fix: corrected C# tree-sitter query

[mubeen-zulfiqar]

Related GitHub Issue

Closes: #5238 #6048

Description

Problem Summary

Roo-Code was failing to process and embed C# files despite having complete infrastructure support for C# language parsing. C# files were being detected but not properly indexed, resulting in missing code embeddings.

Root Cause Analysis

Issue Location

File: src/services/tree-sitter/queries/c-sharp.ts

Problem: Incorrect tree-sitter query patterns using malformed syntax that captured only node names instead of full AST definitions.

Technical Details

How Roo-Code Processes Files

1. Language Detection: src/services/code-index/processors/file-watcher.ts detects file types

2. Tree-sitter Parsing: src/services/tree-sitter/languageParser.ts loads language parsers

3. AST Query Execution: src/services/code-index/processors/parser.ts executes queries

4. Embedding Generation: Captured code blocks are sent for embedding

Failure Point

In src/services/code-index/processors/parser.ts, the processing logic:

 
const captures = query.captures(tree.rootNode);
 
console.log(`[DEBUG] Query captures: ${captures.length}`);
 
if (captures.length === 0) {
 
    console.log('[DEBUG] No captures found - language may be unsupported');
 
    // Falls back to basic text processing
 
}
 

Result: C# queries returned captures.length === 0, triggering "unsupported language" fallback behavior.

Broken Query Patterns

Original (Incorrect):

 
// BROKEN: Only captures node names as text, not full AST nodes
 
queries.push(`(class_declaration name: (identifier) @name.definition.class)`);
 
queries.push(`(method_declaration name: (identifier) @name.definition.method)`);
 
queries.push(`(property_declaration name: (identifier) @name.definition.property)`);
 

Problem: The syntax @name.definition.class attempts to capture nested attributes, but tree-sitter expects separate capture names.

Query Pattern Syntax Explanation

Why We Use Separate Captures:

• ✅ CORRECT: (node_type name: (identifier) @name) @definition.type
• ❌ INCORRECT: (node_type name: (identifier) @name.definition.type)

Technical Reason:
The correct syntax uses separate captures (@name) @definition.type) to capture both:

1. @name - captures the identifier text for the name
2. @definition.type - captures the full AST node for the definition

The incorrect nested syntax (@name.definition.type) attempts to create nested capture attributes which tree-sitter does not support, resulting in zero captures and causing the "unsupported language" fallback behavior.

Future Contributors: Always use separate captures to prevent regressions. This pattern is consistent across all working language queries (Python, JavaScript, etc.).

Fix Applied

Updated C# Query Patterns

Fixed Syntax:

 
// Classes
 
queries.push(`(class_declaration name: (identifier) @name) @definition.class`);
 
// Methods  
 
queries.push(`(method_declaration name: (identifier) @name) @definition.method`);
 
// Properties
 
queries.push(`(property_declaration name: (identifier) @name) @definition.property`);
 
// Events
 
queries.push(`(event_field_declaration (variable_declarator name: (identifier) @name)) @definition.event`);
 
// Delegates
 
queries.push(`(delegate_declaration name: (identifier) @name) @definition.delegate`);
 
// Namespaces (including file-scoped)
 
queries.push(`(namespace_declaration name: (qualified_name) @name) @definition.namespace`);
 
queries.push(`(file_scoped_namespace_declaration name: (qualified_name) @name) @definition.namespace`);
 
// Interfaces
 
queries.push(`(interface_declaration name: (identifier) @name) @definition.interface`);
 
// Structs
 
queries.push(`(struct_declaration name: (identifier) @name) @definition.struct`);
 
// Enums
 
queries.push(`(enum_declaration name: (identifier) @name) @definition.enum`);
 
 

Key Changes

1. Separated Captures: Changed from @name.definition.type to @name) @definition.type

2. Proper Node Targeting: Each query now captures both the identifier name and the full AST node

3. Consistent Syntax: Matches the working pattern used by Python, JavaScript, and other supported languages

4. Qualified Namespace Support: Uses (qualified_name) instead of (identifier) to capture nested namespaces like MyCompany.MyProduct.MyModule

5. File-Scoped Namespace Support: Added support for C# 10+ file-scoped namespace declarations

Verification Process

Debug Test Results

Before Fix:

• Query captures: 0

• Status: "language may be unsupported"

• Embedding: Fallback to full file content

After Fix:

• Query captures: >0 (depends on C# file content)

• Status: Proper semantic parsing

• Embedding: Individual code blocks (classes, methods, properties, etc.)

Test Procedure

This update has been tested on a project containing several hundred C# files, delivering excellent results. An example test is shown below.

