perf: Optimize Chart control performance with reduced allocations and improved algorithmic complexity

[PaulAndersonS]

Root Cause of the Issue

Several hot paths in the Chart control used LINQ queries, List.Contains(), List.IndexOf(), and repeated GetType().Name reflection calls that introduced unnecessary allocations and O(n²) complexity in data-intensive rendering paths.

Description of Change

10 targeted micro-optimizations across 5 files to reduce allocations and improve algorithmic complexity in the Chart control:

1. CategoryAxis.cs — O(n²) IndexOf → Dictionary lookup
   GroupData() called List.IndexOf() inside a LINQ select over every data point, making it O(n²). Replaced with a pre-built Dictionary<string, int> for O(1) index lookups, and used pre-allocated List<double> instead of LINQ materialization.

2. CategoryAxis.cs — List.Contains() → HashSet
   The deduplication loop in GroupData() used List.Contains() (O(n)) on every iteration. Replaced with a HashSet<string> seeded from existing values for O(1) membership checks.

3. CartesianAxisLayout.cs — LINQ ToList() for single-item lookup → simple for loop
   GetAxisByName() created a full materialized List just to return the first match. Replaced with an early-return for loop that avoids all allocation and stops at the first match.

4. CartesianChartArea.cs — Repeated GetType().Name reflection → cached result
   CalculateStackingValues() called GetType().Name.Contains(...) twice per data point inside nested loops (once per if branch). Cached the type name and the boolean result before the inner loop to eliminate repeated reflection.

5. ErrorBarSegment.cs — 4× LINQ Where/Min/Max → single pass loops
   Computing xMin/xMax/yMin/yMax required 4 separate LINQ chains, each iterating the full collection. Replaced with 2 for loops that compute both min and max in a single pass per axis, with NaN fallback to 0.

6. CartesianChartArea.cs — LINQ Where().Sum() → for loop in GetYValue
GetYValue() used a LINQ chain with closure allocation on every call. Replaced with a plain for loop over the pre-typed List<StackingSeriesBase>.

7. CartesianSeries.cs — LINQ-generated index sequences → pre-allocated List<double>
GetXValues() used (from val in list select (xIndexValues++)).ToList() to produce sequential [0, 1, 2, …] index lists, allocating an iterator and boxing closures. Replaced with a new List<double>(count) filled with a for loop, also removing the unused xIndexValues variable.

8. ErrorBarSegment.cs — LINQ Where/Select/Min/Max for YRange → single loop
   The percentage-mode YRange calculation enumerated _topPointCollection and _bottomPointCollection 6 times (.Where, .Select, .Any, .Min, .Max × 2 collections). Replaced with two for loops that find the overall min/max in a single pass with a hasValid guard.

9. ErrorBarSegment.cs — Eliminate list allocations in GetSdErrorValue
GetSdErrorValue() allocated a filtered List, then two more List<double> (dev, sQDev) just to accumulate a sum of squares. Replaced with two inline loops that compute sum, count, and sumSqDev without any heap allocations.

10. CartesianChartArea.cs — Values.ToList() inside loops → cache before loop
    UpdateSBS() and GetTotalWidth() called SideBySideSeriesPosition.Values.ToList() on every loop iteration, creating a fresh List snapshot each time. Moved the call outside the loop and iterated the cached list.

Issues Fixed

No specific issue — proactive performance improvements.

Screenshots

N/A — internal algorithmic changes with no visible behavior difference.