gptdiff 0.1.5__tar.gz → 0.1.7__tar.gz

{gptdiff-0.1.5 → gptdiff-0.1.7}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: gptdiff
-Version: 0.1.5
+Version: 0.1.7
 Summary: A tool to generate and apply git diffs using LLMs
 Author: 255labs
 Classifier: License :: OSI Approved :: MIT License

{gptdiff-0.1.5 → gptdiff-0.1.7}/gptdiff/gptdiff.py

@@ -4,6 +4,7 @@ import openai
 from openai import OpenAI
 
 import tiktoken
+import time
 
 import os
 import json
@@ -20,7 +21,7 @@ import threading
 from pkgutil import get_data
 
 diff_context = contextvars.ContextVar('diffcontent', default="")
-def create_toolbox():
+def create_diff_toolbox():
     toolbox = Toolbox()
     
     def diff(content: str):
@@ -47,6 +48,25 @@ a/file.py b/file.py
     )
     return toolbox
 
+def create_think_toolbox():
+    toolbox = Toolbox()
+    
+    def think(content: str):
+        print("Swallowed thoughts", content)
+
+    toolbox.add_tool(
+        name="think",
+        fn=think,
+        args={
+            "content": {
+                "type": "string",
+                "description": "Thoughts"
+            }
+        },
+        description=""
+    )
+    return toolbox
+
 
 def load_gitignore_patterns(gitignore_path):
     with open(gitignore_path, 'r') as f:
@@ -145,11 +165,13 @@ def load_prepend_file(file):
         return f.read()
 
 # Function to call GPT-4 API and calculate the cost
-def call_gpt4_api(system_prompt, user_prompt, files_content, model, temperature=0.7, max_tokens=2500, api_key=None, base_url=None):
+def call_llm_for_diff(system_prompt, user_prompt, files_content, model, temperature=0.7, max_tokens=30000, api_key=None, base_url=None):
+    enc = tiktoken.get_encoding("o200k_base")
+    start_time = time.time()
 
     parser = FlatXMLParser("diff")
     formatter = FlatXMLPromptFormatter(tag="diff")
-    toolbox = create_toolbox()
+    toolbox = create_diff_toolbox()
     tool_prompt = formatter.usage_prompt(toolbox)
     system_prompt += "\n"+tool_prompt
 
@@ -164,7 +186,7 @@ def call_gpt4_api(system_prompt, user_prompt, files_content, model, temperature=
     print("SYSTEM PROMPT")
     print(system_prompt)
     print("USER PROMPT")
-    print(user_prompt, "+", len(files_content), "characters of file content")
+    print(user_prompt, "+", len(enc.encode(files_content)), "tokens of file content")
 
     if api_key is None:
         api_key = os.getenv('GPTDIFF_LLM_API_KEY')
@@ -180,6 +202,12 @@ def call_gpt4_api(system_prompt, user_prompt, files_content, model, temperature=
     completion_tokens = response.usage.completion_tokens
     total_tokens = response.usage.total_tokens
 
+    elapsed = time.time() - start_time
+    minutes, seconds = divmod(int(elapsed), 60)
+    time_str = f"{minutes}m {seconds}s" if minutes else f"{seconds}s"
+    print(f"Diff creation time: {time_str}")
+    print("-" * 40)
+
     # Now, these rates are updated to per million tokens
     cost_per_million_prompt_tokens = 30
     cost_per_million_completion_tokens = 60
@@ -187,7 +215,6 @@ def call_gpt4_api(system_prompt, user_prompt, files_content, model, temperature=
 
     full_response = response.choices[0].message.content.strip()
 
-
     events = parser.parse(full_response)
     for event in events:
         toolbox.use(event)
@@ -216,7 +243,7 @@ def generate_diff(environment, goal, model=None, temperature=0.7, max_tokens=320
         prepend = ""
     
     system_prompt = prepend+f"Output a git diff into a <diff> block."
-    _, diff_text, _, _, _, _ = call_gpt4_api(
+    _, diff_text, _, _, _, _ = call_llm_for_diff(
         system_prompt, 
         goal, 
         environment, 
@@ -266,7 +293,10 @@ def smartapply(diff_text, files, model=None, api_key=None, base_url=None):
     if model is None:
         model = os.getenv('GPTDIFF_MODEL', 'deepseek-reasoner')
     parsed_diffs = parse_diff_per_file(diff_text)    
-    print("SMARTAPPLY", diff_text)
+    print("-" * 40)
+    print("SMARTAPPLY")
+    print(diff_text)
+    print("-" * 40)
 