Test File Used

 
// test-math-functions.cs
 
using System;
 
namespace MathFunctions
 
{
 
    public static class MathFunctions1
 
    {
 
        public static long Fibonacci(int n)
 
        {
 
            if (n <= 0) return 0;
 
            if (n == 1) return 1;
 
            long a = 0, b = 1;
 
            for (int i = 2; i <= n; i++)
 
            {
 
                long temp = a + b;
 
                a = b;
 
                b = temp;
 
            }
 
            return b;
 
        }
 
        public static double AreaOfTriangle(double baseLength, double height)
 
        {
 
            if (baseLength <= 0 || height <= 0)
 
                throw new ArgumentException("Base and height must be positive values");
 
            return 0.5 * baseLength * height;
 
        }
 
        public static long Factorial(int n)
 
        {
 
            if (n < 0)
 
                throw new ArgumentException("Factorial is not defined for negative numbers");
 
            if (n == 0 || n == 1) return 1;
 
            long result = 1;
 
            for (int i = 2; i <= n; i++)
 
            {
 
                result *= i;
 
            }
 
            return result;
 
        }
 
}
 

Build Result

• Extension: bin/roo-cline-3.27.0.vsix (27.22 MB)

• Download: Download the VSIX

• Status: Successfully built with C# fixes

• Files Updated: (src/services/tree-sitter/queries/c-sharp.ts)

Impact

• C# Support: Now fully functional with proper semantic parsing

• Other Languages: Unaffected (already working correctly)

• Performance: No degradation - same processing pipeline

• Compatibility: Maintains all existing Roo-Code functionality

Get in Touch

Discord: mubeen_zulfiqar

Email: [email protected]

---

Important

Fixes C# tree-sitter query patterns in c-sharp.ts to enable proper AST node capturing and embedding generation.

• Behavior:
  • Fixes C# tree-sitter query patterns in c-sharp.ts to correctly capture AST nodes.
  • Resolves issue where C# files were detected but not indexed, leading to missing embeddings.
• Technical Details:
  • Changes capture syntax from @name.definition.type to @name) @definition.type.
  • Supports qualified and file-scoped namespaces, classes, methods, properties, events, delegates, interfaces, structs, enums, records, attributes, type parameters, and LINQ expressions.
• Verification:
  • Debugging shows query captures now return >0, enabling proper semantic parsing and embedding of C# code blocks.

^{This description was created by for aa9213b. You can customize this summary. It will automatically update as commits are pushed.}

[roomote]

Thank you for your contribution! I've reviewed the changes and the fix correctly addresses the C# indexing issues reported in #5238 and #6048. The query syntax change aligns with tree-sitter's expected format and all tests pass. I have some suggestions for improvement below.

[mubeen-zulfiqar]

@mrubens @jr @cte @hannesrudolph I wanted to start by saying that Roo Code is a huge contribution to open source—its versatility and widespread usage are truly inspiring, and I’m proud to contribute.

I’ve just submitted PR #7813, “fix: corrected C# tree-sitter query,” which resolves the issue where C# files were detected but not properly indexed, resulting in fallback to basic text processing.

Key improvements:
• Root cause: malformed capture syntax (@name.definition.type) resulted in zero AST captures.
• Fix: separated captures properly (for example, (identifier) @name) @definition.class), bringing C# indexing in line with other supported languages .
• Now supports a broad range of C# constructs—classes, methods, properties, events, delegates, namespaces (including file-scoped), interfaces, structs, enums, and more .
• Tests on several hundred C# files now show positive capture counts, enabling semantic parsing and embedding without regressions or performance issues .

Automated Review Feedback: roomote[bot] has already reviewed the change and confirmed that it successfully addresses the indexing issues described in issue #5238 and #6048 .

Could you please take a look when you have a moment? I’d appreciate any feedback or approval so we can merge this and restore full C# support.

Thanks, and keep up the amazing work with Roo Code!

Best regards,
@mubeen-zulfiqar

[adamhill]

@mubeen-zulfiqar This is so awesome! I had a strange feeling about CBI being worse than originally. And turns out it was because 95% of my codebases are C#.

Thank you! ❤️

@KJ7LNW Do you think we need to vet all the queries for at least basic parsing. I dont understand how this slipped by with all the corpus tests there are in the most t-s project directories.

[mubeen-zulfiqar]

@adamhill Thanks a lot! 🙌 Seeing someone hit that issue in real life makes this work worthwhile — this definitely helps folks whose projects are heavily C#. I appreciate you sharing that feedback.

Really glad to be contributing alongside you on Roo Code!

[KJ7LNW]

@KJ7LNW Do you think we need to vet all the queries for at least basic parsing. I dont understand how this slipped by with all the corpus tests there are in the most t-s project directories.

I am not familiar enough with C# to check this directly, but make sure that src/services/tree-sitter/__tests__/parseSourceCodeDefinitions.c-sharp.spec.ts has a test for every item the you plan to validate to prevent future regressions

[daniel-lxs]

Thank you @mubeen-zulfiqar!

[mubeen-zulfiqar]

@mrubens @daniel-lxs
I truly appreciate Roo-Code’s open source work and feel honored to contribute. Thank you! 🙏