gptdiff 0.1.18__tar.gz → 0.1.21__tar.gz

{gptdiff-0.1.18 → gptdiff-0.1.21}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: gptdiff
-Version: 0.1.18
+Version: 0.1.21
 Summary: A tool to generate and apply git diffs using LLMs
 Author: 255labs
 Classifier: License :: OSI Approved :: MIT License
@@ -10,7 +10,7 @@ Description-Content-Type: text/markdown
 License-File: LICENSE.txt
 Requires-Dist: openai>=1.0.0
 Requires-Dist: tiktoken>=0.5.0
-Requires-Dist: ai_agent_toolbox>=0.1.13
+Requires-Dist: ai-agent-toolbox>=0.1.15
 Provides-Extra: test
 Requires-Dist: pytest; extra == "test"
 Requires-Dist: pytest-mock; extra == "test"

{gptdiff-0.1.18 → gptdiff-0.1.21}/gptdiff/gptdiff.py

@@ -202,7 +202,7 @@ def call_llm_for_diff(system_prompt, user_prompt, files_content, model, temperat
     tool_prompt = formatter.usage_prompt(toolbox)
     system_prompt += "\n"+tool_prompt
 
-    if model == "gemini-2.0-flash-thinking-exp-01-21":
+    if 'gemini' in model:
         user_prompt = system_prompt+"\n"+user_prompt
 
     messages = [
@@ -241,6 +241,9 @@ def call_llm_for_diff(system_prompt, user_prompt, files_content, model, temperat
     cost = (prompt_tokens / 1_000_000 * cost_per_million_prompt_tokens) + (completion_tokens / 1_000_000 * cost_per_million_completion_tokens)
 
     full_response = response.choices[0].message.content.strip()
+    full_response, reasoning = swallow_reasoning(full_response)
+    if reasoning and len(reasoning) > 0:
+        print("Swallowed reasoning", reasoning)
 
