SemanticEdit API changes: Obsolete PreserveLocalVariables, add RuntimeRudeEdits

[tmat]

Background and Motivation

Whether or not to preserve local variables is determined by presence of SyntaxMap. PreserveLocalVariables is superfluous. This proposed API change removes the parameter from the constructor. Implemented in #70449.

In addition, to support new lambda editing capabilities we propose to add RuntimeRudeEdit type and a parameter to SemanticEdit. Implemented and used in PR #70418.

Proposed API

class SemanticEdit
{
+  [Obsolete("Use other overload")]
+  [EditorBrowsable(EditorBrowsableState.Never)]
   public SemanticEdit(
     SemanticEditKind kind,
     ISymbol? oldSymbol,
     ISymbol? newSymbol,
     Func<SyntaxNode, SyntaxNode?>? syntaxMap,
     bool preserveLocalVariables,
     MethodInstrumentation instrumentation)

+  public SemanticEdit(
+    SemanticEditKind kind,
+    ISymbol? oldSymbol,
+    ISymbol? newSymbol,
+    Func<SyntaxNode, SyntaxNode?>? syntaxMap = null,
+    Func<SyntaxNode, RuntimeRudeEdit?>? runtimeRudeEdit = null,
+    MethodInstrumentation instrumentation = default)

+  /// <summary>
+  /// Specifies runtime rude edit information associated with a given syntax node, if any.
+  /// </summary>
+  public Func<SyntaxNode, RuntimeRudeEdit?>? RuntimeRudeEdit { get; }
}

+ /// <summary>
+ /// Describes rude edit to be reported at runtime.
+ /// </summary>
+ /// <param name="message">Error message.</param>
+ public readonly struct RuntimeRudeEdit(string message)
+ {
+     public string Message { get; } = message;
+ }

[333fred]

API Review

• Should we put all the optional state into a struct so we don't have to keep obsoleting old things?
• preserveLocalVariables wasn't being used
• What happens when runtimeRudeEdit is null?
  • Can't work in a fallback with lambdas
  • Don't have a clear answer at the moment, need that to understand what we should do with both old constructors.
• Should runtimeRudeEdit have more info in the callback?
  • It's already a struct, we can add more info in the future as needed

Needs work to understand what the fallback behavior (if any) is and how that should affect the existing backcompat overloaded constructor. We'll try to do this over email.

[tmat]

Implemented by #70418