Merge fbdf3ce823 into 3caab85931

aider/args.py

@@ -263,6 +263,21 @@ def get_parser(default_config_files, git_root):
         help="Multiplier for map tokens when no files are specified (default: 2)",
     )
 
+    ##########
+    group = parser.add_argument_group("Auto mode settings")
+    group.add_argument(
+        "--deep-context-search",
+        action=argparse.BooleanOptionalAction,
+        default=True,
+        help="Enable/disable enhanced context finding in auto mode (default: True)",
+    )
+    group.add_argument(
+        "--min-identifier-length",
+        type=int,
+        default=3,
+        help="Minimum length of identifiers to consider for context finding (default: 3)",
+    )
+
     ##########
     group = parser.add_argument_group("History Files")
     default_input_history_file = (

aider/coders/__init__.py

@@ -1,5 +1,6 @@
 from .architect_coder import ArchitectCoder
 from .ask_coder import AskCoder
+from .auto_coder import AutoCoder
 from .base_coder import Coder
 from .context_coder import ContextCoder
 from .editblock_coder import EditBlockCoder
@@ -31,4 +32,5 @@ __all__ = [
     EditorWholeFileCoder,
     EditorDiffFencedCoder,
     ContextCoder,
+    AutoCoder,
 ]

aider/coders/auto_coder.py

@@ -0,0 +1,165 @@
+from .context_coder import ContextCoder
+from .auto_prompts import AutoPrompts
+import re
+from pathlib import Path
+
+
+class AutoCoder(ContextCoder):
+    """Automatically identify files and make changes without confirmation."""
+
+    edit_format = "auto"
+    gpt_prompts = AutoPrompts()
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        # Set yes_to_all to bypass confirmations
+        self.io.yes = True
+
+        # Ensure auto_accept_architect is True
+        self.auto_accept_architect = True
+
+        # Enable auto-linting and auto-testing if configured
+        self.auto_lint = kwargs.get('auto_lint', True)
+        self.auto_test = kwargs.get('auto_test', False)
+
+        # Enhanced context finding settings
+        self.deep_context_search = kwargs.get('deep_context_search', True)
+        self.min_identifier_length = kwargs.get('min_identifier_length', 3)  # Shorter than default (5)
+
+        # Increase repo map tokens for better context
+        if self.repo_map:
+            self.repo_map.max_map_tokens *= 1.5  # Increase token allocation for repo map
+            self.repo_map.refresh = "always"  # Always refresh the repo map
+
+    def get_enhanced_file_mentions(self, content):
+        """Enhanced method to find file mentions in content with better heuristics."""
+        # Get standard file mentions
+        standard_mentions = self.get_file_mentions(content, ignore_current=True)
+
+        # Get identifiers that might be related to files
+        identifiers = self.get_ident_mentions(content)
+
+        # Use a lower threshold for identifier length
+        all_fnames = {}
+        for fname in self.get_all_relative_files():
+            if not fname or fname == ".":
+                continue
+
+            try:
+                path = Path(fname)
+
+                # Add the file's stem (name without extension)
+                base = path.stem.lower()
+                if len(base) >= self.min_identifier_length:
+                    if base not in all_fnames:
+                        all_fnames[base] = set()
+                    all_fnames[base].add(fname)
+
+                # Add the file's parent directory name
+                if path.parent.name:
+                    parent = path.parent.name.lower()
+                    if len(parent) >= self.min_identifier_length:
+                        if parent not in all_fnames:
+                            all_fnames[parent] = set()
+                        all_fnames[parent].add(fname)
+
+                # Add the full path components
+                parts = [p.lower() for p in path.parts if p and len(p) >= self.min_identifier_length]
+                for part in parts:
+                    if part not in all_fnames:
