Feedback: a few ideas/notes from my testing

[AlexTzk]

Great little tool, its incredibly fast and has a lot of potential. I have tested it and here are a few constructive feedback ideas from my POV:

• No relationship awareness:
  the tool is great at returning data quickly and finding functions but it's not able to identify/map relationships such as who calls function_x?

• Chunk based, not function-based;
  this logic makes the tool easy to chunk bits that may cut away the beginning or the end of a function.

• Relevance scores are low and hard to calibrate;
  in my testing, i got a bunch of scores in the range 0.008-0.026 which was quite low as a confidence score and may throw CC off.

• No test filtering;
  this kinda ties back into the aforementioned, the tool is scoring HITS in the test suite higher than the actual original function in the code which again may throw CC off

• Archictecture queries do not return architecture;
  asking project architecture overview pipeline stages returns an import block in my main application endpoint alongside a worker.py docstring which is not really a codebase map.

I hope this helps, I am happy to test again or provide more feedback if needed.

[stephantul]

Hey, thanks for your feedback,

• The tool is inherently not relational. AFAIK no semantic search tool is, you can use a tree sitter grammar to obtain these things, however.
• This is something we're actively investigating, however, just chunking based on functions leads to very large chunks.
• The scoring is used to rank, and does not represent a confidence and is not calibrated. Does it actually throw CC off?
• This looks like a genuine bug, can you open another issue describing when this happens, which languages etc.? We should by default ignore test files.
• What is an architecture query in this case?