use std::time::{Duration, Instant};
use second_brain_core::embedding::{EmbedTimeoutError, Embedder, query_prompt};
#[test]
fn query_prompt_prefixes_with_bge_retrieval_instruction() {
assert_eq!(
query_prompt("foo"),
"Represent this sentence for searching relevant passages: foo"
);
}
#[test]
fn embed_query_matches_prefixed_embed_and_differs_from_bare() {
let embedder = Embedder::new().unwrap();
let prefixed = embedder
.embed("Represent this sentence for searching relevant passages: foo")
.unwrap();
let queried = embedder.embed_query("foo").unwrap();
let bare = embedder.embed("foo").unwrap();
assert_eq!(
queried, prefixed,
"embed_query must apply the BGE query prefix"
);
assert_ne!(queried, bare, "the prefix must change the embedding");
}
#[test]
fn a_query_with_a_deadline_never_waits_for_a_model_reload() {
let embedder = Embedder::new().unwrap();
let reference = embedder.embed_query("foo").unwrap();
embedder.drop_if_idle(0);
let started = Instant::now();
let err = embedder
.embed_query_within("foo", Duration::from_millis(20))
.unwrap_err();
assert!(err.downcast_ref::<EmbedTimeoutError>().is_some(), "{err:#}");
assert!(
started.elapsed() < Duration::from_secs(1),
"{:?}",
started.elapsed()
);
let warmed = embedder
.embed_query_within("foo", Duration::from_secs(30))
.unwrap();
assert_eq!(warmed.len(), reference.len());
for (w, r) in warmed.iter().zip(&reference) {
assert!((w - r).abs() < 1e-5, "{w} vs {r}");
}
}