Performance of creating new cells

[trykyn]

When creating an excel sheet with millions of cells and thousands of cells per row I noticed in the profiler that the majority of time is spent in a small function inside NPOI: XSSFRow.LastCellNum

Given that a SortedDictionary is used, it should be possible to find the last key in O(log(n)) time instead of O(n)
Unfortunately the available implementation of SortedDictionary doesn't have Min, Max even though the underlying SortedSet does.. but it would be possible to either roll your own which could exposes Min/Max
or more hacky but easy, access it with reflection..
In constructor of XSSFRow:
this._set = typeof(SortedDictionary<int, ICell>).GetField("_set", System.Reflection.BindingFlags.NonPublic | System.Reflection.BindingFlags.Instance).GetValue(this._cells) as SortedSet<KeyValuePair<int,ICell>>;

Then In GetFirstKey/GetLastKey use _set.Max.Key or _set.Min.Key

[trykyn]

For performance, I did some benchmarks and with 1000 cells the reflection solution is 10 times faster, with 10000 cells it is more than 100 times faster

[tonyqus]

Do you have a workable version on your box? We do accept PR.

[delreluca]

If SortedDictionary<int,ICell> in XSSFRow is only used internally (no serialization and no exposing outside) rolling your own is quite easy, we can copy and rename the official one (it is MIT-licensed) and expose the missing properties of the set:

https://github.com/dotnet/runtime/blob/5ca4992150e66a7ac6d48c567f905c24a73e7919/src/libraries/System.Collections/src/System/Collections/Generic/SortedDictionary.cs#L21

Is this approach OK, also from a license perspective? If so, I might find some time to create a PR.

[tonyqus]

Thank you for the tip. I will test if it has any side effect on existing logic.