File size: 8,497 Bytes
f514f40
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from typing import Dict, Any, List

import ast
import tarfile
from ast import AsyncFunctionDef, ClassDef, FunctionDef, Module
import torch
import requests
from transformers import Pipeline
from tqdm.auto import tqdm


def extract_code_and_docs(text: str):
    code_set = set()
    docs_set = set()
    root = ast.parse(text)
    for node in ast.walk(root):
        if not isinstance(node, (AsyncFunctionDef, FunctionDef, ClassDef, Module)):
            continue
        docs = ast.get_docstring(node)
        node_without_docs = node
        if docs is not None:
            docs_set.add(docs)
            # Remove docstrings from the node
            node_without_docs.body = node_without_docs.body[1:]
        if isinstance(node, (AsyncFunctionDef, FunctionDef)):
            code_set.add(ast.unparse(node_without_docs))

    return code_set, docs_set


def get_metadata(repo_name, headers=None):
    api_url = f"https://api.github.com/repos/{repo_name}"
    tqdm.write(f"[+] Getting metadata for {repo_name}")
    try:
        response = requests.get(api_url, headers=headers)
        response.raise_for_status()
        return response.json()
    except requests.exceptions.HTTPError as e:
        tqdm.write(f"[-] Failed to retrieve metadata from {repo_name}: {e}")
        return {}


def extract_information(repos, headers=None):
    extracted_infos = []
    for repo_name in tqdm(repos, disable=len(repos) <= 1):
        # Get metadata
        metadata = get_metadata(repo_name, headers=headers)
        repo_info = {
            "name": repo_name,
            "codes": set(),
            "docs": set(),
            "requirements": set(),
            "readmes": set(),
            "topics": [],
            "license": "",
            "stars": metadata.get("stargazers_count"),
        }
        if metadata.get("topics"):
            repo_info["topics"] = metadata["topics"]
        if metadata.get("license"):
            repo_info["license"] = metadata["license"]["spdx_id"]

        # Download repo tarball bytes
        download_url = f"https://api.github.com/repos/{repo_name}/tarball"
        tqdm.write(f"[+] Downloading {repo_name}")
        try:
            response = requests.get(download_url, headers=headers, stream=True)
            response.raise_for_status()
        except requests.exceptions.HTTPError as e:
            tqdm.write(f"[-] Failed to download {repo_name}: {e}")
            continue

        # Extract python files and parse them
        tqdm.write(f"[+] Extracting {repo_name} info")
        with tarfile.open(fileobj=response.raw, mode="r|gz") as tar:
            for member in tar:
                if (member.name.endswith(".py") and member.isfile()) is False:
                    continue
                try:
                    file_content = tar.extractfile(member).read().decode("utf-8")
                    code_set, docs_set = extract_code_and_docs(file_content)

                    repo_info["codes"].update(code_set)
                    repo_info["docs"].update(docs_set)
                except UnicodeDecodeError as e:
                    tqdm.write(
                        f"[-] UnicodeDecodeError in {member.name}, skipping: \n{e}"
                    )
                except SyntaxError as e:
                    tqdm.write(f"[-] SyntaxError in {member.name}, skipping: \n{e}")

        extracted_infos.append(repo_info)

    return extracted_infos


class RepoPipeline(Pipeline):

    def __init__(self, github_token=None, *args, **kwargs):
        super().__init__(*args, **kwargs)

        # Github token
        self.github_token = github_token
        if self.github_token:
            print("[+] GitHub token set!")
        else:
            print(
                "[*] Please set GitHub token to avoid unexpected errors. \n"
                "For more info, see: "
                "https://docs.github.com/authentication/keeping-your-account-and-data-secure/creating-a-personal-access-token"
            )

    def _sanitize_parameters(self, **pipeline_parameters):
        preprocess_parameters = {}
        if "github_token" in pipeline_parameters:
            preprocess_parameters["github_token"] = pipeline_parameters["github_token"]

        forward_parameters = {}
        if "max_length" in pipeline_parameters:
            forward_parameters["max_length"] = pipeline_parameters["max_length"]

        postprocess_parameters = {}
        return preprocess_parameters, forward_parameters, postprocess_parameters

    def preprocess(self, input_: Any, **preprocess_parameters: Dict) -> List:
        # Making input to list format
        if isinstance(input_, str):
            input_ = [input_]