+                        all_fnames[part] = set()
+                    all_fnames[part].add(fname)
+            except ValueError:
+                continue
+
+        # Match identifiers to files
+        identifier_matches = set()
+        for ident in identifiers:
+            ident_lower = ident.lower()
+            if len(ident_lower) >= self.min_identifier_length and ident_lower in all_fnames:
+                identifier_matches.update(all_fnames[ident_lower])
+
+        # Look for import statements and package references
+        import_pattern = re.compile(r'(?:import|from|require|include)\s+([a-zA-Z0-9_.]+)')
+        imports = import_pattern.findall(content)
+
+        import_matches = set()
+        for imp in imports:
+            parts = imp.split('.')
+            for i in range(len(parts)):
+                partial = '.'.join(parts[:i+1])
+                partial_lower = partial.lower()
+                if partial_lower in all_fnames:
+                    import_matches.update(all_fnames[partial_lower])
+
+                # Also check for file extensions
+                for ext in ['.py', '.js', '.ts', '.java', '.c', '.cpp', '.h', '.hpp']:
+                    with_ext = partial + ext
+                    with_ext_lower = with_ext.lower()
+                    if with_ext_lower in all_fnames:
+                        import_matches.update(all_fnames[with_ext_lower])
+
+        # Combine all matches
+        all_matches = standard_mentions | identifier_matches | import_matches
+
+        return all_matches
+
+    def reply_completed(self):
+        # First use ContextCoder's functionality to identify relevant files
+        content = self.partial_response_content
+        if not content or not content.strip():
+            return True
+
+        # Get files mentioned in the response using enhanced methods
+        current_rel_fnames = set(self.get_inchat_relative_files())
+
+        if self.deep_context_search:
+            mentioned_rel_fnames = self.get_enhanced_file_mentions(content)
+        else:
+            mentioned_rel_fnames = set(self.get_file_mentions(content, ignore_current=True))
+
+        # If the files are different, automatically add the mentioned files
+        if mentioned_rel_fnames != current_rel_fnames:
+            self.abs_fnames = set()
+            for fname in mentioned_rel_fnames:
+                self.add_rel_fname(fname)
+
+            # Now that we've added the files, we need to get the content again
+            # and apply the changes automatically
+            self.io.tool_output(f"Automatically added files: {', '.join(mentioned_rel_fnames)}")
+
+            # Refresh the repository map if needed
+            if self.repo_map:
+                self.get_repo_map(force_refresh=True)
+
+            # Create a new message to apply the changes
+            self.reflected_message = "I've identified the relevant files. Now I'll make the requested changes."
+            return False
+
+        # If we already have all the files, apply the changes
+        edited = self.apply_updates()
+
+        if edited:
+            self.io.tool_output(f"Automatically applied changes to: {', '.join(edited)}")
+            self.aider_edited_files.update(edited)
+            saved_message = self.auto_commit(edited)
+
+            if saved_message:
+                self.move_back_cur_messages(saved_message)
+
+            # Run linting if enabled
+            if self.auto_lint:
+                lint_errors = self.lint_edited(edited)
+                if lint_errors:
+                    self.io.tool_output("Linting found errors. Attempting to fix...")
+                    self.reflected_message = lint_errors
+                    return False
+
+            # Run tests if enabled
+            if self.auto_test:
+                test_output = self.run_tests()
+                if test_output:
+                    self.io.tool_output(f"Test results: {test_output}")
+
+        return True

