Skip to content

Bench/issue 2296 completion benchmark #4178

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
kz357 wants to merge 2 commits into
base: main
Choose a base branch
from
+299 −13
Open

Bench/issue 2296 completion benchmark #4178

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
a3342dd
Fix short reads of large LSP responses
kz357 Jul 17, 2026
fc27de8
Add completion mode to the LSP benchmark
kz357 Jul 17, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 17 additions & 1 deletion scripts/benchmark/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -56,13 +56,20 @@ Results are saved as JSON in `results/` (or the directory specified by `-o`).
Each run produces a dated file (e.g., `benchmark_2025-03-17.json`) and a
`latest.json` symlink.

## LSP Benchmark (Go to Definition)
## LSP Benchmark (Go to Definition / Completion)

Measures `textDocument/definition` latency across LSP servers. For each run,
the script picks a random Python file and identifier, starts each server,
and times the Go to Definition response. All servers are run in parallel on
the same position for fair comparison.

With `--mode completion` it instead measures `textDocument/completion`. Each run
picks a member access on an imported module (e.g. `pl.DataFrame` after `import polars as pl`),
truncates the member to `--prefix-len` chars, and requests completion at that point, so the
server sees a partially-typed member access (in this case `pl.D`) exactly as it would while
you type. Because the original member name is known, the benchmark also checks that the
server offered it back, which is what the `valid` column reports in completion mode.

`lsp_benchmark.py` has two modes:

1. **Multi-package mode** (`--install-envs`) — auto-clones packages from
Expand Down Expand Up @@ -111,6 +118,13 @@ python3 lsp_benchmark.py \
--pyrefly-cmd "pyrefly lsp" \
--seed 42 --runs 50

# Benchmark completion (`pl.d`-style member access) instead of Go to Definition
python3 lsp_benchmark.py \
--root ~/projects/some-python-repo \
--pyrefly-cmd "pyrefly lsp" \
--mode completion \
--seed 42 --runs 50

# All four servers
python3 lsp_benchmark.py \
--root ~/projects/some-python-repo \
Expand Down Expand Up @@ -144,6 +158,8 @@ python3 lsp_benchmark.py \
| `-r, --runs N` | Runs per package (default: 100 multi, 1 single) | both |
| `-s, --seed N` | RNG seed for reproducibility | both |
| `--timeout SECS` | Per-request timeout (default: 10s) | both |
| `--mode {definition,completion}` | Which LSP request to benchmark (default: definition) | both |
| `--prefix-len N` | Completion mode: characters of the member name left typed (default: 1, i.e. `pl.D`) | both |

### Output

Expand Down
Loading
Loading