added scripts

.gitignore

@@ -0,0 +1 @@
+node_modules

package.json

@@ -0,0 +1,6 @@
+{
+  "dependencies": {
+    "@phenomnomnominal/tsquery": "^5.0.1",
+    "typescript": "^5.0.4"
+  }
+}

parse.py

@@ -0,0 +1,47 @@
+import subprocess
+from datasets import load_dataset, Dataset
+import json 
+from tqdm import tqdm
+
+# dataset streaming (will only download the data as needed)
+ds = load_dataset("../datasets/the-stack-smol", data_dir='data/typescript')
+print('loaded')
+
+def split_ts_into_chunks(ts_code):
+    result = subprocess.run(
+        ['node', 'parse_ts.js'],
+        input=ts_code,
+        text=True,
+    )
+
+    if result.returncode != 0:
+        raise Exception('Error in TypeScript parsing')
+
+    with open('semantic_chunks.jsonl', 'r') as file:
+        lines = file.read().splitlines()
+
+    chunks = [json.loads(line) for line in lines]
+    with open('semantic_chunks.jsonl', 'w'):
+        pass
+
+    return chunks
+
+
+def chunk_ts_file(data):
+    funcs = split_ts_into_chunks(data['content'])
+    for i in range(len(funcs)):
+        funcs[i]['repo'] = data['repository_name']
+        funcs[i]['path'] = data['path']
+        funcs[i]['language'] = data['lang']
+    return funcs
+
+chunks = []
+for i in tqdm(range(len(ds['train']))):
+    chunk = chunk_ts_file(ds['train'][i])
+    chunks +=(chunk)
+    if i%100 == 0:
+        print(len(chunks))
+
+dataset = Dataset.from_list(chunks)
+print(dataset)
+dataset.to_json('ts-chunks.json')

parse_ts.js

@@ -0,0 +1,45 @@
+// parse_ts.js
+
+const fs = require('fs');
+const ts = require('typescript');
+const { tsquery } = require('@phenomnomnominal/tsquery');
+
+let data = '';
+
+process.stdin.on('data', function (chunk) {
+    data += chunk;
+});
+
+process.stdin.on('end', function () {
+    const sourceFile = ts.createSourceFile('temp.ts', data.toString(), ts.ScriptTarget.ES2015, true);
+
+    // Define the node types that constitute a 'semantic chunk'
+    const semanticChunkNodeTypes = [
+        'FunctionDeclaration',
+        'ArrowFunction',
+        'ClassDeclaration',
+        'InterfaceDeclaration',
+        'EnumDeclaration',
+        'TypeAliasDeclaration',
+        'MethodDeclaration',
+    ];
+
+    const semanticChunks = semanticChunkNodeTypes.flatMap(nodeType =>
+        tsquery(sourceFile, nodeType)
+    );
+
+    const jsonl = semanticChunks.map(chunk => {
+        const comments = ts.getLeadingCommentRanges(sourceFile.getFullText(), chunk.getFullStart()) || [];
+        const commentTexts = comments.map(comment => sourceFile.getFullText().slice(comment.pos, comment.end)).join('\n');
+
+        // Append the leading comments to the front of the chunk's content
+        const contentWithComments = commentTexts + '\n' + chunk.getText(sourceFile);
+
+        return JSON.stringify({
+            type: ts.SyntaxKind[chunk.kind],
+            content: contentWithComments
+        });
+    }).join('\n');
+
+    fs.writeFileSync('semantic_chunks.jsonl', jsonl);
+});