Slow performance for viewing changes for large git repos

[amay]

Hello, I'm trying out this project to help me orchestrate agent work async so I can scale my output and we have a very large git repo. I was noticing 70s+ delays in viewing diffs for task run code changes. I've used claude to help me dig into the implementation and have a changeset I can share that takes that time down to ~0.5s (136x speed up). Here is a commit you can look at to see what I did.

Note: I don't have good internal understanding of the project so I have low context on whether this approach is backwards compatible (I did have claude analyze this). Really appreciate the hard work the team has put into this!

Here is an analysis from claude on the change:

---

Performance Optimization: Git Diff Operation

Issue Identified

The application was experiencing severe performance issues when viewing diffs, with loading times exceeding one minute. Investigation revealed that the bottleneck was in the diff_status method in git_cli.rs, specifically the git add -A operation which was taking 70-75 seconds to complete on large repositories.

Root Cause

The original implementation used a temporary Git index and executed git add -A to stage all files before computing diffs. This approach is inefficient for large repositories as it processes all files regardless of whether they've changed.

Solution Implemented

Replaced the expensive git add -A operation with a targeted approach that only adds changed files:

1. Use git status --porcelain to identify which files have actually changed
2. Parse the status output to extract file paths
3. Add only the changed files to the Git index in batches of 100
4. Properly handle renamed files by extracting both old and new paths

Results

• Before: ~75 seconds to generate diffs
• After: ~0.5-0.55 seconds (136x performance improvement)
• The approach maintains full backward compatibility with the existing design
• No functional changes to the API or output format

This optimization is especially beneficial for large repositories, where the git add -A operation was scanning and processing all files unnecessarily.