File size: 13,624 Bytes
374f3ad
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
import re
import math
import lizard
import numpy as np
from collections import Counter


def strip_metadata_headers(code_text: str) -> str:
    """Xóa các khối header/metadata (ví dụ: JSON metadata) ở đầu file nếu có."""
    code_text = re.sub(r'^\s*\{.*?"code"\s*:\s*".*?"\s*\}\s*', '', code_text, flags=re.DOTALL)
    
    lines = code_text.split('\n')
    cleaned_lines = []
    header_passed = False
    for line in lines:
        if not header_passed:
            if line.strip().startswith('//') and any(kw in line.upper() for kw in ['DATASET', 'MODEL', 'GEMINI', 'GPT', 'CATEGORY']):
                continue
            if line.strip() != '':
                header_passed = True
        cleaned_lines.append(line)
    return "\n".join(cleaned_lines).strip()


class CppFeatureExtractorV8:
    """
    Trích xuất 44 đặc trưng tĩnh từ mã nguồn C++.
    Sử dụng regex và thư viện lizard để đánh giá complexity.
    """
    def __init__(self):
        self.feature_names = [
            'comment_ratio', 'empty_line_ratio', 'emoji_marker_score', 'avg_line_length', 'max_line_length',
            'tab_vs_space_ratio', 'trailing_space_ratio', 'brace_style_consistency', 'avg_identifier_length',
            'identifier_length_variance', 'single_char_var_ratio', 'unique_identifier_ratio', 'keyword_to_identifier_ratio',
            'avg_cyclomatic_complexity', 'num_functions', 'avg_function_loc', 'halstead_volume',
            'halstead_difficulty', 'halstead_effort', 'halstead_bugs', 'maintainability_index',
            'code_to_comment_ratio', 'max_nesting_depth', 'total_includes', 'has_bits_stdc',
            'macro_count', 'modern_cpp_ratio', 'const_usage_ratio', 'has_fast_io', 'newline_style_ratio',
            'shannon_entropy', 'bigram_entropy', 'whitespace_entropy', 'class_count', 'struct_count',
            'has_inheritance', 'access_specifier_ratio', 'virtual_override_ratio', 'using_std_ratio',
            'try_catch_ratio', 'raw_pointer_ratio', 'getter_setter_ratio', 'std_prefix_ratio',
            'cpp_cast_ratio'
        ]

        self.string_pattern = re.compile(r'".*?(?<!\\)"|\'.*?(?<!\\)\'')
        self.comment_line_pattern = re.compile(r'//.*')
        self.comment_block_pattern = re.compile(r'/\*.*?\*/', re.DOTALL)

        self.include_pattern = re.compile(r'#include\s*[<"].*[>"]')
        self.macro_pattern = re.compile(r'#define\s+')
        self.bits_stdc_pattern = re.compile(r'bits/stdc\+\+\.h')
        self.fast_io_pattern = re.compile(r'ios_base::sync_with_stdio|cin\.tie')

        self.modern_cpp_pattern = re.compile(r'\b(auto|nullptr|unique_ptr|shared_ptr|constexpr|lambda)\b')
        self.const_pattern = re.compile(r'\b(const|constexpr)\b')
        self.exception_pattern = re.compile(r'\b(try|catch|throw)\b')
        self.single_char_var_pattern = re.compile(r'\b[a-zA-Z]\b')
        
        self.emote_tokens = [
            ":)", ":-)", ":D", ":-D", ";)", ";-)", "<3",
            "^_^", "T_T", "xD", "XD", "UwU", "uwu", "O_O", ":3",
            "¯\\_(ツ)_/¯"
        ]

        self.operators_pattern = re.compile(r'(\+|-|\*|/|%|=|==|!=|<|>|<=|>=|&&|\|\||!|&|\||\^|~|<<|>>|\+\+|--|->|\.|::|\?|:)')
        self.keywords = {'int', 'void', 'if', 'else', 'while', 'for', 'return', 'class', 'public', 'private', 'struct', 'bool', 'char', 'float', 'double', 'std', 'cout', 'cin', 'endl', 'break', 'continue'}

        self.class_pattern = re.compile(r'\bclass\s+[a-zA-Z_]\w*')
        self.struct_pattern = re.compile(r'\bstruct\s+[a-zA-Z_]\w*')
        self.inheritance_pattern = re.compile(r'\bclass\s+[a-zA-Z_]\w*\s*:\s*(public|private|protected)\b')
        self.access_specifier_pattern = re.compile(r'\b(public|private|protected)\s*:')
        self.virtual_override_pattern = re.compile(r'\b(virtual|override)\b')
        self.using_std_pattern = re.compile(r'\busing\s+namespace\s+std\s*;')
        self.try_catch_pattern = re.compile(r'\b(try|catch)\b')
        self.raw_pointer_pattern = re.compile(r'\b(int|float|double|char|void|[a-zA-Z_]\w*)\s*\*+\s*[a-zA-Z_]\w*')
        self.getter_setter_pattern = re.compile(r'\b(get|set)[A-Z]\w*\b')
        self.cpp_cast_pattern = re.compile(r'\b(static_cast|dynamic_cast|reinterpret_cast|const_cast)\b')

