autocoder-nano 0.1.26__py3-none-any.whl → 0.1.28__py3-none-any.whl

autocoder_nano/edit/__init__.py

@@ -0,0 +1,20 @@
+from typing import Optional
+
+from autocoder_nano.edit.actions import ActionPyProject, ActionSuffixProject
+from autocoder_nano.llm_client import AutoLLM
+from autocoder_nano.llm_types import AutoCoderArgs
+
+
+class Dispacher:
+    def __init__(self, args: AutoCoderArgs, llm: Optional[AutoLLM] = None):
+        self.args = args
+        self.llm = llm
+
+    def dispach(self):
+        dispacher_actions = [
+            ActionPyProject(args=self.args, llm=self.llm),
+            ActionSuffixProject(args=self.args, llm=self.llm)
+        ]
+        for action in dispacher_actions:
+            if action.run():
+                return

autocoder_nano/edit/actions.py

@@ -0,0 +1,136 @@
+import time
+
+from loguru import logger
+
+from autocoder_nano.edit.code.generate_editblock import CodeAutoGenerateEditBlock
+from autocoder_nano.edit.code.merge_editblock import CodeAutoMergeEditBlock
+from autocoder_nano.index.entry import build_index_and_filter_files
+from autocoder_nano.llm_client import AutoLLM
+from autocoder_nano.llm_types import AutoCoderArgs
+from autocoder_nano.project import PyProject, SuffixProject
+
+
+class BaseAction:
+    @staticmethod
+    def _get_content_length(content: str) -> int:
+        return len(content)
+
+
+class ActionPyProject(BaseAction):
+    def __init__(self, args: AutoCoderArgs, llm: AutoLLM) -> None:
+        self.args = args
+        self.llm = llm
+        self.pp = None
+
+    def run(self):
+        if self.args.project_type != "py":
+            return False
+        pp = PyProject(llm=self.llm, args=self.args)
+        self.pp = pp
+        pp.run()
+        source_code = pp.output()
+        if self.llm:
+            source_code = build_index_and_filter_files(args=self.args, llm=self.llm, sources=pp.sources)
+        self.process_content(source_code)
+        return True
+
+    def process_content(self, content: str):
+        # args = self.args
+        if self.args.execute and self.llm:
+            content_length = self._get_content_length(content)
+            if content_length > self.args.model_max_input_length:
+                logger.warning(
+                    f"发送给模型的内容长度为 {content_length} 个 token（可能收集了过多文件），"
+                    f"已超过最大输入长度限制 {self.args.model_max_input_length}。"
+                )
+
+        if self.args.execute:
+            logger.info("正在自动生成代码...")
+            start_time = time.time()
+            # diff, strict_diff, editblock 是代码自动生成或合并的不同策略, 通常用于处理代码的变更或生成
+            # diff 模式,基于差异生成代码,生成最小的变更集,适用于局部优化,代码重构
+            # strict_diff 模式,严格验证差异,确保生成的代码符合规则,适用于代码审查,自动化测试
+            # editblock 模式,基于编辑块生成代码，支持较大范围的修改,适用于代码重构,功能扩展
+            if self.args.auto_merge == "editblock":
+                generate = CodeAutoGenerateEditBlock(args=self.args, llm=self.llm, action=self)
+            else:
+                generate = None
+
+            if self.args.enable_multi_round_generate:
+                generate_result = generate.multi_round_run(query=self.args.query, source_content=content)
+            else:
+                generate_result = generate.single_round_run(query=self.args.query, source_content=content)
+            logger.info(f"代码生成完成，耗时 {time.time() - start_time:.2f} 秒")
+
+            if self.args.auto_merge:
+                logger.info("正在自动合并代码...")