aider/coders/auto_prompts.py

@@ -0,0 +1,52 @@
+# flake8: noqa: E501
+
+from .context_prompts import ContextPrompts
+
+
+class AutoPrompts(ContextPrompts):
+    main_system = """Act as an expert code analyst and developer with deep understanding of software architecture.
+First, thoroughly analyze the user's request to determine ALL existing source files which will need to be modified or referenced.
+Then, make the necessary changes to implement the requested feature or fix the issue.
+
+Your task has two phases:
+1. Identify all relevant files that need to be modified or referenced
+2. Make the necessary changes to implement the requested feature or fix
+
+For phase 1 (Context Discovery):
+- Perform a comprehensive analysis to identify ALL files that might be relevant
+- Consider not just files that need direct modification, but also:
+  * Files containing related classes, interfaces, or types
+  * Files with dependent functionality
+  * Configuration files that might affect the behavior
+  * Test files that will need to be updated
+- Return the *complete* list of files which will need to be modified or referenced
+- Explain why each file is needed, including names of key classes/functions/methods/variables
+- Be sure to include or omit the names of files already added to the chat, based on whether they are actually needed or not
+- Think about imports, inheritance hierarchies, and dependency relationships
+
+For phase 2 (Implementation):
+- Implement the requested changes in the identified files
+- Follow the codebase's style and conventions
+- Ensure your changes are complete and functional
+- Consider edge cases and error handling
+- Update any related tests
+- Explain the changes you've made and why they address the user's request
+
+The user will use every file you mention, regardless of your commentary.
+So *ONLY* mention the names of relevant files.
+If a file is not relevant DO NOT mention it.
+
+Remember to consider:
+- Class hierarchies and inheritance relationships
+- Interface implementations
+- Import dependencies
+- Configuration settings
+- Related test files
+"""
+
+    system_reminder = """Remember to:
+1. First identify ALL relevant files needed for the task
+2. Then implement the changes
+3. Only mention file names that are actually relevant
+4. Consider dependencies, imports, and inheritance relationships
+"""

aider/coders/base_coder.py

@@ -335,6 +335,8 @@ class Coder:
         file_watcher=None,
         auto_copy_context=False,
         auto_accept_architect=True,
+        deep_context_search=True,
+        min_identifier_length=3,
     ):
         # Fill in a dummy Analytics if needed, but it is never .enable()'d
         self.analytics = analytics if analytics is not None else Analytics()
@@ -349,6 +351,10 @@ class Coder:
         self.auto_copy_context = auto_copy_context
         self.auto_accept_architect = auto_accept_architect
 
+        # Auto mode settings
+        self.deep_context_search = deep_context_search
+        self.min_identifier_length = min_identifier_length
+
         self.ignore_mentions = ignore_mentions
         if not self.ignore_mentions:
             self.ignore_mentions = set()
@@ -592,6 +598,7 @@ class Coder:
 
     def get_abs_fnames_content(self):
         for fname in list(self.abs_fnames):
+            try:
                 content = self.io.read_text(fname)
 
                 if content is None:
@@ -600,15 +607,25 @@ class Coder:
                     self.abs_fnames.remove(fname)
                 else:
                     yield fname, content
+            except UnicodeDecodeError:
+                # Skip binary files that can't be decoded as text
+                relative_fname = self.get_rel_fname(fname)
+                self.io.tool_warning(f"Dropping binary file {relative_fname} from the chat.")
+                self.abs_fnames.remove(fname)
 
     def choose_fence(self):
         all_content = ""
         for _fname, content in self.get_abs_fnames_content():
             all_content += content + "\n"
         for _fname in self.abs_read_only_fnames:
+            try:
                 content = self.io.read_text(_fname)
                 if content is not None:
                     all_content += content + "\n"
+            except UnicodeDecodeError:
+                # Skip binary files that can't be decoded as text
+                relative_fname = self.get_rel_fname(_fname)
+                self.io.tool_warning(f"Skipping binary file {relative_fname} when choosing fence.")
 
         lines = all_content.splitlines()
         good = False
@@ -654,6 +671,7 @@ class Coder:
     def get_read_only_files_content(self):
         prompt = ""
         for fname in self.abs_read_only_fnames:
+            try:
                 content = self.io.read_text(fname)
                 if content is not None and not is_image_file(fname):
                     relative_fname = self.get_rel_fname(fname)
@@ -662,6 +680,10 @@ class Coder:
                     prompt += f"\n{self.fence[0]}\n"
                     prompt += content
                     prompt += f"{self.fence[1]}\n"
+            except UnicodeDecodeError:
+                # Skip binary files that can't be decoded as text
+                relative_fname = self.get_rel_fname(fname)
+                self.io.tool_warning(f"Skipping binary file {relative_fname} from read-only files.")
         return prompt
 
     def get_cur_message_text(self):
@@ -1706,6 +1728,11 @@ class Coder:
                 )
             ]
 