        # Building token
        github_token = preprocess_parameters["preprocess_parameters"]
        headers = {"Accept": "application/vnd.github+json"}
        token = github_token or self.github_token
        if token:
            headers["Authorization"] = f"Bearer {token}"

        # Getting repositories' information: input_ means series of repositories
        extracted_infos = extract_information(input_, headers=headers)

        return extracted_infos

    def encode(self, text, max_length):
        assert max_length < 1024

        tokenizer = self.tokenizer
        tokens = (
                [tokenizer.cls_token, "<encoder-only>", tokenizer.sep_token]
                + tokenizer.tokenize(text)[: max_length - 4]
                + [tokenizer.sep_token]
        )
        tokens_id = tokenizer.convert_tokens_to_ids(tokens)
        source_ids = torch.tensor([tokens_id]).to(self.device)

        token_embeddings = self.model(source_ids)[0]
        sentence_embeddings = token_embeddings.mean(dim=1)

        return sentence_embeddings

    def generate_embeddings(self, text_sets, max_length):
        assert max_length < 1024
        return torch.concat([self.encode(text, max_length) for text in text_sets], dim=0) \
            if text_sets is None or len(text_sets) == 0 \
            else torch.zeros((1, 768), device=self.device)

    def _forward(self, extracted_infos: List, **forward_parameters: Dict) -> List:
        max_length = 512 if forward_parameters["max_length"] is None else forward_parameters["max_length"]

        model_outputs = []
        num_repos = len(extracted_infos)
        with tqdm(total=num_repos) as progress_bar:
            # For each repository
            for repo_info in extracted_infos:
                repo_name = repo_info["name"]
                info = {
                    "name": repo_name,
                    "topics": repo_info["topics"],
                    "license": repo_info["license"],
                    "stars": repo_info["stars"],
                }
                progress_bar.set_description(f"Processing {repo_name}")

                # Code embeddings
                tqdm.write(f"[*] Generating code embeddings for {repo_name}")
                code_embeddings = self.generate_embeddings(repo_info["codes"], max_length)
                info["code_embeddings"] = code_embeddings.item()
                info["mean_code_embedding"] = torch.mean(code_embeddings, dim=0).item()

                # Doc embeddings
                tqdm.write(f"[*] Generating doc embeddings for {repo_name}")
                doc_embeddings = self.generate_embeddings(repo_info["docs"], max_length)
                info["doc_embeddings"] = doc_embeddings.item()
                info["mean_doc_embedding"] = torch.mean(doc_embeddings, dim=0).item()

                # Requirement embeddings
                tqdm.write(f"[*] Generating requirement embeddings for {repo_name}")
                requirement_embeddings = self.generate_embeddings(repo_info["requirements"], max_length)
                info["requirement_embeddings"] = requirement_embeddings.item()
                info["mean_requirement_embedding"] = torch.mean(requirement_embeddings, dim=0).item()

                # Requirement embeddings
                tqdm.write(f"[*] Generating readme embeddings for {repo_name}")
                readme_embeddings = self.generate_embeddings(repo_info["readmes"], max_length)
                info["readme_embeddings"] = readme_embeddings.item()
                info["mean_readme_embedding"] = torch.mean(readme_embeddings, dim=0).item()

                progress_bar.update(1)
                model_outputs.append(info)

        return model_outputs

    def postprocess(self, model_outputs: List, **postprocess_parameters: Dict) -> List:
        return model_outputs