+                if self.args.auto_merge == "editblock":
+                    code_merge = CodeAutoMergeEditBlock(args=self.args, llm=self.llm)
+                    merge_result = code_merge.merge_code(generate_result=generate_result)
+                else:
+                    merge_result = None
+
+                content = merge_result.contents[0]
+            else:
+                content = generate_result.contents[0]
+
+            with open(self.args.target_file, "w") as file:
+                file.write(content)
+
+
+class ActionSuffixProject(BaseAction):
+    def __init__(self, args: AutoCoderArgs, llm: AutoLLM) -> None:
+        self.args = args
+        self.llm = llm
+        self.pp = None
+
+    def run(self):
+        pp = SuffixProject(llm=self.llm, args=self.args)
+        self.pp = pp
+        pp.run()
+        source_code = pp.output()
+        if self.llm:
+            source_code = build_index_and_filter_files(args=self.args, llm=self.llm, sources=pp.sources)
+        self.process_content(source_code)
+
+    def process_content(self, content: str):
+        if self.args.execute and self.llm:
+            content_length = self._get_content_length(content)
+            if content_length > self.args.model_max_input_length:
+                logger.warning(
+                    f"发送给模型的内容长度为 {content_length} 个 token（可能收集了过多文件），"
+                    f"已超过最大输入长度限制 {self.args.model_max_input_length}。"
+                )
+
+        if self.args.execute:
+            logger.info("正在自动生成代码...")
+            start_time = time.time()
+            # diff, strict_diff, editblock 是代码自动生成或合并的不同策略, 通常用于处理代码的变更或生成
+            # diff 模式,基于差异生成代码,生成最小的变更集,适用于局部优化,代码重构
+            # strict_diff 模式,严格验证差异,确保生成的代码符合规则,适用于代码审查,自动化测试
+            # editblock 模式,基于编辑块生成代码，支持较大范围的修改,适用于代码重构,功能扩展
+            if self.args.auto_merge == "editblock":
+                generate = CodeAutoGenerateEditBlock(args=self.args, llm=self.llm, action=self)
+            else:
+                generate = None
+
+            if self.args.enable_multi_round_generate:
+                generate_result = generate.multi_round_run(query=self.args.query, source_content=content)
+            else:
+                generate_result = generate.single_round_run(query=self.args.query, source_content=content)
+            logger.info(f"代码生成完成，耗时 {time.time() - start_time:.2f} 秒")
+
+            if self.args.auto_merge:
+                logger.info("正在自动合并代码...")
+                if self.args.auto_merge == "editblock":
+                    code_merge = CodeAutoMergeEditBlock(args=self.args, llm=self.llm)
+                    merge_result = code_merge.merge_code(generate_result=generate_result)
+                else:
+                    merge_result = None
+
+                content = merge_result.contents[0]
+            else:
+                content = generate_result.contents[0]
+
+            with open(self.args.target_file, "w") as file:
+                file.write(content)

autocoder_nano/edit/code/generate_editblock.py

@@ -0,0 +1,403 @@
+from autocoder_nano.llm_client import AutoLLM
+from autocoder_nano.llm_prompt import prompt
+from autocoder_nano.llm_types import AutoCoderArgs, CodeGenerateResult
+
+
+class CodeAutoGenerateEditBlock:
+    def __init__(self, args: AutoCoderArgs, llm: AutoLLM, action=None):
+        self.llm = llm
+        self.args = args
+        self.action = action
+        if not self.llm:
+            raise ValueError("Please provide a valid model instance to use for code generation.")
+        self.llms = [self.llm]
+
+    @prompt()
+    def single_round_instruction(self, instruction: str, content: str, context: str = ""):
+        """
+        如果你需要生成代码，对于每个需要更改的文件,你需要按 *SEARCH/REPLACE block* 的格式进行生成。
+
+        # *SEARCH/REPLACE block* Rules:
+
+        Every *SEARCH/REPLACE block* must use this format:
+        1. The opening fence and code language, eg: ```python
+        2. The file path alone on a line, starting with "##File:" and verbatim. No bold asterisks, no quotes around it,
+        no escaping of characters, etc.