+    def get_enhanced_file_mentions(self, content):
+        """Base implementation of enhanced file mentions - just returns standard file mentions.
+        This method is overridden in AutoCoder to provide more sophisticated context finding."""
+        return self.get_file_mentions(content, ignore_current=True)
+
     def get_file_mentions(self, content, ignore_current=False):
         words = set(word for word in content.split())
 

aider/commands.py

@@ -163,6 +163,10 @@ class Commands:
                     "context",
                     "Automatically identify which files will need to be edited.",
                 ),
+                (
+                    "auto",
+                    "Automatically identify files and make changes without confirmation.",
+                ),
             ]
         )
 
@@ -1157,6 +1161,9 @@ class Commands:
     def completions_context(self):
         raise CommandCompletionException()
 
+    def completions_auto(self):
+        raise CommandCompletionException()
+
     def cmd_ask(self, args):
         """Ask questions about the code base without editing any files. If no prompt provided, switches to ask mode."""  # noqa
         return self._generic_chat_command(args, "ask")
@@ -1169,6 +1176,10 @@ class Commands:
         """Enter architect/editor mode using 2 different models. If no prompt provided, switches to architect/editor mode."""  # noqa
         return self._generic_chat_command(args, "architect")
 
+    def cmd_auto(self, args):
+        """Enter auto mode to automatically identify files and make changes without confirmation. If no prompt provided, switches to auto mode."""  # noqa
+        return self._generic_chat_command(args, "auto")
+
     def cmd_context(self, args):
         """Enter context mode to see surrounding code context. If no prompt provided, switches to context mode."""  # noqa
         return self._generic_chat_command(args, "context", placeholder=args.strip() or None)

aider/exceptions.py

@@ -58,8 +58,10 @@ class LiteLLMExceptions:
         self._load()
 
     def _load(self, strict=False):
+        # Import litellm - json.load is already patched in aider.llm
         import litellm
 
+        # Load litellm exceptions
         for var in dir(litellm):
             if var.endswith("Error"):
                 if var not in self.exception_info:

aider/llm.py

@@ -1,6 +1,8 @@
 import importlib
+import json
 import os
 import warnings
+from pathlib import Path
 
 from aider.dump import dump  # noqa: F401
 
@@ -17,6 +19,27 @@ os.environ["LITELLM_MODE"] = "PRODUCTION"
 
 VERBOSE = False
 
+# Patch json.load to handle UTF-8 encoding for litellm
+original_json_load = json.load
+
+def patched_json_load(fp, *args, **kwargs):
+    try:
+        # First try the original method
+        return original_json_load(fp, *args, **kwargs)
+    except UnicodeDecodeError:
+        # If it fails with UnicodeDecodeError, try with UTF-8 encoding
+        try:
+            # Read the file content with UTF-8 encoding
+            content = Path(fp.name).read_text(encoding='utf-8')
+            # Parse the content as JSON
+            return json.loads(content, *args, **kwargs)
+        except Exception:
+            # If that also fails, re-raise the original exception
+            raise
+
+# Apply the monkey patch
+json.load = patched_json_load
+
 
 class LazyLiteLLM:
     _lazy_module = None

aider/main.py

@@ -996,6 +996,8 @@ def main(argv=None, input=None, output=None, force_git_root=None, return_coder=F
             detect_urls=args.detect_urls,
             auto_copy_context=args.copy_paste,
             auto_accept_architect=args.auto_accept_architect,
+            deep_context_search=args.deep_context_search,
+            min_identifier_length=args.min_identifier_length,
         )
     except UnknownEditFormat as err:
         io.tool_error(str(err))