    def calculate_entropy(self, text):
        if not text: return 0.0
        prob = [float(c) / len(text) for c in dict(Counter(text)).values()]
        return -sum(p * math.log2(p) for p in prob)

    def calculate_bigram_entropy(self, text):
        if len(text) < 2: return 0.0
        bigrams = [text[i:i+2] for i in range(len(text)-1)]
        prob = [float(c) / len(bigrams) for c in dict(Counter(bigrams)).values()]
        return -sum(p * math.log2(p) for p in prob)

    def count_emoji_markers(self, text):
        count = 0
        for char in text:
            codepoint = ord(char)
            if (0x1F300 <= codepoint <= 0x1FAFF) or (0x2600 <= codepoint <= 0x27BF):
                count += 1
        for token in self.emote_tokens:
            count += text.count(token)
        return count

    def score_emoji_markers(self, marker_count):
        if marker_count <= 0: return 0.0
        if marker_count == 1: return 0.65
        if marker_count == 2: return 0.9
        return 1.0

    def calculate_halstead_metrics(self, pure_code, identifiers):
        ops = self.operators_pattern.findall(pure_code)
        kw_found = [w for w in re.findall(r'\b[a-zA-Z_]\w*\b', pure_code) if w in self.keywords]
        all_operators = ops + kw_found

        N1 = len(all_operators)
        n1 = len(set(all_operators))

        strings = self.string_pattern.findall(pure_code)
        numbers = re.findall(r'\b\d+(\.\d+)?\b', pure_code)
        all_operands = identifiers + strings + numbers

        N2 = len(all_operands)
        n2 = len(set(all_operands))

        n = n1 + n2
        N = N1 + N2

        Volume = N * math.log2(n) if n > 0 else 0
        Difficulty = (n1 / 2) * (N2 / n2) if n2 > 0 else 0
        Effort = Difficulty * Volume
        Bugs = Volume / 3000

        return Volume, Difficulty, Effort, Bugs

    def extract(self, code_raw):
        features = {}
        lines = code_raw.split('\n')
        total_chars = len(code_raw)
        pure_lines = [l for l in lines if l.strip() and not l.strip().startswith('//')]
        total_loc = len(pure_lines) if len(pure_lines) > 0 else 1

        line_comments = self.comment_line_pattern.findall(code_raw)
        block_comments = self.comment_block_pattern.findall(code_raw)
        comments_text = "\n".join(line_comments + block_comments)
        string_literals = self.string_pattern.findall(code_raw)

        pure_code = self.string_pattern.sub('', code_raw)
        pure_code = self.comment_block_pattern.sub('', pure_code)
        pure_code = self.comment_line_pattern.sub('', pure_code)

        # --- [A] LAYOUT & FORMATTING ---
        features['comment_ratio'] = len(comments_text) / total_chars if total_chars > 0 else 0
        features['empty_line_ratio'] = sum(1 for line in lines if not line.strip()) / max(1, len(lines))
        marker_text = comments_text + "\n" + "\n".join(string_literals)
        emoji_marker_count = self.count_emoji_markers(marker_text)
        features['emoji_marker_score'] = self.score_emoji_markers(emoji_marker_count)

        line_lengths = [len(l.strip()) for l in pure_lines]
        features['avg_line_length'] = np.mean(line_lengths) if line_lengths else 0
        features['max_line_length'] = np.max(line_lengths) if line_lengths else 0

        spaces = code_raw.count(' ')
        tabs = code_raw.count('\t')
        features['tab_vs_space_ratio'] = tabs / (spaces + tabs) if (spaces + tabs) > 0 else 0
        features['trailing_space_ratio'] = sum(1 for l in lines if l.endswith(' ') or l.endswith('\t')) / max(1, len(lines))

        kr_brace = len(re.findall(r'\S\s*\{', pure_code))
        allman_brace = len(re.findall(r'^\s*\{', pure_code, re.MULTILINE))
        total_braces = kr_brace + allman_brace
        allman_ratio = allman_brace / total_braces if total_braces > 0 else 0.5
        features['brace_style_consistency'] = abs(allman_ratio - 0.5) * 2

        # --- [B] NAMING CONVENTIONS ---
        identifiers = re.findall(r'\b[a-zA-Z_][a-zA-Z0-9_]*\b', pure_code)
        custom_ids = [w for w in identifiers if w not in self.keywords]

        if custom_ids:
            id_lengths = [len(w) for w in custom_ids]
            features['avg_identifier_length'] = np.mean(id_lengths)
            features['identifier_length_variance'] = np.var(id_lengths)
            features['single_char_var_ratio'] = len(self.single_char_var_pattern.findall(pure_code)) / len(custom_ids)
            features['unique_identifier_ratio'] = len(set(custom_ids)) / len(custom_ids)
        else:
            features['avg_identifier_length'] = 0
            features['identifier_length_variance'] = 0
            features['single_char_var_ratio'] = 0
            features['unique_identifier_ratio'] = 0

        features['keyword_to_identifier_ratio'] = len([w for w in identifiers if w in self.keywords]) / max(1, len(custom_ids))

