx.SetCellHyperLink("Sheet1", "A5", "https://github.com/360EntSecGroup-Skylar/excelize", "External")
-	xlsx.RemoveRow("Sheet1", -1)
-	xlsx.RemoveRow("Sheet1", 4)
-	xlsx.MergeCell("Sheet1", "B3", "B5")
-	xlsx.RemoveRow("Sheet1", 2)
-	xlsx.RemoveRow("Sheet1", 4)
-
-	err := xlsx.AutoFilter("Sheet1", "A2", "A2", `{"column":"A","expression":"x != blanks"}`)
-	if !assert.NoError(t, err) {
-		t.FailNow()
-	}
-
-	xlsx.RemoveRow("Sheet1", 0)
-	xlsx.RemoveRow("Sheet1", 1)
-	xlsx.RemoveRow("Sheet1", 0)
-
-	assert.NoError(t, xlsx.SaveAs(filepath.Join("test", "TestRemoveRow.xlsx")))
-}
-
 func TestConditionalFormat(t *testing.T) {
 	xlsx := NewFile()
 	for j := 1; j <= 10; j++ {
@@ -1604,50 +1235,6 @@ func TestSetSheetRow(t *testing.T) {
 	assert.NoError(t, xlsx.SaveAs(filepath.Join("test", "TestSetSheetRow.xlsx")))
 }
 
-func TestRows(t *testing.T) {
-	xlsx, err := OpenFile(filepath.Join("test", "Book1.xlsx"))
-	if !assert.NoError(t, err) {
-		t.FailNow()
-	}
-
-	rows, err := xlsx.Rows("Sheet2")
-	if !assert.NoError(t, err) {
-		t.FailNow()
-	}
-
-	rowStrs := make([][]string, 0)
-	var i = 0
-	for rows.Next() {
-		i++
-		columns := rows.Columns()
-		rowStrs = append(rowStrs, columns)
-	}
-
-	if !assert.NoError(t, rows.Error()) {
-		t.FailNow()
-	}
-
-	dstRows := xlsx.GetRows("Sheet2")
-	if !assert.Equal(t, len(rowStrs), len(dstRows)) {
-		t.FailNow()
-	}
-
-	for i := 0; i < len(rowStrs); i++ {
-		if !assert.Equal(t, trimSliceSpace(dstRows[i]), trimSliceSpace(rowStrs[i])) {
-			t.FailNow()
-		}
-	}
-}
-
-func TestRowsError(t *testing.T) {
-	xlsx, err := OpenFile(filepath.Join("test", "Book1.xlsx"))
-	if !assert.NoError(t, err) {
-		t.FailNow()
-	}
-	_, err = xlsx.Rows("SheetN")
-	assert.EqualError(t, err, "Sheet SheetN is not exist")
-}
-
 func TestOutlineLevel(t *testing.T) {
 	xlsx := NewFile()
 	xlsx.NewSheet("Sheet2")
@@ -1733,17 +1320,6 @@ func TestUnprotectSheet(t *testing.T) {
 	assert.NoError(t, xlsx.SaveAs(filepath.Join("test", "TestUnprotectSheet.xlsx")))
 }
 
-func trimSliceSpace(s []string) []string {
-	for {
-		if len(s) > 0 && s[len(s)-1] == "" {
-			s = s[:len(s)-1]
-		} else {
-			break
-		}
-	}
-	return s
-}
-
 func prepareTestBook1() (*File, error) {
 	xlsx, err := OpenFile(filepath.Join("test", "Book1.xlsx"))
 	if err != nil {

+ 60 - 47
rows.go

@@ -203,6 +203,9 @@ func (f *File) getTotalRowsCols(name string) (int, int) {
 //
 func (f *File) SetRowHeight(sheet string, row int, height float64) {
 	xlsx := f.workSheetReader(sheet)
+	if row < 1 {
+		return
+	}
 	cells := 0
 	rowIdx := row - 1
 	completeRow(xlsx, row, cells)
@@ -230,6 +233,9 @@ func (f *File) getRowHeight(sheet string, row int) int {
 //
 func (f *File) GetRowHeight(sheet string, row int) float64 {
 	xlsx := f.workSheetReader(sheet)
+	if row < 1 || row > len(xlsx.SheetData.Row) {
+		return defaultRowHeightPixels // it will be better to use 0, but we take care with BC
+	}
 	for _, v := range xlsx.SheetData.Row {
 		if v.R == row && v.Ht != 0 {
 			return v.Ht
@@ -279,80 +285,88 @@ func (xlsx *xlsxC) getValueFrom(f *File, d *xlsxSST) (string, error) {
 	}
 }
 