     def process_file(path, patch):
         original = files.get(path, '')
@@ -275,12 +305,12 @@ def smartapply(diff_text, files, model=None, api_key=None, base_url=None):
             if path in files:
                 del files[path]
         else:
-            updated = call_llm_for_apply(path, original, patch, model, api_key=api_key, base_url=base_url)
+            updated = call_llm_for_apply_with_think_tool_available(path, original, patch, model, api_key=api_key, base_url=base_url)
             files[path] = updated.strip()
-            
+
     for path, patch in parsed_diffs:
         process_file(path, patch)
-    
+
     return files
 
 # Function to apply diff to project files
@@ -307,6 +337,7 @@ def parse_arguments():
                         help='Call the GPT-4 API. Writes the full prompt to prompt.txt if not specified.')
     parser.add_argument('files', nargs='*', default=[], help='Specify additional files or directories to include.')
     parser.add_argument('--temperature', type=float, default=0.7, help='Temperature parameter for model creativity (0.0 to 2.0)')
+    parser.add_argument('--max_tokens', type=int, default=30000, help='Temperature parameter for model creativity (0.0 to 2.0)')
     parser.add_argument('--model', type=str, default=None, help='Model to use for the API call.')
 
     parser.add_argument('--nowarn', action='store_true', help='Disable large token warning')
@@ -320,20 +351,20 @@ def absolute_to_relative(absolute_path):
 
 def parse_diff_per_file(diff_text):
     """Parse unified diff text into individual file patches.
-    
+
     Splits a multi-file diff into per-file entries for processing. Handles:
     - File creations (+++ /dev/null)
     - File deletions (--- /dev/null)
     - Standard modifications
-    
+
     Args:
         diff_text: Unified diff string as generated by `git diff`
-        
+
     Returns:
         List of tuples (file_path, patch) where:
         - file_path: Relative path to modified file
         - patch: Full diff fragment for this file
-        
+
     Note:
         Uses 'b/' prefix detection from git diffs to determine target paths
     """
@@ -373,6 +404,26 @@ def parse_diff_per_file(diff_text):
 
     return diffs
 
+def call_llm_for_apply_with_think_tool_available(file_path, original_content, file_diff, model, api_key=None, base_url=None):
+    parser = FlatXMLParser("think")
+    formatter = FlatXMLPromptFormatter(tag="think")
+    toolbox = create_think_toolbox()
+    full_response = call_llm_for_apply(file_path, original_content, file_diff, model, api_key=None, base_url=None)
+    notool_response = ""
+    events = parser.parse(full_response)
+    is_in_tool = False
+    appended_content = ""
+    for event in events:
+        if event.mode == 'append':
+            appended_content += event.content
+        if event.mode == 'close' and appended_content and event.tool is None:
+            notool_response += appended_content
+        if event.mode == 'close':
+            appended_content = ""
+        toolbox.use(event)
+
+    return notool_response
+
 def call_llm_for_apply(file_path, original_content, file_diff, model, api_key=None, base_url=None):
     """AI-powered diff application with conflict resolution.
     
@@ -409,7 +460,8 @@ def call_llm_for_apply(file_path, original_content, file_diff, model, api_key=No
 
 1. Carefully apply all changes from the diff
 2. Preserve surrounding context that isn't changed
-3. Only return the final file content, do not add any additional markup and do not add a code block"""
+3. Only return the final file content, do not add any additional markup and do not add a code block
+4. You must return the entire file. It overwrites the existing file."""
 
     user_prompt = f"""File: {file_path}
 File contents:
@@ -434,12 +486,19 @@ Diff to apply:
     if base_url is None:
         base_url = os.getenv('GPTDIFF_LLM_BASE_URL', "https://nano-gpt.com/api/v1/")
     client = OpenAI(api_key=api_key, base_url=base_url)
+    start_time = time.time()
     response = client.chat.completions.create(model=model,
         messages=messages,
         temperature=0.0,
         max_tokens=30000)
+    full_response = response.choices[0].message.content
 
