UtilityAnalyzer: Move to a syntax based classification of identifiers in the token type utility analyzer

[martin-strecker-sonarsource]

Blocked by #7289, #7368

The token type analyzer calls TokenClassifierBase.ClassifyIdentifier for each identifier.

sonar-dotnet/analyzers/src/SonarAnalyzer.Common/Rules/Utilities/TokenTypeAnalyzerBase.cs

Lines 140 to 154 in 2728acd

	private TokenInfo ClassifyIdentifier(SyntaxToken token)
	{
	if (semanticModel.GetDeclaredSymbol(token.Parent) is { } declaration)
	{
	return ClassifyIdentifier(token, declaration);
	}
	else if (GetBindableParent(token) is { } parent && semanticModel.GetSymbolInfo(parent).Symbol is { } symbol)
	{
	return ClassifyIdentifier(token, symbol);
	}
	else
	{
	return null;
	}
	}

This method calls semanticModel.GetDeclaredSymbol(token.Parent) and optionally semanticModel.GetSymbolInfo(token.Parent) for the identifier token. Therefore an ISymbol is created, and a mapping from SyntaxNode to ISymbol is added to the semantic model. This adds a lot of pressure to any shared semantic model, as the ISymbol and the mapping need to be cached by the semantic model in a thread-safe manner. The code snippet below shows how many identifiers are present in a simple code snippet.

using System;                     // +1
using System.Collections.Generic; // +3

namespace A.B.C;                  // +3

public class D                    // +1
{
    public D()                    // +1
    {
    }
    public void M()               // +1
    {
        List<D> myList;           // +3
    }
}

TokenClassifierBase.ClassifyIdentifier can only have two outcomes:

• The identifier is considered TokenType.TypeName (with some special casing for types that are classified as keywords)
• The identifier is considered to be unknown.

This classification can often be done on a syntactical level. In the sample above, all identifiers can be classified without querying the semantic model, saving 20 calls to the semantic model (13 identifiers, where 6 are declarations -> 13 + 7) and the allocation of 11 symbols.

To do a proper classification on the syntax level, the test infrastructure needs to be extended and made more powerful. #7289 describes how to do it, and #7108 implements this infrastructure. Therefore, this issue is blocked by #7289

Related:
#4217
#7288
#6674

[martin-strecker-sonarsource]

Closed as fixed by #7788 and #7775