-// SetRowVisible provides a function to set visible of a single row by given
-// worksheet name and row index. For example, hide row 2 in Sheet1:
+// SetRowVisible2 provides a function to set visible of a single row by given
+// worksheet name and Excel row number. For example, hide row 2 in Sheet1:
 //
 //    xlsx.SetRowVisible("Sheet1", 2, false)
 //
-func (f *File) SetRowVisible(sheet string, rowIndex int, visible bool) {
+func (f *File) SetRowVisible(sheet string, row int, visible bool) {
 	xlsx := f.workSheetReader(sheet)
-	rows := rowIndex + 1
+	if row < 1 {
+		return
+	}
 	cells := 0
-	completeRow(xlsx, rows, cells)
+	completeRow(xlsx, row, cells)
+	rowIdx := row - 1
 	if visible {
-		xlsx.SheetData.Row[rowIndex].Hidden = false
+		xlsx.SheetData.Row[rowIdx].Hidden = false
 		return
 	}
-	xlsx.SheetData.Row[rowIndex].Hidden = true
+	xlsx.SheetData.Row[rowIdx].Hidden = true
 }
 
-// GetRowVisible provides a function to get visible of a single row by given
-// worksheet name and row index. For example, get visible state of row 2 in
+// GetRowVisible2 provides a function to get visible of a single row by given
+// worksheet name and Excel row number.
+// For example, get visible state of row 2 in
 // Sheet1:
 //
 //    xlsx.GetRowVisible("Sheet1", 2)
 //
-func (f *File) GetRowVisible(sheet string, rowIndex int) bool {
+func (f *File) GetRowVisible(sheet string, row int) bool {
 	xlsx := f.workSheetReader(sheet)
-	rows := rowIndex + 1
+	if row < 1 || row > len(xlsx.SheetData.Row) {
+		return false
+	}
+	rowIndex := row - 1
 	cells := 0
-	completeRow(xlsx, rows, cells)
+	completeRow(xlsx, row, cells)
 	return !xlsx.SheetData.Row[rowIndex].Hidden
 }
 
 // SetRowOutlineLevel provides a function to set outline level number of a
-// single row by given worksheet name and row index. For example, outline row
+// single row by given worksheet name and Excel row number. For example, outline row
 // 2 in Sheet1 to level 1:
 //
 //    xlsx.SetRowOutlineLevel("Sheet1", 2, 1)
 //
-func (f *File) SetRowOutlineLevel(sheet string, rowIndex int, level uint8) {
+func (f *File) SetRowOutlineLevel(sheet string, row int, level uint8) {
 	xlsx := f.workSheetReader(sheet)
-	rows := rowIndex + 1
+	if row < 1 {
+		return
+	}
 	cells := 0
-	completeRow(xlsx, rows, cells)
-	xlsx.SheetData.Row[rowIndex].OutlineLevel = level
+	completeRow(xlsx, row, cells)
+	xlsx.SheetData.Row[row-1].OutlineLevel = level
 }
 
 // GetRowOutlineLevel provides a function to get outline level number of a
-// single row by given worksheet name and row index. For example, get outline
-// number of row 2 in Sheet1:
+// single row by given worksheet name and Exce row number.
+// For example, get outline number of row 2 in Sheet1:
 //
 //    xlsx.GetRowOutlineLevel("Sheet1", 2)
 //
-func (f *File) GetRowOutlineLevel(sheet string, rowIndex int) uint8 {
+func (f *File) GetRowOutlineLevel(sheet string, row int) uint8 {
 	xlsx := f.workSheetReader(sheet)
-	rows := rowIndex + 1
-	cells := 0
-	completeRow(xlsx, rows, cells)
-	return xlsx.SheetData.Row[rowIndex].OutlineLevel
+	if row < 1 || row > len(xlsx.SheetData.Row) {
+		return 0
+	}
+	return xlsx.SheetData.Row[row-1].OutlineLevel
 }
 
