Fix Qwen3 content extraction breaking code formatting (#661)

Problem:
- qwen3::extract_content_during_parsing() used aggressive regex to collapse multiple newlines
- This broke proper code formatting (e.g., PEP 8's 2 empty lines between functions)
- Affected non-tool-call streaming output where formatting is critical

Solution:
- Replace aggressive std::regex_replace(R"(\n\s*\n)", "\n") with gentle string_strip()
- Follow original llama.cpp patterns: only trim leading/trailing whitespace
- Preserve internal formatting including multiple newlines
- Add proper include for common.h to access string_strip function

Changes:
- examples/server/parsers/qwen3_parser.hpp: Replace whitespace cleanup with string_strip()
- tests/test-function-calls.cpp: Add test_qwen3_whitespace_preservation() to prevent regression

Testing:
-  PEP 8 compliance: 2 empty lines between functions preserved
-  Tool call parsing: All Qwen3 tests continue to pass
-  No regressions: Existing functionality maintained
-  Follows original llama.cpp whitespace handling patterns
This commit is contained in:
Anton Sokolchenko
2025-08-07 07:22:01 +02:00
committed by GitHub
parent f4051d9c3e
commit 05a61510b9
2 changed files with 38 additions and 6 deletions

View File

@@ -1,6 +1,7 @@
#pragma once
#include "json.hpp"
#include "../../common/common.h"
#include <string>
#include <regex>
@@ -102,12 +103,8 @@ static std::string extract_content_during_parsing(const std::string& text, bool
}
}
// Clean up extra whitespace
content = std::regex_replace(content, std::regex(R"(\n\s*\n)"), "\n");
// Trim leading/trailing whitespace
content.erase(0, content.find_first_not_of(" \t\n\r"));
content.erase(content.find_last_not_of(" \t\n\r") + 1);
// Only trim leading/trailing whitespace, preserve internal formatting
content = string_strip(content);
} catch (const std::exception&) {
// Return original text on regex errors

View File

@@ -2237,6 +2237,40 @@ void test_xml_tool_call_parsing() {
std::cout << " ✅ XML tool call parsing works correctly!" << std::endl;
}
// Test whitespace preservation in qwen3 content extraction
void test_qwen3_whitespace_preservation() {
std::cout << "\n🧹 Testing Qwen3 Whitespace Preservation Fix:" << std::endl;
// Test case with PEP 8 style: 2 empty lines between functions
const std::string pep8_content = R"(def celsius_to_fahrenheit(celsius):
return celsius * 9/5 + 32
def fahrenheit_to_celsius(fahrenheit):
return (fahrenheit - 32) * 5/9)";
std::cout << "🎯 Testing PEP 8 compliance (2 empty lines between functions)..." << std::endl;
std::cout << "Original content has: 2 empty lines between functions" << std::endl;
// Test the qwen3 content extraction directly
std::string result = qwen3::extract_content_during_parsing(pep8_content, false);
// Check if the double newlines are preserved (should have \n\n\n for 2 empty lines)
bool has_double_empty_lines = result.find("\n\n\n") != std::string::npos;
std::cout << "Result content: '" << result << "'" << std::endl;
std::cout << "Has 2 empty lines preserved: " << (has_double_empty_lines ? "YES" : "NO") << std::endl;
test_assert(has_double_empty_lines, "Qwen3: PEP 8 double empty lines preserved");
// Additional test: ensure no excessive trimming
test_assert(!result.empty(), "Qwen3: Content not empty after processing");
test_assert(result.find("celsius_to_fahrenheit") != std::string::npos, "Qwen3: Function content preserved");
test_assert(result.find("fahrenheit_to_celsius") != std::string::npos, "Qwen3: Second function preserved");
std::cout << " ✅ Qwen3 whitespace preservation working correctly!" << std::endl;
}
// Test the streaming tool calls fix implementation
void test_streaming_tool_calls_fix() {
std::cout << "\n=== Streaming Tool Calls Fix Validation ===" << std::endl;
@@ -2797,6 +2831,7 @@ int main() {
test_content_cleaning();
test_contamination_reproduction(); // Added this test
test_mixed_formats();
test_qwen3_whitespace_preservation(); // Test whitespace fix
std::cout << "\n🌍 Unicode & International Tests:" << std::endl;
test_unicode_support();