refactor: Update context handling and repo map logic

Author: paul-gauthier

aider/coders/context_coder.py

@@ -8,6 +8,16 @@ class ContextCoder(Coder):
     edit_format = "context"
     gpt_prompts = ContextPrompts()
 
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+        if not self.repo_map:
+            return
+
+        self.repo_map.refresh = "always"
+        self.repo_map.max_map_tokens *= self.repo_map.map_mul_no_files
+        self.repo_map.map_mul_no_files = 1.0
+
     def reply_completed(self):
         content = self.partial_response_content
         if not content or not content.strip():

aider/coders/context_prompts.py

@@ -5,19 +5,41 @@
 
 class ContextPrompts(CoderPrompts):
     main_system = """Act as an expert code analyst.
-Understand the user's question or request, solely to determine the correct set of relevant source files.
-Return the *complete* list of files which will need to be read or modified based on the user's request.
+Understand the user's question or request, solely to determine ALL the existing sources files which will need to be modified.
+Return the *complete* list of files which will need to be modified based on the user's request.
 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.
 
-Be selective!
-Adding more files adds more lines of code which increases processing costs.
-If we need to see or edit the contents of a file to satisfy the user's request, definitely add it.
-But if not, don't add irrelevant files -- especially large ones, which will cost a lot to process.
+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.
+
+Only return files that will need to be modified, not files that contain useful/relevant functions.
+
+You are only to discuss EXISTING files and symbols.
+Only return existing files, don't suggest the names of new files we will need to create.
 
 Always reply to the user in {language}.
 
-Return a simple bulleted list:
+Be concise in your replies.
+Return:
+1. A bulleted list of files the will need to be edited, and symbols that are highly relevant to the user's request.
+2. A list of classes/functions/methods/variables that are located OUTSIDE those files which will need to be understood. Just the symbols names, *NOT* file names.
+
+Here an example response, use this format:
+
+## Files to modify, with their relevant symbols:
+
+- alarms/buzz.py
+  - `Buzzer` class which can make the needed sound
+  - `Buzzer.buzz_buzz()` method triggers the sound
+- alarms/time.py
+  - `Time.set_alarm(hour, minute)` to set the alarm
+
+## Relevant symbols from OTHER files:
+
+- AlarmManager class for setup/teardown of alarms
+- SoundFactory will be used to create a Buzzer
 """
 
     example_messages = []
@@ -46,6 +68,8 @@ class ContextPrompts(CoderPrompts):
 """
 
     try_again = """I have updated the set of files added to the chat.
-Review them to decide if this is the correct set of files or if we need to add more.
+Review them to decide if this is the correct set of files or if we need to add more or remove files.
+
 If this is the right set, just return the current list of files.
+Or return a smaller or larger set of files which need to be edited, with symbols that are highly relevant to the user's request.
 """

aider/commands.py

@@ -149,7 +149,7 @@ def cmd_chat_mode(self, args):
                 ),
                 (
                     "context",
-                    "Work with surrounding code context for more contextually-aware edits.",
+                    "Automatically identify which files will need to be edited.",
                 ),
             ]
         )

aider/models.py

@@ -748,7 +748,6 @@ def send_completion(self, messages, functions, stream, temperature=None):
 
         kwargs = dict(
             model=self.name,
-            messages=messages,
             stream=stream,
         )
 
@@ -779,6 +778,8 @@ def send_completion(self, messages, functions, stream, temperature=None):
             kwargs["timeout"] = request_timeout
         if self.verbose:
             dump(kwargs)
+        kwargs["messages"] = messages
+
         res = litellm.completion(**kwargs)
         return hash_object, res
 

aider/repomap.py

@@ -447,16 +447,17 @@ def get_ranked_tags(
             definers = defines[ident]
 
             mul = 1.0
-            # Check for snake_case (contains underscore, no uppercase)
-            if "_" in ident and not any(c.isupper() for c in ident):
-                mul *= 10
-            # Check for camelCase (no underscore, starts with lowercase, has uppercase)
-            elif not "_" in ident and ident[0].islower() and any(c.isupper() for c in ident):
-                mul *= 10
+
+            is_snake = ("_" in ident) and any(c.isalpha() for c in ident)
+            is_camel = any(c.isupper() for c in ident) and any(c.islower() for c in ident)
             if ident in mentioned_idents:
                 mul *= 10
+            if (is_snake or is_camel) and len(ident) >= 8:
+                mul *= 10
             if ident.startswith("_"):
                 mul *= 0.1
+            if len(defines[ident]) > 5:
+                mul *= 0.1
 
             for referencer, num_refs in Counter(references[ident]).items():
                 for definer in definers: