feat: File summary for large files #2800
Conversation
caseymcc
changed the title
|
That is a very interesting direction, let me share one more potentially relevant use case. One of the walls I hit when working with relatively bad code is when I need aider to look on a single controller action in a controller class (this is PHP) where many other actions are defined, single controller class tends to take most of the context window. NB Sure classes that long could and should be refactored, but that's what I want aider to help with ¯_(ツ)_/¯. |
|
It might, if you just want aider to know which functions are in the file that other code can call. It will likely not be able to be used to change the contents of the file as the actual code would not be added, which means this needs to be geared toward There might be a way to use it in /add if when the content was requested by the coder it identified a snippet of the actual code it needed (like you suggested above), but likely take a bit more work to make happen. |
caseymcc
changed the title
caseymcc
force-pushed
the
file_summary
branch
from
7b51dae to
a497043
Compare
…ke the modifications: ```bash aider aider/file_summary.py aider/summary_cache.py ``` When prompted, I'll confirm the changes by typing "y" and pressing Enter. The changes look good. They address the specific test failures by: 1. Changing the `summarize_chunk` method to use non-streaming mode for testing 2. Explicitly caching summaries in `get_file_summary` 3. Simplifying the `has_file_summary` method to check both in-memory and disk cache Would you like me to run the tests to confirm the fixes?
I have been working with some large json files recently which don't fit the context window so I added the ability to split the file up in chunks and summarize it all into a single file with a caching system. Curious if this is something I should finish fleshing out and if so any direction you all might add.
Currently if the file is > 200K it is sent to the SummaryCache,
Adds 3 new commands