-// RemoveRow provides a function to remove single row by given worksheet name
-// and row index. For example, remove row 3 in Sheet1:
+// RemoveRow2 provides a function to remove single row by given worksheet name
+// and Excel row number. For example, remove row 3 in Sheet1:
 //
-//    xlsx.RemoveRow("Sheet1", 2)
+//    xlsx.RemoveRow("Sheet1", 3)
 //
 // 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
 // worksheet, it will cause a file error when you open it. The excelize only
 // partially updates these references currently.
 func (f *File) RemoveRow(sheet string, row int) {
-	if row < 0 {
+	xlsx := f.workSheetReader(sheet)
+	if row < 1 || row > len(xlsx.SheetData.Row) {
 		return
 	}
-	xlsx := f.workSheetReader(sheet)
-	row++
 	for i, r := range xlsx.SheetData.Row {
 		if r.R == row {
 			xlsx.SheetData.Row = append(xlsx.SheetData.Row[:i], xlsx.SheetData.Row[i+1:]...)
@@ -362,20 +376,19 @@ func (f *File) RemoveRow(sheet string, row int) {
 	}
 }
 
-// InsertRow provides a function to insert a new row after given row index.
-// For example, create a new row before row 3 in Sheet1:
+// InsertRow2 provides a function to insert a new row after given Excel row number
+// starting from 1. For example, create a new row before row 3 in Sheet1:
 //
-//    xlsx.InsertRow("Sheet1", 2)
+//    xlsx.InsertRow("Sheet1", 3)
 //
 func (f *File) InsertRow(sheet string, row int) {
-	if row < 0 {
+	if row < 1 {
 		return
 	}
-	row++
 	f.adjustHelper(sheet, -1, row, 1)
 }
 
-// DuplicateRow inserts a copy of specified row below specified
+// DuplicateRow inserts a copy of specified row (by it Excel row number) below
 //
 //    xlsx.DuplicateRow("Sheet1", 2)
 //
@@ -387,8 +400,8 @@ func (f *File) DuplicateRow(sheet string, row int) {
 	f.DuplicateRowTo(sheet, row, row+1)
 }
 
-// DuplicateRowTo inserts a copy of specified row at specified row position
-// moving down exists rows after target position
+// DuplicateRowTo inserts a copy of specified row by it Excel number
+// to specified row position moving down exists rows after target position
 //
 //    xlsx.DuplicateRowTo("Sheet1", 2, 7)
 //
@@ -397,18 +410,18 @@ func (f *File) DuplicateRow(sheet string, row int) {
 // worksheet, it will cause a file error when you open it. The excelize only
 // partially updates these references currently.
 func (f *File) DuplicateRowTo(sheet string, row, row2 int) {
-	if row <= 0 || row2 <= 0 || row == row2 {
+	xlsx := f.workSheetReader(sheet)
+
+	if row < 1 || row > len(xlsx.SheetData.Row) || row2 < 1 || row == row2 {
 		return
 	}
 
-	ws := f.workSheetReader(sheet)
-
 	var ok bool
 	var rowCopy xlsxRow
 
-	for i, r := range ws.SheetData.Row {
+	for i, r := range xlsx.SheetData.Row {
 		if r.R == row {
-			rowCopy = ws.SheetData.Row[i]
+			rowCopy = xlsx.SheetData.Row[i]
 			ok = true
 			break
 		}
@@ -420,13 +433,13 @@ func (f *File) DuplicateRowTo(sheet string, row, row2 int) {
 	f.adjustHelper(sheet, -1, row2, 1)
 
 	idx2 := -1
-	for i, r := range ws.SheetData.Row {
+	for i, r := range xlsx.SheetData.Row {
 		if r.R == row2 {
 			idx2 = i
 			break
 		}
 	}
-	if idx2 == -1 && len(ws.SheetData.Row) >= row2 {
+	if idx2 == -1 && len(xlsx.SheetData.Row) >= row2 {
 		return
 	}
 
@@ -434,9 +447,9 @@ func (f *File) DuplicateRowTo(sheet string, row, row2 int) {
 	f.ajustSingleRowDimensions(&rowCopy, row2)
 
 	if idx2 != -1 {
-		ws.SheetData.Row[idx2] = rowCopy
+		xlsx.SheetData.Row[idx2] = rowCopy
 	} else {
-		ws.SheetData.Row = append(ws.SheetData.Row, rowCopy)
+		xlsx.SheetData.Row = append(xlsx.SheetData.Row, rowCopy)
 	}
 }
 

+ 507 - 0
rows_test.go

@@ -0,0 +1,507 @@
+package excelize
+
+import (
+	"fmt"
+	"path/filepath"
+	"strconv"
+	"testing"
+
+	"github.com/stretchr/testify/assert"
+)
+
+func TestRows(t *testing.T) {
+	xlsx, err := OpenFile(filepath.Join("test", "Book1.xlsx"))
+	if !assert.NoError(t, err) {
+		t.FailNow()
+	}
+
+	rows, err := xlsx.Rows("Sheet2")
+	if !assert.NoError(t, err) {
+		t.FailNow()
+	}
+
+	rowStrs := make([][]string, 0)
+	var i = 0
+	for rows.Next() {
+		i++
+		columns := rows.Columns()
+		rowStrs = append(rowStrs, columns)
+	}
+
+	if !assert.NoError(t, rows.Error()) {
+		t.FailNow()
+	}
+
+	dstRows := xlsx.GetRows("Sheet2")
+	if !assert.Equal(t, len(rowStrs), len(dstRows)) {
+		t.FailNow()
+	}
+
+	for i := 0; i < len(rowStrs); i++ {
+		if !assert.Equal(t, trimSliceSpace(dstRows[i]), trimSliceSpace(rowStrs[i])) {
+			t.FailNow()
+		}
+	}
+}
+
+func TestRowsError(t *testing.T) {
+	xlsx, err := OpenFile(filepath.Join("test", "Book1.xlsx"))
+	if !assert.NoError(t, err) {
+		t.FailNow()
+	}
+	_, err = xlsx.Rows("SheetN")
+	assert.EqualError(t, err, "Sheet SheetN is not exist")
+}
+
+func TestRowHeight(t *testing.T) {
+	xlsx := NewFile()
+	sheet1 := xlsx.GetSheetName(1)
+
+	xlsx.SetRowHeight(sheet1, 0, defaultRowHeightPixels+1.0) // should no effect
+	assert.Equal(t, defaultRowHeightPixels, xlsx.GetRowHeight("Sheet1", 0))
+
+	xlsx.SetRowHeight(sheet1, 1, 111.0)
+	assert.Equal(t, 111.0, xlsx.GetRowHeight(sheet1, 1))
+
+	xlsx.SetRowHeight(sheet1, 4, 444.0)
+	assert.Equal(t, 444.0, xlsx.GetRowHeight(sheet1, 4))
+
+	err := xlsx.SaveAs(filepath.Join("test", "TestRowHeight.xlsx"))
+	if !assert.NoError(t, err) {
+		t.FailNow()
+	}
+
+	convertColWidthToPixels(0)
+}
+
+func TestRemoveRow(t *testing.T) {
+	xlsx := NewFile()
+	sheet1 := xlsx.GetSheetName(1)
+	r := xlsx.workSheetReader(sheet1)
+
+	const (
+		cellCount = 10
+		rowCount  = 10
+	)
+	for j := 1; j <= cellCount; j++ {
+		for i := 1; i <= rowCount; i++ {
+			axis := ToAlphaString(i) + strconv.Itoa(j)
+			xlsx.SetCellStr(sheet1, axis, axis)
+		}
+	}
+	xlsx.SetCellHyperLink(sheet1, "A5", "https://github.com/360EntSecGroup-Skylar/excelize", "External")
+
+	xlsx.RemoveRow(sheet1, -1)
+	if !assert.Len(t, r.SheetData.Row, rowCount) {
+		t.FailNow()
+	}
+
+	xlsx.RemoveRow(sheet1, 0)
+	if !assert.Len(t, r.SheetData.Row, rowCount) {
+		t.FailNow()
+	}
+
+	xlsx.RemoveRow(sheet1, 4)
+	if !assert.Len(t, r.SheetData.Row, rowCount-1) {
+		t.FailNow()
+	}
+
+	xlsx.MergeCell(sheet1, "B3", "B5")
+
+	xlsx.RemoveRow(sheet1, 2)
+	if !assert.Len(t, r.SheetData.Row, rowCount-2) {
+		t.FailNow()
+	}
+
+	xlsx.RemoveRow(sheet1, 4)
+	if !assert.Len(t, r.SheetData.Row, rowCount-3) {
+		t.FailNow()
+	}
+
+	err := xlsx.AutoFilter(sheet1, "A2", "A2", `{"column":"A","expression":"x != blanks"}`)
+	if !assert.NoError(t, err) {
+		t.FailNow()
+	}
+
+	xlsx.RemoveRow(sheet1, 1)
+	if !assert.Len(t, r.SheetData.Row, rowCount-4) {
+		t.FailNow()
+	}
+
+	xlsx.RemoveRow(sheet1, 2)
+	if !assert.Len(t, r.SheetData.Row, rowCount-5) {
+		t.FailNow()
+	}
+
+	xlsx.RemoveRow(sheet1, 1)
+	if !assert.Len(t, r.SheetData.Row, rowCount-6) {
