Raw row access for increased performance.

[Splizard]

At the moment, any sort of filter operation where rows are (iterated in any order) and then deleted is very slow.
This is mainly due to the fact that the entire Row slice is copied for each RemoveRow operation.

xlsx.SheetData.Row = append(xlsx.SheetData.Row[:rowIdx],
	xlsx.SheetData.Row[rowIdx+1:]...)[:len(xlsx.SheetData.Row)-1]

Modifying the Row slice directly (ie filtering in-place) is fast. However there is no public interface to access this slice, it is accessed by the private method workSheetReader.

Could workSheetReader be made public? I realize the R value needs to be updated to reflect the row number and this could be confusing to users of the API but this could be documented and I think the performance gains are worth it.

Alternatively could a 'Rows' struct be created with methods for slicing, appending 'Row' objects?
Then you can check that the rows have proper R values.

Here is an example interface, (the types can just be a struct).

type Row interface {
    SetValue(column int, value interface{}) error
    SetColumns([]string) error
    Columns() []string
}

type Rows interface {
    Replace(int, Row)
    Index(int) Row
    Append(Row) Rows
    AppendRows(Rows) Rows
    Slice(i, j int) Rows
}

I can knock something together and create a pull request if you like.

[xuri]

Hi @Splizard, thanks for your issue. I propose deleting a row by filter in-place in the RemoveRow function and hold workSheetReader unexported.