+        3. The start of search block: <<<<<<< SEARCH
+        4. A contiguous chunk of lines to search for in the existing source code
+        5. The dividing line: =======
+        6. The lines to replace into the source code
+        7. The end of the replacement block: >>>>>>> REPLACE
+        8. The closing fence: ```
+
+        Every *SEARCH* section must *EXACTLY MATCH* the existing source code, character for character,
+        including all comments, docstrings, etc.
+
+        *SEARCH/REPLACE* blocks will replace *all* matching occurrences.
+        Include enough lines to make the SEARCH blocks unique.
+
+        Include *ALL* the code being searched and replaced!
+
+        To move code within a file, use 2 *SEARCH/REPLACE* blocks: 1 to delete it from its current location,
+        1 to insert it in the new location.
+
+        If you want to put code in a new file, use a *SEARCH/REPLACE block* with:
+        - A new file path, including dir name if needed
+        - An empty `SEARCH` section
+        - The new file's contents in the `REPLACE` section
+
+        ONLY EVER RETURN CODE IN A *SEARCH/REPLACE BLOCK*!
+
+        下面我们来看一个例子：
+
+        当前项目目录结构：
+        1. 项目根目录： /tmp/projects/mathweb
+        2. 项目子目录/文件列表(类似tree 命令输出)
+        flask/
+            app.py
+            templates/
+                index.html
+            static/
+                style.css
+
+        用户需求： Change get_factorial() to use math.factorial
+
+        回答： To make this change we need to modify `/tmp/projects/mathweb/flask/app.py` to:
+
+        1. Import the math package.
+        2. Remove the existing factorial() function.
+        3. Update get_factorial() to call math.factorial instead.
+
+        Here are the *SEARCH/REPLACE* blocks:
+
+        ```python
+        ##File: /tmp/projects/mathweb/flask/app.py
+        <<<<<<< SEARCH
+        from flask import Flask
+        =======
+        import math
+        from flask import Flask
+        >>>>>>> REPLACE
+        ```
+
+        ```python
+        ##File: /tmp/projects/mathweb/flask/app.py
+        <<<<<<< SEARCH
+        def factorial(n):
+            "compute factorial"
+
+            if n == 0:
+                return 1
+            else:
+                return n * factorial(n-1)
+
+        =======
+        >>>>>>> REPLACE
+        ```
+
+        ```python
+        ##File: /tmp/projects/mathweb/flask/app.py
+        <<<<<<< SEARCH
+            return str(factorial(n))
+        =======
+            return str(math.factorial(n))
+        >>>>>>> REPLACE
+        ```
+
+        用户需求： Refactor hello() into its own file.
+
+        回答：To make this change we need to modify `main.py` and make a new file `hello.py`:
+
+        1. Make a new hello.py file with hello() in it.
+        2. Remove hello() from main.py and replace it with an import.
+
+        Here are the *SEARCH/REPLACE* blocks:
+
+        ```python
+        ##File: /tmp/projects/mathweb/hello.py
+        <<<<<<< SEARCH
+        =======
+        def hello():
+            "print a greeting"
+
+            print("hello")
+        >>>>>>> REPLACE
+        ```
+
+        ```python
+        ##File: /tmp/projects/mathweb/main.py
+        <<<<<<< SEARCH
+        def hello():
+            "print a greeting"
+
+            print("hello")
+        =======
+        from hello import hello
+        >>>>>>> REPLACE
+        ```
+
+        现在让我们开始一个新的任务:
+
+        {%- if structure %}
+        {{ structure }}
+        {%- endif %}
+
+        {%- if content %}
+        下面是一些文件路径以及每个文件对应的源码：
+        <files>
+        {{ content }}
+        </files>
+        {%- endif %}
+
+        {%- if context %}
+        <extra_context>
+        {{ context }}
+        </extra_context>
+        {%- endif %}
+
+        下面是用户的需求：
+
+        {{ instruction }}
+
+        """
+
+    @prompt()
+    def auto_implement_function(self, instruction: str, content: str) -> str:
+        """
+        下面是一些文件路径以及每个文件对应的源码：
+
+        {{ content }}
+
+        请参考上面的内容，重新实现所有文件下方法体标记了如下内容的方法：
+
+        ```python
+        raise NotImplementedError("This function should be implemented by the model.")