+		t.FailNow()
+	}
+
+	assert.NoError(t, xlsx.SaveAs(filepath.Join("test", "TestRemoveRow.xlsx")))
+}
+
+func TestInsertRow(t *testing.T) {
+	xlsx := NewFile()
+	sheet1 := xlsx.GetSheetName(1)
+	r := xlsx.workSheetReader(sheet1)
+
+	const (
+		cellCount = 10
+		rowCount  = 10
+	)
+	for j := 1; j <= cellCount; j++ {
+		for i := 1; i < rowCount; i++ {
+			axis := ToAlphaString(i) + strconv.Itoa(j)
+			xlsx.SetCellStr(sheet1, axis, axis)
+		}
+	}
+	xlsx.SetCellHyperLink(sheet1, "A5", "https://github.com/360EntSecGroup-Skylar/excelize", "External")
+
+	xlsx.InsertRow(sheet1, -1)
+	if !assert.Len(t, r.SheetData.Row, rowCount) {
+		t.FailNow()
+	}
+
+	xlsx.InsertRow(sheet1, 0)
+	if !assert.Len(t, r.SheetData.Row, rowCount) {
+		t.FailNow()
+	}
+
+	xlsx.InsertRow(sheet1, 1)
+	if !assert.Len(t, r.SheetData.Row, rowCount+1) {
+		t.FailNow()
+	}
+
+	xlsx.InsertRow(sheet1, 4)
+	if !assert.Len(t, r.SheetData.Row, rowCount+2) {
+		t.FailNow()
+	}
+
+	assert.NoError(t, xlsx.SaveAs(filepath.Join("test", "TestInsertRow.xlsx")))
+}
+
+// Testing internal sructure state after insert operations.
+// It is important for insert workflow to be constant to avoid side effect with functions related to internal structure.
+func TestInsertRowInEmptyFile(t *testing.T) {
+	xlsx := NewFile()
+	sheet1 := xlsx.GetSheetName(1)
+	r := xlsx.workSheetReader(sheet1)
+	xlsx.InsertRow(sheet1, 1)
+	assert.Len(t, r.SheetData.Row, 0)
+	xlsx.InsertRow(sheet1, 2)
+	assert.Len(t, r.SheetData.Row, 0)
+	xlsx.InsertRow(sheet1, 99)
+	assert.Len(t, r.SheetData.Row, 0)
+	assert.NoError(t, xlsx.SaveAs(filepath.Join("test", "TestInsertRowInEmptyFile.xlsx")))
+}
+
+func TestDuplicateRow(t *testing.T) {
+	const sheet = "Sheet1"
+	outFile := filepath.Join("test", "TestDuplicateRow.%s.xlsx")
+
+	cells := map[string]string{
+		"A1": "A1 Value",
+		"A2": "A2 Value",
+		"A3": "A3 Value",
+		"B1": "B1 Value",
+		"B2": "B2 Value",
+		"B3": "B3 Value",
+	}
+
+	newFileWithDefaults := func() *File {
+		f := NewFile()
+		for cell, val := range cells {
+			f.SetCellStr(sheet, cell, val)
+
+		}
+		return f
+	}
+
+	t.Run("FromSingleRow", func(t *testing.T) {
+		xlsx := NewFile()
+		xlsx.SetCellStr(sheet, "A1", cells["A1"])
+		xlsx.SetCellStr(sheet, "B1", cells["B1"])
+
+		xlsx.DuplicateRow(sheet, 1)
+		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.FromSingleRow_1"))) {
+			t.FailNow()
+		}
+		expect := map[string]string{
+			"A1": cells["A1"], "B1": cells["B1"],
+			"A2": cells["A1"], "B2": cells["B1"],
+		}
+		for cell, val := range expect {
+			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
+				t.FailNow()
+			}
+		}
+
+		xlsx.DuplicateRow(sheet, 2)
+		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.FromSingleRow_2"))) {
+			t.FailNow()
+		}
+		expect = map[string]string{
+			"A1": cells["A1"], "B1": cells["B1"],
+			"A2": cells["A1"], "B2": cells["B1"],
+			"A3": cells["A1"], "B3": cells["B1"],
+		}
+		for cell, val := range expect {
+			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
+				t.FailNow()
+			}
+		}
+	})
+
+	t.Run("UpdateDuplicatedRows", func(t *testing.T) {
+		xlsx := NewFile()
+		xlsx.SetCellStr(sheet, "A1", cells["A1"])
+		xlsx.SetCellStr(sheet, "B1", cells["B1"])
+
+		xlsx.DuplicateRow(sheet, 1)
+
+		xlsx.SetCellStr(sheet, "A2", cells["A2"])
+		xlsx.SetCellStr(sheet, "B2", cells["B2"])
+
+		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.UpdateDuplicatedRows"))) {
+			t.FailNow()
+		}
+		expect := map[string]string{
+			"A1": cells["A1"], "B1": cells["B1"],
+			"A2": cells["A2"], "B2": cells["B2"],
+		}
+		for cell, val := range expect {
+			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
+				t.FailNow()
+			}
+		}
+	})
+
+	t.Run("FirstOfMultipleRows", func(t *testing.T) {
+		xlsx := newFileWithDefaults()
+
+		xlsx.DuplicateRow(sheet, 1)
+
+		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.FirstOfMultipleRows"))) {
+			t.FailNow()
+		}
+		expect := map[string]string{
+			"A1": cells["A1"], "B1": cells["B1"],
+			"A2": cells["A1"], "B2": cells["B1"],
+			"A3": cells["A2"], "B3": cells["B2"],
+			"A4": cells["A3"], "B4": cells["B3"],
+		}
+		for cell, val := range expect {
+			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
+				t.FailNow()
+			}
+		}
+	})
+
+	t.Run("ZeroWithNoRows", func(t *testing.T) {
+		xlsx := NewFile()
+
+		xlsx.DuplicateRow(sheet, 0)
+
+		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.ZeroWithNoRows"))) {
+			t.FailNow()
+		}
+		assert.Equal(t, "", xlsx.GetCellValue(sheet, "A1"))
+		assert.Equal(t, "", xlsx.GetCellValue(sheet, "B1"))
+		assert.Equal(t, "", xlsx.GetCellValue(sheet, "A2"))
+		assert.Equal(t, "", xlsx.GetCellValue(sheet, "B2"))
+		expect := map[string]string{
+			"A1": "", "B1": "",
+			"A2": "", "B2": "",
+		}
+		for cell, val := range expect {
+			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
+				t.FailNow()
+			}
+		}
+	})
+
+	t.Run("MiddleRowOfEmptyFile", func(t *testing.T) {
+		xlsx := NewFile()
+
+		xlsx.DuplicateRow(sheet, 99)
+
+		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.MiddleRowOfEmptyFile"))) {
+			t.FailNow()
+		}
+		expect := map[string]string{
+			"A98":  "",
+			"A99":  "",
+			"A100": "",
+		}
+		for cell, val := range expect {
+			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
+				t.FailNow()
+			}
+		}
+	})
+
+	t.Run("WithLargeOffsetToMiddleOfData", func(t *testing.T) {
+		xlsx := newFileWithDefaults()
+
+		xlsx.DuplicateRowTo(sheet, 1, 3)
+
+		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.WithLargeOffsetToMiddleOfData"))) {
+			t.FailNow()
+		}
+		expect := map[string]string{
+			"A1": cells["A1"], "B1": cells["B1"],
+			"A2": cells["A2"], "B2": cells["B2"],
+			"A3": cells["A1"], "B3": cells["B1"],
+			"A4": cells["A3"], "B4": cells["B3"],
+		}
+		for cell, val := range expect {
+			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
+				t.FailNow()
+			}
+		}
+	})
+
+	t.Run("WithLargeOffsetToEmptyRows", func(t *testing.T) {
+		xlsx := newFileWithDefaults()
+
+		xlsx.DuplicateRowTo(sheet, 1, 7)
+
+		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.WithLargeOffsetToEmptyRows"))) {
+			t.FailNow()
+		}
+		expect := map[string]string{