     events = parser.parse(full_response)
     for event in events:
@@ -643,6 +646,9 @@ def call_llm_for_apply_with_think_tool_available(file_path, original_content, fi
     formatter = FlatXMLPromptFormatter(tag="think")
     toolbox = create_think_toolbox()
     full_response = call_llm_for_apply(file_path, original_content, file_diff, model, api_key=api_key, base_url=base_url, extra_prompt=extra_prompt, max_tokens=max_tokens)
+    full_response, reasoning = swallow_reasoning(full_response)
+    if reasoning and len(reasoning) > 0:
+        print("Swallowed reasoning", reasoning)
     notool_response = ""
     events = parser.parse(full_response)
     is_in_tool = False
@@ -707,7 +713,7 @@ Diff to apply:
 ```"""
     if extra_prompt:
         user_prompt += f"\n\n{extra_prompt}"
-    if model == "gemini-2.0-flash-thinking-exp-01-21":
+    if 'gemini' in model:
         user_prompt = system_prompt+"\n"+user_prompt
     messages = [
         {"role": "system", "content": system_prompt},
@@ -967,5 +973,35 @@ def main():
     print(f"Total tokens: {total_tokens}")
     #print(f"Total cost: ${cost:.4f}")
 
+def swallow_reasoning(full_response: str) -> (str, str):
+    """
+    Extracts and swallows the chain-of-thought reasoning section from the full LLM response.
+    Assumes the reasoning block starts with a line beginning with "> Reasoning"
+    and ends with a line matching 'Reasoned for <number> seconds'.
+
+    Returns:
+        A tuple (final_content, reasoning) where:
+         - final_content: The response with the reasoning block removed.
+         - reasoning: The extracted reasoning block, or an empty string if not found.
+    """
+    pattern = re.compile(
+        r"(?P<reasoning>>\s*Reasoning.*?Reasoned.*?seconds)",
+        re.DOTALL
+    )
+    match = pattern.search(full_response)
+    if match:
+        raw_reasoning = match.group("reasoning")
+        # Remove any leading '+' characters and extra whitespace from each line
+        reasoning_lines = [line.lstrip('+').strip() for line in raw_reasoning.splitlines()]
+        reasoning = "\n".join(reasoning_lines).strip()
+
+        # Remove the reasoning block from the response using its exact span
+        final_content = full_response[:match.start()] + full_response[match.end():]
+        final_content = final_content.strip()
+    else:
+        reasoning = ""
+        final_content = full_response.strip()
+    return final_content, reasoning
+
 if __name__ == "__main__":
-    main()
+    main()

{gptdiff-0.1.18 → gptdiff-0.1.21}/gptdiff.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: gptdiff
-Version: 0.1.18
+Version: 0.1.21
 Summary: A tool to generate and apply git diffs using LLMs
 Author: 255labs
 Classifier: License :: OSI Approved :: MIT License
@@ -10,7 +10,7 @@ Description-Content-Type: text/markdown
 License-File: LICENSE.txt
 Requires-Dist: openai>=1.0.0
 Requires-Dist: tiktoken>=0.5.0
-Requires-Dist: ai_agent_toolbox>=0.1.13
+Requires-Dist: ai-agent-toolbox>=0.1.15
 Provides-Extra: test
 Requires-Dist: pytest; extra == "test"
 Requires-Dist: pytest-mock; extra == "test"

{gptdiff-0.1.18 → gptdiff-0.1.21}/gptdiff.egg-info/SOURCES.txt

@@ -15,4 +15,5 @@ tests/test_applydiff_edgecases.py
 tests/test_diff_parse.py
 tests/test_failing_case.py
 tests/test_parse_diff_per_file.py
-tests/test_smartapply.py
+tests/test_smartapply.py
+tests/test_swallow_reasoning.py

{gptdiff-0.1.18 → gptdiff-0.1.21}/gptdiff.egg-info/requires.txt

@@ -1,6 +1,6 @@
 openai>=1.0.0
 tiktoken>=0.5.0
-ai_agent_toolbox>=0.1.13
+ai-agent-toolbox>=0.1.15
 
 [docs]
 mkdocs

{gptdiff-0.1.18 → gptdiff-0.1.21}/setup.py

@@ -2,7 +2,7 @@ from setuptools import setup, find_packages
 
 setup(
     name='gptdiff',
-    version='0.1.18',
+    version='0.1.21',
     description='A tool to generate and apply git diffs using LLMs',
     author='255labs',
     packages=find_packages(),  # Use find_packages() to automatically discover packages
@@ -12,7 +12,7 @@ setup(
     install_requires=[
         'openai>=1.0.0',
         'tiktoken>=0.5.0',
-        'ai_agent_toolbox>=0.1.13'
+        'ai-agent-toolbox>=0.1.15'
     ],
     extras_require={
         'test': ['pytest', 'pytest-mock'],

gptdiff-0.1.21/tests/test_swallow_reasoning.py

@@ -0,0 +1,156 @@
+import pytest
+
+from gptdiff.gptdiff import swallow_reasoning
+
+def test_swallow_reasoning_extraction():
+    llm_response = (
+        "+> Reasoning\n"
+        "+None\n"
+        "+Reasoned about summary drawer button 변경 for 15 seconds\n"
+        "+def new():\n"
+        "```"
+    )
+    final_content, reasoning = swallow_reasoning(llm_response)
+    expected_reasoning = (
+        "> Reasoning\n"
+        "**Applying the diff**\n"
+        "I'm piecing together how to efficiently apply a diff to a file...\n"
+        "**Returning the result**\n"
+        "I'm finalizing the method to apply the diff updates...\n"
+        "Reasoned for 6 seconds"
+    )
+    assert reasoning == expected_reasoning
+    # The final content should no longer contain the reasoning block.
+    assert expected_reasoning not in final_content
+    # And it should contain the diff block.
+    assert "```diff" in final_content
+
+
+def test_swallow_reasoning_with_untested_response():
+    llm_response = (
+        "> Reasoning\n"
+        "**Considering the request**\n"
