Skip to content

The fastest, cheapest, most efficient open-source agent harness. Run more than 500 agents on a $10 VPS.

test(native-channel): cover image and file captions separately

Steven Enamakelcf180cc1ff19
diff --git a/tests/native_channel_session_restart_e2e.rs b/tests/native_channel_session_restart_e2e.rs
index f0a37b43ea..639abdde9d 100644
--- a/tests/native_channel_session_restart_e2e.rs
+++ b/tests/native_channel_session_restart_e2e.rs
@@ -147,8 +147,8 @@ async fn get_file(Json(body): Json<Value>) -> Json<Value> {
async fn download_file(path: axum::extract::Path<String>) -> Response {
let (content_type, bytes) = match path.0.as_str() {
- "photos/native.png" => ("application/octet-stream", NATIVE_IMAGE_PNG),
- "documents/native.txt" => ("application/octet-stream", NATIVE_FILE_BYTES),
+ "photos/native.png" => ("image/png", NATIVE_IMAGE_PNG),
+ "documents/native.txt" => ("text/plain", NATIVE_FILE_BYTES),
_ => return StatusCode::NOT_FOUND.into_response(),
};
(
@@ -248,6 +248,10 @@ fn media_update(
document: bool,
text: Option<&str>,
) -> Value {
+ assert!(
+ !(image && document),
+ "a Telegram message has one media object"
+ );
let mut message = json!({
"message_id": message_id,
"date": 1_800_000_000,
@@ -271,7 +275,11 @@ fn media_update(
});
}
if let Some(text) = text {
- message["text"] = json!(text);
+ if image || document {
+ message["caption"] = json!(text);
+ } else {
+ message["text"] = json!(text);
+ }
}
json!({ "update_id": update_id, "message": message })
}
@@ -766,9 +774,14 @@ async fn native_telegram_typed_media_reaches_the_model_without_provider_paths()
wait_for_completion(&mocks.state, 2).await;
push_media_update(
&mocks.state,
- media_update(202, 22, true, true, Some("before typed media after")),
+ media_update(202, 22, true, false, Some("image caption before media")),
);
wait_for_completion(&mocks.state, 3).await;
+ push_media_update(
+ &mocks.state,
+ media_update(203, 23, false, true, Some("file caption before media")),
+ );
+ wait_for_completion(&mocks.state, 4).await;
let requests = mocks
.state
@@ -776,24 +789,36 @@ async fn native_telegram_typed_media_reaches_the_model_without_provider_paths()
.lock()
.expect("model requests")
.clone();
+ assert_eq!(
+ requests.len(),
+ 4,
+ "each media update reaches the model once"
+ );
assert!(model_saw_media_bytes(&requests[0], NATIVE_IMAGE_PNG));
assert!(model_saw_media_bytes(&requests[1], NATIVE_FILE_BYTES));
- let mixed_content = last_user_content(&requests[2]).expect("mixed user message is present");
- let mixed = serde_json::to_string(mixed_content).expect("serialize mixed user content");
- let before = mixed
- .find("before typed media")
- .expect("caption text is present");
- let image = mixed
+ let image_mixed = serde_json::to_string(
+ last_user_content(&requests[2]).expect("mixed image user message is present"),
+ )
+ .expect("serialize mixed image user content");
+ let image_caption = image_mixed
+ .find("image caption before media")
+ .expect("image caption text is present");
+ let image = image_mixed
.find(&base64::engine::general_purpose::STANDARD.encode(NATIVE_IMAGE_PNG))
.expect("mixed image bytes reach the model");
- let file = mixed
+ assert!(image_caption < image, "image part order: {image_mixed}");
+
+ let file_mixed = serde_json::to_string(
+ last_user_content(&requests[3]).expect("mixed file user message is present"),
+ )
+ .expect("serialize mixed file user content");
+ let file_caption = file_mixed
+ .find("file caption before media")
+ .expect("file caption text is present");
+ let file = file_mixed
.find(&base64::engine::general_purpose::STANDARD.encode(NATIVE_FILE_BYTES))
.expect("mixed file bytes reach the model");
- let after = mixed.find("after").expect("trailing text is present");
- assert!(
- before < image && image < file && file < after,
- "typed part order: {mixed}"
- );
+ assert!(file_caption < file, "file part order: {file_mixed}");
for request in &requests {
let serialized = serde_json::to_string(request).unwrap();
assert!(!serialized.contains("typed-image-file"));