Navigating a codebase with an API

Robert McKnight — Wed, 06 Nov 2024 20:23:34 +0000

Today, I'm going to show you how to navigate a code workspace with lsproxy, an open-source dev container for code exploration workflows like AI coding agents.

Installation and setup

Download the python sdk.

pip install lsproxy-sdk

Pull and run the docker container. WORKSPACE_PATH is the absolute path to your code workspace (likely the repository root).

docker pull agenticlabs/lsproxy

docker run \
    -p 4444:4444 \
    -v $WORKSPACE_PATH:/mnt/workspace \
    agenticlabs/lsproxy

Set up the client in python.

# First create the API client
from lsproxy import (
    Lsproxy,
    FileRange,
    GetReferencesRequest,
    GetDefinitionRequest,
)

client = Lsproxy()

Find definitions in a file

symbols = client.definitions_in_file("src/sample_file.rs")

Gives you lots of information about the symbols defined within.

Get source code and references

# Pull source code
file_range = FileRange(
    path="src/sample_file.rs",
    start=symbol.range.start,
    end=symbol.range.end,
)
source_code = client.read_source_code(file_range).source_code

# Get references with surrounding code context
reference_request = GetReferencesRequest(
    identifier_position=symbol.identifier_position,
    include_code_context_lines=2,
)
references = client.find_references(reference_request)

Returns definition source code and references across the workspace. For example if we chose UploadIngestionMessage symbol defined in the file.

Code

pub struct UploadIngestionMessage {
    pub ingest_specific_chunk_metadata: IngestSpecificChunkMetadata,
    pub chunk: ChunkReqPayload,
    pub dataset_id: uuid::Uuid,
    pub upsert_by_tracking_id: bool,
}

References
server/src/bin/ingestion-worker.rs

   15: use trieve_server::handlers::chunk_handler::{
   16:     BulkUploadIngestionMessage, FullTextBoost, SemanticBoost, UpdateIngestionMessage,
   17:     UploadIngestionMessage,
     : ____^
   18: };
   19: 
@@@@@: -----
  793: #[tracing::instrument(skip(payload, web_pool))]
  794: async fn upload_chunk(
  795:     mut payload: UploadIngestionMessage,
     : _________________^
  796:     dataset_config: DatasetConfiguration,
  797: 
@@@@@: -----
server/src/handlers/chunk_handler.rs
  210:     pub attempt_number: usize,
  211:     pub dataset_id: uuid::Uuid,
  212:     pub ingestion_messages: Vec<UploadIngestionMessage>,
     : ________________________________^
  213: }
  214:

server/src/operators/chunk_operator.rs

    4: };
    5: use crate::handlers::chunk_handler::{BulkUploadIngestionMessage, ChunkReqPayload};
    6: use crate::handlers::chunk_handler::{ChunkFilter, UploadIngestionMessage};
     : __________________________________________________^
    7: use crate::operators::group_operator::{
    8: 
@@@@@: -----
 2514:         chunk_only_group_ids.group_tracking_ids = None;
 2515: 
 2516:         let upload_message = UploadIngestionMessage {
     : _____________________________^
 2517:             ingest_specific_chunk_metadata: IngestSpecificChunkMetadata {
 2518:

Go to definition

If you have a reference to a symbol and want to find its definition, you can do that, too.

definition_request = GetDefinitionRequest(position=reference)

# Some languages allow multiple definitions so we get a list
definitions = client.find_definition(definition_request).definitions

Check out the interactive tutorial to see this and more in-depth examples, and the API spec for supported endpoints:

Tutorial: https://demo.lsproxy.dev
API Spec: https://docs.lsproxy.dev/api-reference

DEV Community: Robert McKnight

Navigating a codebase with an API

Installation and setup

Find definitions in a file

Get source code and references

Go to definition