valknar/llmx
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Auto compact at ~90% (#5292)

Browse Source 
Users now hit a window exceeded limit and they usually don't know what
to do. This starts auto compact at ~90% of the window.
This commit is contained in:
Ahmed Ibrahim
2025-10-20 11:29:49 -07:00
committed by GitHub
parent cda6db6ccf
commit 049a61bcfc
21 changed files with 236 additions and 110 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
codex-rs/core/tests/suite/client.rs
View File

@@ -858,8 +858,8 @@ async fn token_count_includes_rate_limits_snapshot() {
"reasoning_output_tokens": 0,
"total_tokens": 123
},
// Default model is gpt-5-codex in tests → 272000 context window
"model_context_window": 272000
// Default model is gpt-5-codex in tests → 95% usable context window
"model_context_window": 258400
},
"rate_limits": {
"primary": {
@@ -985,6 +985,8 @@ async fn context_window_error_sets_total_tokens_to_model_window() -> anyhow::Res
skip_if_no_network!(Ok(()));
let server = MockServer::start().await;
const EFFECTIVE_CONTEXT_WINDOW: i64 = (272_000 * 95) / 100;
responses::mount_sse_once_match(
&server,
body_string_contains("trigger context window"),
@@ -1056,8 +1058,11 @@ async fn context_window_error_sets_total_tokens_to_model_window() -> anyhow::Res
.info
.expect("token usage info present when context window is exceeded");
assert_eq!(info.model_context_window, Some(272_000));
assert_eq!(info.total_token_usage.total_tokens, 272_000);
assert_eq!(info.model_context_window, Some(EFFECTIVE_CONTEXT_WINDOW));
assert_eq!(
info.total_token_usage.total_tokens,
EFFECTIVE_CONTEXT_WINDOW
);
let error_event = wait_for_event(&codex, |ev| matches!(ev, EventMsg::Error(_))).await;
let expected_context_window_message = CodexErr::ContextWindowExceeded.to_string();