Use CodeQL or grep search for potential affected places in code across repository

[Konard]

This will dramatically reduce the number of requests to GPT-4 directly for huge projects.
Or maybe it should be also guided by user to look at specific places as in #31

Or maybe we should build the graph of code with all implicit and explicit dependencies, including monkey patches and so on. And the moonshot idea - to see how users use the code in other repositories.

Related to: #90

[irgolic]

There's a couple ways we could improve this, yeah. I think some of the lowest hanging fruit is:

• semantic search (among file summaries)
• Global CTRL+F style search

I outline some of this in #55.

When you talk about a code graph, at what part of the process would you use it, and how would you query it?

[Konard]

The idea for code graph is updated it each time files in the repository are changed.
The graph should have links that define which function/code block depends on which other code block/function/field. The simple select query will directly give an answer that code should be investigated before making a change. In this query, we also will be able to directly get relevant code parts.
For storing the code graph I recommend to use associative storage from Deep as in https://deep.foundation

[irgolic]

Right now it only operates over filenames, so better context gathering from something akin to a code graph will improve it a lot. It'd be great to include the ability to query context from the code graph during both the planning and code writing stages.

Things to check out:

• https://lsif.dev/
• https://about.sourcegraph.com/blog/announcing-scip