From baaf87aa239dcbba3bb27aef103c2ca0d392a9d4 Mon Sep 17 00:00:00 2001 From: Bennet Fenner Date: Tue, 28 Oct 2025 15:30:47 +0100 Subject: [PATCH] Fix `unit_evals.yml` (#41377) Release Notes: - N/A --- .github/workflows/unit_evals.yml | 2 +- crates/agent/Cargo.toml | 2 +- crates/agent/src/edit_agent/evals.rs | 18 +++++++++--------- 3 files changed, 11 insertions(+), 11 deletions(-) diff --git a/.github/workflows/unit_evals.yml b/.github/workflows/unit_evals.yml index c03cf8b087188f3e10a298e52a8278e63765c4f0..53ed33a1af300d6b641b3b9430de0bb6846b27cc 100644 --- a/.github/workflows/unit_evals.yml +++ b/.github/workflows/unit_evals.yml @@ -63,7 +63,7 @@ jobs: - name: Run unit evals shell: bash -euxo pipefail {0} - run: cargo nextest run --workspace --no-fail-fast --features eval --no-capture -E 'test(::eval_)' + run: cargo nextest run --workspace --no-fail-fast --features unit-eval --no-capture -E 'test(::eval_)' env: ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }} diff --git a/crates/agent/Cargo.toml b/crates/agent/Cargo.toml index e0f2d9dcb97e298dd3c906e3f902974821efcdc0..e962c876a38f788607706aad4e53ee5e0488b08d 100644 --- a/crates/agent/Cargo.toml +++ b/crates/agent/Cargo.toml @@ -11,7 +11,7 @@ path = "src/agent.rs" [features] test-support = ["db/test-support"] eval = [] -edit-agent-eval = [] +unit-eval = [] e2e = [] [lints] diff --git a/crates/agent/src/edit_agent/evals.rs b/crates/agent/src/edit_agent/evals.rs index 48977df1974cc104bc10fdf8975ed09172a1a938..84cdd101f57546a0bfbc86a290bf1f453e69a979 100644 --- a/crates/agent/src/edit_agent/evals.rs +++ b/crates/agent/src/edit_agent/evals.rs @@ -31,7 +31,7 @@ use std::{ use util::path; #[test] -#[cfg_attr(not(feature = "edit-agent-eval"), ignore)] +#[cfg_attr(not(feature = "unit-eval"), ignore)] fn eval_extract_handle_command_output() { // Test how well agent generates multiple edit hunks. // @@ -108,7 +108,7 @@ fn eval_extract_handle_command_output() { } #[test] -#[cfg_attr(not(feature = "edit-agent-eval"), ignore)] +#[cfg_attr(not(feature = "unit-eval"), ignore)] fn eval_delete_run_git_blame() { // Model | Pass rate // ----------------------------|---------- @@ -171,7 +171,7 @@ fn eval_delete_run_git_blame() { } #[test] -#[cfg_attr(not(feature = "edit-agent-eval"), ignore)] +#[cfg_attr(not(feature = "unit-eval"), ignore)] fn eval_translate_doc_comments() { // Model | Pass rate // ============================================ @@ -234,7 +234,7 @@ fn eval_translate_doc_comments() { } #[test] -#[cfg_attr(not(feature = "edit-agent-eval"), ignore)] +#[cfg_attr(not(feature = "unit-eval"), ignore)] fn eval_use_wasi_sdk_in_compile_parser_to_wasm() { // Model | Pass rate // ============================================ @@ -360,7 +360,7 @@ fn eval_use_wasi_sdk_in_compile_parser_to_wasm() { } #[test] -#[cfg_attr(not(feature = "edit-agent-eval"), ignore)] +#[cfg_attr(not(feature = "unit-eval"), ignore)] fn eval_disable_cursor_blinking() { // Model | Pass rate // ============================================ @@ -446,7 +446,7 @@ fn eval_disable_cursor_blinking() { } #[test] -#[cfg_attr(not(feature = "edit-agent-eval"), ignore)] +#[cfg_attr(not(feature = "unit-eval"), ignore)] fn eval_from_pixels_constructor() { // Results for 2025-06-13 // @@ -656,7 +656,7 @@ fn eval_from_pixels_constructor() { } #[test] -#[cfg_attr(not(feature = "edit-agent-eval"), ignore)] +#[cfg_attr(not(feature = "unit-eval"), ignore)] fn eval_zode() { // Model | Pass rate // ============================================ @@ -763,7 +763,7 @@ fn eval_zode() { } #[test] -#[cfg_attr(not(feature = "edit-agent-eval"), ignore)] +#[cfg_attr(not(feature = "unit-eval"), ignore)] fn eval_add_overwrite_test() { // Model | Pass rate // ============================================ @@ -995,7 +995,7 @@ fn eval_add_overwrite_test() { } #[test] -#[cfg_attr(not(feature = "edit-agent-eval"), ignore)] +#[cfg_attr(not(feature = "unit-eval"), ignore)] fn eval_create_empty_file() { // Check that Edit Agent can create a file without writing its // thoughts into it. This issue is not specific to empty files, but