Reduce func allocations

[TheTonttu]

Fixes

• Improves Intermediate heap allocations #2725
• Eliminates major sources of Func<,> allocations originating from LINQ lambda outer variable captures by replacing them with equivalent loops.*

* Measured in most likely skewed scenario described in the "Data collection scenario" section.

Proposed Changes/Todos

• Refactor LineCanvas to reduce majority of Func allocations
  • LineCanvas.All(...)
  • LineCanvas.GetCellMap(...)
  • LineCanvas.Has(...)
• Refactor Region to reduce majority of Func allocations
  • Region.Contains(...)
  • Region.Intersect(...)
• Refactor PosAlign.CalculateMinDimension(...) to reduce majority of Func allocations
• Refactor TextFormatter.GetRuneWidth(...) to reduce majority of Func allocations

Pull Request checklist

• I've named my PR in the form of "Fixes #issue. Terse description."
• My code follows the style guidelines of Terminal.Gui - if you use Visual Studio, hit CTRL-K-D to automatically reformat your files before committing.
• My code follows the Terminal.Gui library design guidelines
• I ran dotnet test before commit
• I have made corresponding changes to the API documentation (using /// style comments)
• My changes generate no new warnings
• I have checked my code and corrected any poor grammar or misspellings
• I conducted basic QA to assure all features are working

Disclaimer

I like LINQ and the convenience it offers and use it most of time over loops when appropriate. Unfortunately when used in application hot paths the heap allocations can get out of hand due to hidden allocations from captured outer variables.

Data collection scenario

Run in Windows Terminal v1.21.10351.0 120x30 Command Prompt tab.

1. Start UICatalog.
2. Navigate down the Categories list using arrow key down.
   • Pause for a while (~1 sec) between key presses so that the Scenarios view has time to load properly.
3. Once at the bottom of the list navigate up using the arrow key up in similar manner until back at the top of the Categories list.
4. Quit application.
5. ???
6. Profit

This does not necessarily represent a typical application use and probably skews the statistics but at least reveals some allocation issues when using similar list navigation.

Ideally there would be some sort of automation that emulates more typical application.

Object allocation statistics

Data collected in Release build using Visual Studio Performance Profiler .NET Object Allocation Tracking.

Before

After

As a bonus, also list and enumerator allocations decreased slightly.

The main focus Func<,> dropped way down the allocation list.

There are still some unnecessary Func<,> allocations but the PR is already getting quite bloated. Also some of them are not worth the effort unless of course profiling other parts of the UICatalog tells otherwise. I nope'd out quite quickly after seeing DimAuto.Calculate(...). 😆

Refactors

LineCanvas.All

LINQ All replaced with equivalent foreach loop.

LineCanvas.GetCellMap

Replaced array allocation inside nested loops + LINQ lambdas with List buffer outside the nested. If IntersectionDefinition would be struct then stackalloc/ArrayPool would have made more sense.

Maybe the biggest change due to ReadOnlySpan replacing arrays in the methods further down the line, which forced to replace LINQ lambdas from those methods as well. GetRuneForIntersects(...) being the biggest "victim" of this change. Tried to keep it readable with local methods.

LineCanvas.Has

Similar change as in LineCanvas.All(...).

Region.Contains

Replaced LINQ Any with foreach loop.

Region.Intersect

Replaced list allocation and LINQ lambdas with foreach loop utilizing stackalloc buffer or pooled array depending on the amount of region rectangles. The original rectangles list is cleared and reused instead of creating new list every time. The list can still of course internally create new array if capacity is too small.

PosAlign.CalculateMinDimension

Group list and sum list allocations with multiple loops replaced with single loop combining the group check and summing.

TextFormatter.GetRuneWidth

Removed intermediate list allocation and just looped through the StringRuneEnumerator and summing the widths.

[tig]

With:

Without:

2 sec improvement. :-)