-    return response.choices[0].message.content
+    elapsed = time.time() - start_time
+    minutes, seconds = divmod(int(elapsed), 60)
+    time_str = f"{minutes}m {seconds}s" if minutes else f"{seconds}s"
+    print(f"Smartapply time: {time_str}")
+    print("-" * 40)
+    return full_response
 
 def build_environment_from_filelist(file_list, cwd):
     """Build environment string from list of file paths"""
@@ -531,10 +590,11 @@ def main():
             if confirmation != 'y':
                 print("Request canceled")
                 sys.exit(0)
-        full_text, diff_text, prompt_tokens, completion_tokens, total_tokens, cost = call_gpt4_api(system_prompt, user_prompt, files_content, args.model, 
+        full_text, diff_text, prompt_tokens, completion_tokens, total_tokens, cost = call_llm_for_diff(system_prompt, user_prompt, files_content, args.model, 
                                                                                                     temperature=args.temperature,
                                                                                                     api_key=os.getenv('GPTDIFF_LLM_API_KEY'),
-                                                                                                    base_url=os.getenv('GPTDIFF_LLM_BASE_URL', "https://nano-gpt.com/api/v1/") 
+                                                                                                    base_url=os.getenv('GPTDIFF_LLM_BASE_URL', "https://nano-gpt.com/api/v1/"),
+                                                                                                    max_tokens=args.max_tokens
                                                                                                     ) 
 
     if(diff_text.strip() == ""):
@@ -587,8 +647,17 @@ def main():
                         print(f"Skipping binary file {file_path}")
                         return
 
+                print("-" * 40)
+                print("SMARTAPPLY")
+                print(file_diff)
+                print("-" * 40)
                 try:
-                    updated_content = call_llm_for_apply(file_path, original_content, file_diff, args.model)
+                    updated_content = call_llm_for_apply_with_think_tool_available(file_path, original_content, file_diff, args.model)
+
+                    if updated_content.strip() == "":
+                        print("Cowardly refusing to write empty file to", file_path, "merge failed")
+                        return
+
                     full_path.parent.mkdir(parents=True, exist_ok=True)
                     full_path.write_text(updated_content)
                     print(f"\033[1;32mSuccessful 'smartapply' update {file_path}.\033[0m")

{gptdiff-0.1.5 → gptdiff-0.1.7}/gptdiff.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: gptdiff
-Version: 0.1.5
+Version: 0.1.7
 Summary: A tool to generate and apply git diffs using LLMs
 Author: 255labs
 Classifier: License :: OSI Approved :: MIT License

{gptdiff-0.1.5 → gptdiff-0.1.7}/setup.py

@@ -2,7 +2,7 @@ from setuptools import setup, find_packages
 
 setup(
     name='gptdiff',
-    version='0.1.5',
+    version='0.1.7',
     description='A tool to generate and apply git diffs using LLMs',
     author='255labs',
     packages=find_packages(),  # Use find_packages() to automatically discover packages

{gptdiff-0.1.5 → gptdiff-0.1.7}/tests/test_smartapply.py

@@ -35,6 +35,32 @@ def test_smartapply_file_modification():
      print('Hello')
 +
 +def goodbye():
++    print('Goodbye')'''
+
+    original_hello = "def hello():\n    print('Hello')"
+    original_files = {
+        "hello.py": original_hello
+    }
+
+    # Mock LLM to return modified content
+    with patch('gptdiff.gptdiff.call_llm_for_apply',
+               return_value="\ndef goodbye():\n    print('Goodbye')"):
+
+        updated_files = smartapply(diff_text, original_files)
+        
+        assert "hello.py" in updated_files
+        assert original_hello != updated_files["hello.py"]
+
+def test_smartapply_think_then_modify():
+    """Test that smartapply correctly handles file modification diffs"""
+    diff_text = '''diff --git a/hello.py b/hello.py
+--- a/hello.py
++++ b/hello.py
+@@ -1,2 +1,5 @@
+ def hello():
+     print('Hello')
++
++def goodbye():
 +    print('Goodbye')'''
 
     original_files = {
@@ -43,12 +69,13 @@ def test_smartapply_file_modification():
 
     # Mock LLM to return modified content
     with patch('gptdiff.gptdiff.call_llm_for_apply',
-               return_value="def hello():\n    print('Hello')\n\ndef goodbye():\n    print('Goodbye')"):
+               return_value="<think>Hello from thoughts</think>\ndef goodbye():\n    print('Goodbye')"):
 
         updated_files = smartapply(diff_text, original_files)
         
         assert "hello.py" in updated_files
-        assert original_files["hello.py"] != updated_files["hello.py"]
+        assert updated_files["hello.py"] == "def goodbye():\n    print('Goodbye')"
+
 
 def test_smartapply_new_file_creation():
     """Test that smartapply handles new file creation through diffs"""