+			"A1": cells["A1"], "B1": cells["B1"],
+			"A2": cells["A2"], "B2": cells["B2"],
+			"A3": cells["A3"], "B3": cells["B3"],
+			"A7": cells["A1"], "B7": cells["B1"],
+		}
+		for cell, val := range expect {
+			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
+				t.FailNow()
+			}
+		}
+	})
+
+	t.Run("InsertBefore", func(t *testing.T) {
+		xlsx := newFileWithDefaults()
+
+		xlsx.DuplicateRowTo(sheet, 2, 1)
+
+		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.InsertBefore"))) {
+			t.FailNow()
+		}
+
+		expect := map[string]string{
+			"A1": cells["A2"], "B1": cells["B2"],
+			"A2": cells["A1"], "B2": cells["B1"],
+			"A3": cells["A2"], "B3": cells["B2"],
+			"A4": cells["A3"], "B4": cells["B3"],
+		}
+		for cell, val := range expect {
+			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell), cell) {
+				t.FailNow()
+			}
+		}
+	})
+
+	t.Run("InsertBeforeWithLargeOffset", func(t *testing.T) {
+		xlsx := newFileWithDefaults()
+
+		xlsx.DuplicateRowTo(sheet, 3, 1)
+
+		if !assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, "TestDuplicateRow.InsertBeforeWithLargeOffset"))) {
+			t.FailNow()
+		}
+
+		expect := map[string]string{
+			"A1": cells["A3"], "B1": cells["B3"],
+			"A2": cells["A1"], "B2": cells["B1"],
+			"A3": cells["A2"], "B3": cells["B2"],
+			"A4": cells["A3"], "B4": cells["B3"],
+		}
+		for cell, val := range expect {
+			if !assert.Equal(t, val, xlsx.GetCellValue(sheet, cell)) {
+				t.FailNow()
+			}
+		}
+	})
+}
+
+func TestDuplicateRowInvalidRownum(t *testing.T) {
+	const sheet = "Sheet1"
+	outFile := filepath.Join("test", "TestDuplicateRowInvalidRownum.%s.xlsx")
+
+	cells := map[string]string{
+		"A1": "A1 Value",
+		"A2": "A2 Value",
+		"A3": "A3 Value",
+		"B1": "B1 Value",
+		"B2": "B2 Value",
+		"B3": "B3 Value",
+	}
+
+	testRows := []int{-2, -1}
+
+	testRowPairs := []struct {
+		row1 int
+		row2 int
+	}{
+		{-1, -1},
+		{-1, 0},
+		{-1, 1},
+		{0, -1},
+		{0, 0},
+		{0, 1},
+		{1, -1},
+		{1, 1},
+		{1, 0},
+	}
+
+	for i, row := range testRows {
+		name := fmt.Sprintf("TestRow_%d", i+1)
+		t.Run(name, func(t *testing.T) {
+			xlsx := NewFile()
+			for col, val := range cells {
+				xlsx.SetCellStr(sheet, col, val)
+			}
+			xlsx.DuplicateRow(sheet, row)
+
+			for col, val := range cells {
+				if !assert.Equal(t, val, xlsx.GetCellValue(sheet, col)) {
+					t.FailNow()
+				}
+			}
+			assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, name)))
+		})
+	}
+
+	for i, pair := range testRowPairs {
+		name := fmt.Sprintf("TestRowPair_%d", i+1)
+		t.Run(name, func(t *testing.T) {
+			xlsx := NewFile()
+			for col, val := range cells {
+				xlsx.SetCellStr(sheet, col, val)
+			}
+			xlsx.DuplicateRowTo(sheet, pair.row1, pair.row2)
+
+			for col, val := range cells {
+				if !assert.Equal(t, val, xlsx.GetCellValue(sheet, col)) {
+					t.FailNow()
+				}
+			}
+			assert.NoError(t, xlsx.SaveAs(fmt.Sprintf(outFile, name)))
+		})
+	}
+}
+
+func trimSliceSpace(s []string) []string {
+	for {
+		if len(s) > 0 && s[len(s)-1] == "" {
+			s = s[:len(s)-1]
+		} else {
+			break
+		}
+	}
+	return s
+}