+        ```
+
+        {{ instruction }}
+
+        """
+
+    def single_round_run(self, query: str, source_content: str) -> CodeGenerateResult:
+        init_prompt = ''
+        if self.args.template == "common":
+            init_prompt = self.single_round_instruction.prompt(
+                instruction=query, content=source_content, context=self.args.context
+            )
+        elif self.args.template == "auto_implement":
+            init_prompt = self.auto_implement_function.prompt(
+                instruction=query, content=source_content
+            )
+
+        with open(self.args.target_file, "w") as file:
+            file.write(init_prompt)
+
+        conversations = [{"role": "user", "content": init_prompt}]
+
+        conversations_list = []
+        results = []
+
+        for llm in self.llms:
+            v = llm.chat_ai(conversations=conversations, model=self.args.code_model)
+            results.append(v.output)
+        for result in results:
+            conversations_list.append(conversations + [{"role": "assistant", "content": result}])
+
+        return CodeGenerateResult(contents=results, conversations=conversations_list)
+
+    @prompt()
+    def multi_round_instruction(self, instruction: str, content: str, context: str = "") -> str:
+        """
+        如果你需要生成代码，对于每个需要更改的文件,你需要按 *SEARCH/REPLACE block* 的格式进行生成。
+
+        # *SEARCH/REPLACE block* Rules:
+
+        Every *SEARCH/REPLACE block* must use this format:
+        1. The opening fence and code language, eg: ```python
+        2. The file path alone on a line, starting with "##File:" and verbatim. No bold asterisks, no quotes around it,
+        no escaping of characters, etc.
+        3. The start of search block: <<<<<<< SEARCH
+        4. A contiguous chunk of lines to search for in the existing source code
+        5. The dividing line: =======
+        6. The lines to replace into the source code
+        7. The end of the replacement block: >>>>>>> REPLACE
+        8. The closing fence: ```
+
+        Every *SEARCH* section must *EXACTLY MATCH* the existing source code, character for character,
+        including all comments, docstrings, etc.
+
+        *SEARCH/REPLACE* blocks will replace *all* matching occurrences.
+        Include enough lines to make the SEARCH blocks unique.
+
+        Include *ALL* the code being searched and replaced!
+
+        To move code within a file, use 2 *SEARCH/REPLACE* blocks: 1 to delete it from its current location,
+        1 to insert it in the new location.
+
+        If you want to put code in a new file, use a *SEARCH/REPLACE block* with:
+        - A new file path, including dir name if needed
+        - An empty `SEARCH` section
+        - The new file's contents in the `REPLACE` section
+
+        ONLY EVER RETURN CODE IN A *SEARCH/REPLACE BLOCK*!
+
+        下面我们来看一个例子：
+
+        当前项目目录结构：
+        1. 项目根目录： /tmp/projects/mathweb
+        2. 项目子目录/文件列表(类似tree 命令输出)
+        flask/
+            app.py
+            templates/
+                index.html
+            static/
+                style.css
+
+        用户需求： Change get_factorial() to use math.factorial
+
+        回答： To make this change we need to modify `/tmp/projects/mathweb/flask/app.py` to:
+
+        1. Import the math package.
+        2. Remove the existing factorial() function.
+        3. Update get_factorial() to call math.factorial instead.