+        "I’m noting that the user wants me to apply a diff to a file and return the result in a block, ensuring the entire file is included.\n"
+        "**Ensuring comprehensive inclusion**\n"
+        "I'm making sure the entire file is included when presenting the result in a block, following the user's request carefully.\n"
+        "**Ensuring clarity**\n"
+        "I’m integrating the diff into the file and ensuring the entire file is returned as requested. This approach maintains precision and clarity in the response.\n"
+        "**Refining the response**\n"
+        "I’m focusing on how to structure the response by carefully integrating the diff and ensuring the entire file is included in a clear block format.\n"
+        "**Connecting the pieces**\n"
+        "I'm mapping out how to apply the diff to the file carefully and ensure the entire file is incorporated into the final block.\n"
+        "Reasoned for a few seconds\n"
+        "\n"
+        "```diff\n"
+        "--- a/file.py\n"
+        "+++ b/file.py\n"
+        "@@ -1,2 +1,2 @@\n"
+        "-def old():\n"
+        "+def new():\n"
+        "```"
+    )
+    final_content, reasoning = swallow_reasoning(llm_response)
+
+    expected_reasoning = (
+        "> Reasoning\n"
+        "**Considering the request**\n"
+        "I’m noting that the user wants me to apply a diff to a file and return the result in a block, ensuring the entire file is included.\n"
+        "**Ensuring comprehensive inclusion**\n"
+        "I'm making sure the entire file is included when presenting the result in a block, following the user's request carefully.\n"
+        "**Ensuring clarity**\n"
+        "I’m integrating the diff into the file and ensuring the entire file is returned as requested. This approach maintains precision and clarity in the response.\n"
+        "**Refining the response**\n"
+        "I’m focusing on how to structure the response by carefully integrating the diff and ensuring the entire file is included in a clear block format.\n"
+        "**Connecting the pieces**\n"
+        "I'm mapping out how to apply the diff to the file carefully and ensure the entire file is incorporated into the final block.\n"
+        "Reasoned for a few seconds"
+    )
+
+    assert reasoning == expected_reasoning
+    # The final content should no longer contain the reasoning block.
+    assert expected_reasoning not in final_content
+    # And it should contain the diff block.
+    assert "```diff" in final_content
+
+def test_swallow_reasoning_extraction():
+    llm_response = (
+        "> Reasoning\n"
+        "**Applying the diff**\n"
+        "I'm piecing together how to efficiently apply a diff to a file...\n"
+        "**Returning the result**\n"
+        "I'm finalizing the method to apply the diff updates...\n"
+        "Reasoned for 6 seconds\n"
+        "\n"
+        "```diff\n"
+        "--- a/file.py\n"
+        "+++ b/file.py\n"
+        "@@ -1,2 +1,2 @@\n"
+        "-def old():\n"
+        "+def new():\n"
+        "```"
+    )
+    final_content, reasoning = swallow_reasoning(llm_response)
+    expected_reasoning = (
+        "> Reasoning\n"
+        "**Applying the diff**\n"
+        "I'm piecing together how to efficiently apply a diff to a file...\n"
+        "**Returning the result**\n"
+        "I'm finalizing the method to apply the diff updates...\n"
+        "Reasoned for 6 seconds"
+    )
+    assert reasoning == expected_reasoning
+    # The final content should no longer contain the reasoning block.
+    assert expected_reasoning not in final_content
+    # And it should contain the diff block.
+    assert "```diff" in final_content
+
+
+def test_swallow_reasoning_no_reasoning():
+    llm_response = (
+        "```diff\n"
+        "--- a/file.py\n"
+        "+++ b/file.py\n"
+        "@@ -1,2 +1,2 @@\n"
+        "-def old():\n"
+        "+def new():\n"
+        "```"
+    )
+    final_content, reasoning = swallow_reasoning(llm_response)
+    assert reasoning == ""
+    assert final_content == llm_response.strip()
+
+def test_swallow_reasoning_inline_newlines():
+    llm_response = (
+        "Prefix text before reasoning and some inline content "
+        "> Reasoning\n"
+        "Inline line 1\n"
+        "Inline line 2\n"
+        "Reasoned for 2 seconds "
+        "and then suffix text.\n"
+        "```diff\n"
+        "--- a/inline.py\n"
+        "+++ b/inline.py\n"
+        "@@ -1,2 +1,2 @@\n"
+        "-print('Old')\n"
+        "+print('New')\n"
+        "```"
+    )
+    final_content, reasoning = swallow_reasoning(llm_response)
+    expected_reasoning = (
+        "> Reasoning\n"
+        "Inline line 1\n"
+        "Inline line 2\n"
+        "Reasoned for 2 seconds"
+    )
+    # Count the newlines in the extracted reasoning block.
+    newline_count = reasoning.count('\n')
+    # There should be 3 newline characters: after "> Reasoning", after "Inline line 1", and after "Inline line 2"
+    assert newline_count == 3, f"Expected 3 newlines, got {newline_count}"
+    assert reasoning == expected_reasoning
+    # Ensure the reasoning block is removed from the final content.
+    assert expected_reasoning not in final_content
+    # Verify that surrounding content remains.
+    assert "Prefix text before reasoning" in final_content
+    assert "and then suffix text." in final_content
+    # Verify that the diff block is still present.
+    assert "```diff" in final_content