Open
Conversation
Introduces `codesearch explain <symbol>` which combines impact analysis with an LLM to produce a structured explanation of a symbol's purpose, data flow, and business requirements. How it works: - Runs BFS impact analysis (same engine as `impact`) to build the call graph up to a configurable depth (default: 3). - Resolves the root symbol's source file via its callees, then reads a 40-line window around each reference site directly from disk. - Sends a structured Markdown prompt — root source + depth-1 caller sources + summary of deeper nodes — to the configured LLM backend. - Supports both Anthropic (/v1/messages) and OpenAI-compatible (/v1/chat/completions, e.g. LM Studio) endpoints via `--llm`. - Opens the database read-only (no write lock) so it can run alongside concurrent indexing or search processes. New files: src/connector/api/controller/explain_controller.rs Changed files: src/cli/mod.rs — Explain command variant src/connector/api/controller/mod.rs — export ExplainController src/connector/api/router.rs — route Commands::Explain src/main.rs — read-only flag for Explain Usage: codesearch explain authenticate codesearch explain MyStruct::new --depth 5 --llm open-ai codesearch explain process_payment --repository my-repo https://claude.ai/code/session_01494XuGs5Ez5SvRxtaN8dFy
Add a new trait method for retrieving all chunks for a given file path without performing a similarity search. Useful for snippet-lookup use cases (e.g. TUI reference navigation). - Trait default: no-op returning empty vec for backwards compatibility - DuckDB adapter: SQL query with optional repository_id filter, ordered by start_line - InMemory adapter: HashMap filter with matching sort
Align method ordering in DuckdbVectorRepository impl block with the convention established in claude/fix-duckdb-tui-state-scA0C: flush and count precede find_chunks_by_file.
The enum controls which LLM provider is used for any LLM call (query expansion, explain command, etc.), so the narrower name was misleading. - cli: QueryExpansionTarget → LlmTarget - container: query_expansion_target field → llm_target - main: expand_query_target field → llm_target - lib: update re-export - explain_controller: update import and match arms
Drop the max_depth cap so the call graph walk runs until all reachable callers are visited. Removes the --depth flag from the impact and explain CLI commands and the depth field from the MCP ImpactToolInput schema.
Replace the flat depth-1-only snippet approach with path-based source gathering. reconstruct_paths() traces each leaf back to the root via via_symbol links (same algorithm as ImpactController), producing one Vec<&ImpactNode> per call chain ordered outermost-caller-first. build_prompt() then: - Collects unique symbols across all paths and reads a source window for each (capped at MAX_UNIQUE_SYMBOLS_WITH_SOURCE=20 to bound prompt size; no depth limit is applied to the traversal itself). - Renders every path as a chain header (A → B → … → root_symbol) with an inline source block per node, giving the LLM full context for each call chain rather than just the first five depth-1 callers.
Read source for every unique symbol across all call paths with no artificial limit. The exploration is now fully unbounded end to end.
📝 WalkthroughWalkthroughThe PR introduces a file-scoped chunk lookup capability via a new VectorRepository method, removes depth constraints from impact analysis traversal, renames the LLM-provider enum for generality, adds a new Explain command with an accompanying ExplainController for LLM-based call-flow analysis, and removes depth parameters from the CLI Impact command and related method signatures. Changes
Sequence DiagramsequenceDiagram
participant CLI as CLI/Router
participant EC as ExplainController
participant IA as ImpactAnalysis
participant VR as VectorRepository
participant LLM as LLM Service
CLI->>EC: explain(symbol, repo, llm_target)
EC->>IA: analyze(symbol, repo)
IA->>VR: query for symbol calls
VR-->>IA: return call graph
IA-->>EC: return ImpactAnalysis
alt No callers found
EC-->>CLI: return "no callers" message
else Callers found
EC->>VR: find_chunks_by_file(repo, file)
VR-->>EC: return code chunks
EC->>EC: reconstruct_paths(impact_nodes)
EC->>EC: build_prompt(symbol, paths, sources)
EC->>LLM: query with prompt + system context
LLM-->>EC: return explanation
EC->>EC: format with metrics
EC-->>CLI: return formatted explanation
end
Estimated code review effort🎯 4 (Complex) | ⏱️ ~50 minutes Possibly related PRs
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 inconclusive)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 3
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@src/connector/api/controller/explain_controller.rs`:
- Around line 139-143: The current maps (node_by_depth_symbol and the similar
map at lines ~199-213) use keys of (node.depth, node.symbol) which can collide
across files/repos; change the key to include a unique source identifier (e.g.
node.source_id, node.file_path, repo_id or node.id) so keys become (node.depth,
node.symbol.as_str(), node.source_id.as_str()) or equivalent unique field;
update the HashMap type signatures and all .entry(...) calls and lookups (both
where node_by_depth_symbol is built and in the later map at ~199-213) to use
that expanded tuple key so path reconstruction and snippet lookup use the full
identity rather than symbol+depth only.
- Around line 75-79: The call to
call_graph.find_callees(...).await.unwrap_or_default() in explain_controller.rs
is swallowing errors; change it to capture the Result, log the error with
tracing::warn! or tracing::error! (including the error value and context like
analysis.root_symbol and cg_query) and then fallback to an empty Vec only after
logging; locate the invocation around the variable callees (the
call_graph.find_callees call) and replace the unwrap_or_default() with explicit
error handling (e.g., match or if let Err(e) = ...) that logs the error before
assigning the default.
- Around line 252-256: Clamp the computed center index to the valid range before
slicing: when computing `center` from `center_line` (the existing `let center =
center_line.saturating_sub(1) as usize`), replace that with a clamped value
using `lines.len().saturating_sub(1)` to ensure `center <= last_index`; then
recompute `start` and `end` (as you already do) and add an early-return check
(e.g., return None) if `start >= end` to avoid slicing with out-of-range bounds
in the function that builds the source window (the block using `center`, `half`,
`start`, `end`, and `lines[start..end].join("\n")`).
ℹ️ Review info
⚙️ Run configuration
Configuration used: defaults
Review profile: CHILL
Plan: Pro
Run ID: fd915bd3-c7e4-43de-bab1-d2203c0313a0
📒 Files selected for processing (13)
src/application/interfaces/vector_repository.rssrc/application/use_cases/impact_analysis.rssrc/cli/mod.rssrc/connector/adapter/duckdb_vector_repository.rssrc/connector/adapter/in_memory_vector_repository.rssrc/connector/adapter/mcp/server.rssrc/connector/api/container.rssrc/connector/api/controller/explain_controller.rssrc/connector/api/controller/impact_controller.rssrc/connector/api/controller/mod.rssrc/connector/api/router.rssrc/lib.rssrc/main.rs
💤 Files with no reviewable changes (1)
- src/application/use_cases/impact_analysis.rs
…urce window - Expand node_by_depth_symbol key from (depth, symbol) to (depth, symbol, repository_id) so the same symbol name at the same depth in different repositories no longer clobbers each other. - Expand source_cache and dedup seen-set keys from symbol to (symbol, file_path) for the same reason. - Replace unwrap_or_default() on find_callees with explicit match that logs the error via tracing::warn! before falling back to an empty Vec. - Clamp center index to lines.len()-1 in read_source_window and add an early None return when start >= end to prevent an out-of-range slice panic when a stored line number exceeds the actual file length.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary by CodeRabbit
Release Notes
New Features
Improvements