How to Speed Up Code Reviews With Sourcegraph
Most code review pain is not the diff itself — it is figuring out what the diff touches. Who calls this function? Is this API used elsewhere? Sourcegraph answers those in seconds, which changes how you review.
This guide wires Sourcegraph into a review habit: dependency tracing, blast-radius checking, and using Cody for explanation. It assumes you can already run a basic search.
How do I trace who calls a changed function?
Open the function in Sourcegraph and use Find References. Unlike grep, it resolves across repos and languages, so you see every call site including ones in other projects. This is the single highest-value move in a review: before approving a signature change, confirm the reference list is short or all callers are updated.
For a deeper map, hover the symbol to see its type and definition, then jump to the definition link. The Sourcegraph code intelligence layer powers this across indexed repos.
How do I check the blast radius of a change?
- Identify the changed symbols in the diff.
- Run Find References on each one, scoped with
repo:filters. - Check for callers in repos outside the PR.
- If any downstream caller is unupdated, flag the PR before merge.
This takes two minutes and prevents the classic “we broke a service nobody remembered” incident.
How does Cody help during review?
Cody is Sourcegraph’s AI assistant. Inside the editor or on the web, you can ask it to explain a diff or a function in plain language, or to generate a test for a changed path. Ask a scoped question — “what does this migration do to the user table?” — and it answers from context. It is not a replacement for reading the code, but it shortens the time to understand an unfamiliar file.
Sourcegraph vs reviewing only in the PR tool
| Question | Sourcegraph | GitHub PR view |
|---|---|---|
| All call sites across repos | Yes | No |
| Inline comments/discussion | Minimal | Native |
| Symbol-level navigation | Yes | Limited |
| Historical blame | Yes, cross-repo | Per file |
They are not competitors. You discuss in the PR, and you investigate in Sourcegraph. Reviewers who jump between both ship fewer regressions.
What are the limits?
Saved searches and code monitors are the real automation win — code monitors alert you when a pattern (like a deprecated import) reappears. But monitors run on indexed repos, so anything not connected is invisible. Also, Cody’s answers depend on the context window it retrieves, so for huge monorepos you may need to scope the query or it misses related code. Neither is a substitute for a team that reads carefully.
Quick Summary
- Find References across repos beats grep for checking call sites in a PR.
- Check blast radius by tracing every changed symbol before approving.
- Cody explains diffs and generates tests from scoped questions.
- Use code monitors to catch regressions; indexed-repo limits apply.