+
+        Here are the *SEARCH/REPLACE* blocks:
+
+        ```python
+        ##File: /tmp/projects/mathweb/flask/app.py
+        <<<<<<< SEARCH
+        from flask import Flask
+        =======
+        import math
+        from flask import Flask
+        >>>>>>> REPLACE
+        ```
+
+        ```python
+        ##File: /tmp/projects/mathweb/flask/app.py
+        <<<<<<< SEARCH
+        def factorial(n):
+            "compute factorial"
+
+            if n == 0:
+                return 1
+            else:
+                return n * factorial(n-1)
+
+        =======
+        >>>>>>> REPLACE
+        ```
+
+        ```python
+        ##File: /tmp/projects/mathweb/flask/app.py
+        <<<<<<< SEARCH
+            return str(factorial(n))
+        =======
+            return str(math.factorial(n))
+        >>>>>>> REPLACE
+        ```
+
+        用户需求： Refactor hello() into its own file.
+
+        回答：To make this change we need to modify `main.py` and make a new file `hello.py`:
+
+        1. Make a new hello.py file with hello() in it.
+        2. Remove hello() from main.py and replace it with an import.
+
+        Here are the *SEARCH/REPLACE* blocks:
+
+
+        ```python
+        ##File: /tmp/projects/mathweb/hello.py
+        <<<<<<< SEARCH
+        =======
+        def hello():
+            "print a greeting"
+
+            print("hello")
+        >>>>>>> REPLACE
+        ```
+
+        ```python
+        ##File: /tmp/projects/mathweb/main.py
+        <<<<<<< SEARCH
+        def hello():
+            "print a greeting"
+
+            print("hello")
+        =======
+        from hello import hello
+        >>>>>>> REPLACE
+        ```
+
+        现在让我们开始一个新的任务:
+
+        {%- if structure %}
+        {{ structure }}
+        {%- endif %}
+
+        {%- if content %}
+        下面是一些文件路径以及每个文件对应的源码：
+        <files>
+        {{ content }}
+        </files>
+        {%- endif %}
+
+        {%- if context %}
+        <extra_context>
+        {{ context }}
+        </extra_context>
+        {%- endif %}
+
+        下面是用户的需求：
+
+        {{ instruction }}
+
+        每次生成一个文件的*SEARCH/REPLACE* blocks，然后询问我是否继续，当我回复继续，
+        继续生成下一个文件的*SEARCH/REPLACE* blocks。当没有后续任务时，请回复 "__完成__" 或者 "__EOF__"。
+        """
+
+    def multi_round_run(self, query: str, source_content: str, max_steps: int = 3) -> CodeGenerateResult:
+        init_prompt = ''
+        if self.args.template == "common":
+            init_prompt = self.multi_round_instruction.prompt(
+                instruction=query, content=source_content, context=self.args.context
+            )
+        elif self.args.template == "auto_implement":
+            init_prompt = self.auto_implement_function.prompt(
+                instruction=query, content=source_content
+            )
+
+        with open(self.args.target_file, "w") as file:
+            file.write(init_prompt)
+
+        results = []
+        conversations = [{"role": "user", "content": init_prompt}]
+
+        code_llm = self.llms[0]
+        v = code_llm.chat_ai(conversations=conversations, model=self.args.code_model)
+        results.append(v.output)
+
+        conversations.append({"role": "assistant", "content": v.output})
+
+        if "__完成__" in v.output or "/done" in v.output or "__EOF__" in v.output:
+            return CodeGenerateResult(contents=["\n\n".join(results)], conversations=[conversations])
+
+        current_step = 0
+
+        while current_step < max_steps:
+            conversations.append({"role": "user", "content": "继续"})
+
+            with open(self.args.target_file, "w") as file:
+                file.write("继续")
+
+            t = code_llm.chat_ai(conversations=conversations, model=self.args.code_model)
+
+            results.append(t.output)
+            conversations.append({"role": "assistant", "content": t.output})
+            current_step += 1
+
+            if "__完成__" in t.output or "/done" in t.output or "__EOF__" in t.output:
+                return CodeGenerateResult(contents=["\n\n".join(results)], conversations=[conversations])
+
+        return CodeGenerateResult(contents=["\n\n".join(results)], conversations=[conversations])