        # --- [C] STRUCTURAL COMPLEXITY ---
        try:
            analysis = lizard.analyze_source_code("test.cpp", code_raw)
            if analysis.function_list:
                cc_list = [f.cyclomatic_complexity for f in analysis.function_list]
                features['avg_cyclomatic_complexity'] = np.mean(cc_list)
                features['num_functions'] = len(analysis.function_list)
                features['avg_function_loc'] = np.mean([f.end_line - f.start_line for f in analysis.function_list])
            else:
                features['avg_cyclomatic_complexity'] = 1.0
                features['num_functions'] = 0
                features['avg_function_loc'] = 0
        except Exception:
            features['avg_cyclomatic_complexity'] = 1.0
            features['num_functions'] = 0
            features['avg_function_loc'] = 0

        V, D, E, B = self.calculate_halstead_metrics(pure_code, custom_ids)
        features['halstead_volume'] = V
        features['halstead_difficulty'] = D
        features['halstead_effort'] = E
        features['halstead_bugs'] = B

        MI = 171 - 5.2 * math.log(max(1, V)) - 0.23 * features['avg_cyclomatic_complexity'] - 16.2 * math.log(max(1, total_loc))
        features['maintainability_index'] = max(0, MI)
        features['code_to_comment_ratio'] = total_loc / max(1, len(comments_text.split('\n')))

        depth, max_depth = 0, 0
        for char in pure_code:
            if char == '{': depth += 1; max_depth = max(max_depth, depth)
            elif char == '}': depth = max(0, depth - 1)
        features['max_nesting_depth'] = max_depth

        # --- [D] CODING HABITS & IDIOMS ---
        features['total_includes'] = len(self.include_pattern.findall(code_raw))
        features['has_bits_stdc'] = 1 if self.bits_stdc_pattern.search(code_raw) else 0
        features['macro_count'] = len(self.macro_pattern.findall(code_raw))

        total_words = len(re.findall(r'\b\w+\b', pure_code))
        features['modern_cpp_ratio'] = len(self.modern_cpp_pattern.findall(pure_code)) / max(1, total_words)
        features['const_usage_ratio'] = len(self.const_pattern.findall(pure_code)) / max(1, total_words)

        features['has_fast_io'] = 1 if self.fast_io_pattern.search(pure_code) else 0

        count_endl = pure_code.count('endl')
        count_n = pure_code.count('\\n')
        features['newline_style_ratio'] = count_n / max(1, (count_n + count_endl))

        # --- [E] INFORMATION THEORY ---
        features['shannon_entropy'] = self.calculate_entropy(pure_code)
        features['bigram_entropy'] = self.calculate_bigram_entropy(pure_code)

        spaces_pattern = "".join(['S' if c == ' ' else 'T' if c == '\t' else 'N' if c == '\n' else '' for c in code_raw])
        features['whitespace_entropy'] = self.calculate_entropy(spaces_pattern)

        # --- [F] NEW OOP & ADVANCED CODING HABITS ---
        class_cnt = len(self.class_pattern.findall(pure_code))
        features['class_count'] = class_cnt
        features['struct_count'] = len(self.struct_pattern.findall(pure_code))
        features['has_inheritance'] = 1 if self.inheritance_pattern.search(pure_code) else 0
        
        access_spec_cnt = len(self.access_specifier_pattern.findall(pure_code))
        features['access_specifier_ratio'] = access_spec_cnt / max(1, class_cnt)
        
        virtual_override_cnt = len(self.virtual_override_pattern.findall(pure_code))
        features['virtual_override_ratio'] = virtual_override_cnt / max(1, features['num_functions'])
        
        features['using_std_ratio'] = 1 if self.using_std_pattern.search(code_raw) else 0
        
        try_catch_cnt = len(self.try_catch_pattern.findall(pure_code))
        features['try_catch_ratio'] = try_catch_cnt / max(1, total_loc)
        
        raw_ptr_cnt = len(self.raw_pointer_pattern.findall(pure_code))
        features['raw_pointer_ratio'] = raw_ptr_cnt / max(1, total_words)
        
        getter_setter_cnt = len(self.getter_setter_pattern.findall(pure_code))
        features['getter_setter_ratio'] = getter_setter_cnt / max(1, len(custom_ids) if custom_ids else 1)
        
        std_prefix_cnt = pure_code.count('std::')
        features['std_prefix_ratio'] = std_prefix_cnt / max(1, total_words)
        
        cpp_cast_cnt = len(self.cpp_cast_pattern.findall(pure_code))
        features['cpp_cast_ratio'] = cpp_cast_cnt / max(1, total_words)

        return {k: features.get(k, 0.0) for k in self.feature_names}