#Good morning I m using VS Code on Ubuntu

1 messages · Page 1 of 1 (latest)

timid obsidian
#

Can you show an example of this

fringe prawn
#

On a related question, is there a way of viewing progress of the indexing?

#

First thing I wanted to look for is confirmation that the codebase was indexed, as the search was returning super limited results, as though its index was incomplete.

fringe prawn
#

Search enabled for workspace:

#

A basic question asked via chat:

#

How the workspace is configured to support search:

#

@timid obsidian Anything else specific that would be useful to see?

#

Here's something interesting...

#

The last question I was just correcting a typo to see what would happen. Oddly it answered the questions very differently.

timid obsidian
#

@spring lava

fringe prawn
#

Thanks @timid obsidian. Any ideas @spring lava?

#

I can grab more information and screens if that helps.

spring lava
#

Yes I think we may have an issue in our logic to decide whether you are asking a general question or a question about your codebase

#

Do you mind retrying the same question few times, clearing the chat history in between and seeing if it ever says something like "Searching X paths"

fringe prawn
#

@spring lava No problem.
While I test that, here is a response which shows a slightly different problem. The context appears to have only checked 4 files instead of the entire codebase:

#

Second time, it searches 6 files:

#

Third time, up to 9 files:

#

Fourth time, the pattern changes and I'm back down to 7:

#

The only thing I am doing between tries is clearing the conversation.

#

No files are open in the editor.

#

I am in a workspace with search enabled.

spring lava
#

Ah in general context chat is not great at answering questions about the codebase holistically, due to limited context length

#

There's no way for it to answer a question that requires simultaneously looking at 5000 files of code, for example

#

But it should be able to answer that question for a narrower scope like, "Find all the functions in 'file x' that could be simplified"

#

To the point about changing behavior, the LLM is not entirely deterministic so it will try different retrieval approaches to the question every time it's asked

fringe prawn
#

RIght! Thanks for clarifying.
My understanding was that with search enabled, chat would use the search index to help answer questions. Is that correct?
But, I think I understand what you're saying. Chat wouldn't know from its index (if its available to chat) which functions are complex and therefore can be simplified, it would have to check each file to figure that out.
Have I understood that correctly?
Do you have examples of what it could do across the entire codebase quickly, via search or chat?

#

I suppose what would be useful there is to understand what type of information gets indexed.

#

Thanks for your help with this @spring lava!

spring lava
#

Yes that's a good question -- we index the code itself, but not necessarily "Suggestions" about the code like "this function is too verbose" or "this function lacks a docstring". that's something we are looking into @feral spear

#
fringe prawn
#

Great, thanks. I have this URL bookmarked and have been trying to understand the suggestions. I will go back and try to figure out where the limitations of chat and the index are likely to be and form some more relevant questions to try.
@spring lava Thanks for your help.

spring lava
#

Of course these are limitations of the current product. We very much appreciate any thougts you have on the capabilities we should be investing in in the future

#

as I said, scanning the entire codebase for a certain kind of issue is something other users have mentioned, and we are currently exploring the right approchaes here

fringe prawn
#

Other more complex things to calculate and index (I imagine) like test coverage would be nice.

#

And general "clean code" design principles (well named functions, few parameters...).