## Summary On untouched `upstream/latest-alpha-cli` (release `0.116.0-alpha.3`), `codex-core` integration tests fail in my environment. ## Repro 1. Checkout upstream commit: - `0a89d20a364f5ca37f880aa4b38365b97ca66c9e` (`Release 0.116.0-alpha.3`) 2. Run: - `cd codex-rs && cargo test -p codex-core --test all` ## Observed `test result: FAILED. 604 passed; 168 failed; 16 ignored; 0 measured; 0 filtered out; finished in 107.78s` `error: test failed, to rerun pass '-p codex-core --test all'`

codex-core --test all fails on latest-alpha-cli (168 failures on 0.116.0-alpha.3)

Resolved 💬 3 comments Opened Mar 19, 2026 by rebroad Closed Mar 19, 2026

Summary

On untouched upstream/latest-alpha-cli (release 0.116.0-alpha.3), codex-core integration tests fail in my environment.

Repro

Checkout upstream commit:

0a89d20a364f5ca37f880aa4b38365b97ca66c9e (Release 0.116.0-alpha.3)

Run:

cd codex-rs && cargo test -p codex-core --test all

Observed

test result: FAILED. 604 passed; 168 failed; 16 ignored; 0 measured; 0 filtered out; finished in 107.78s

error: test failed, to rerun pass '-p codex-core --test all'

The failures include large clusters in suites such as:

suite::apply_patch_cli::*
suite::shell_serialization::*
suite::shell_snapshot::*
suite::tool_harness::*
suite::undo::*
suite::unified_exec::*
suite::view_image::*

Environment

OS: Linux hpenvyx360 6.18.12+deb13-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.18.12-1~bpo13+1 (2026-02-25) x86_64 GNU/Linux
Rust:
rustc 1.93.0 (254b59607 2026-01-19)
host: x86_64-unknown-linux-gnu
LLVM version: 21.1.8

Happy to provide fuller logs if helpful.

View original on GitHub ↗

3 Comments

rebroad · 5 months ago

Fresh data point from a clean repro on the current upstream alpha.

I checked out upstream/latest-alpha-cli, ran git clean -fdx, and then reran the same test command.

Upstream ref: upstream/latest-alpha-cli
Commit: b565c24a8fde7c6db27f49ad835f9fa83ad5cb5f
Version observed during build: codex-core v0.116.0-alpha.10
Command: cd codex-rs && cargo test -p codex-core --test all

Observed:

test result: FAILED. 60 passed; 723 failed; 16 ignored; 0 measured; 0 filtered out; finished in 12.78s

error: test failed, to rerun pass '-p codex-core --test all'

Failure clusters again include broad suites (agent jobs/websocket, apply_patch_cli, shell serialization/snapshot, undo, unified_exec, view_image, web_search, websocket_fallback, etc.), indicating this is not isolated to my previous branch state.

If helpful, I can attach or paste a full failing-test list from /tmp/cargo-test-core-all-upstream.

rebroad · 5 months ago

I added a docs clarification for this repro pitfall and pushed it to my fork on a dedicated branch:

Commit: https://github.com/openai/codex/commit/c5bc24b3a6aa1db9d76a7e21a0c6ed9de1cad5c0

The update adds a Linux-specific note in docs/install.md that cargo test -p codex-core --test all requires building codex-linux-sandbox first to avoid the large false-failure cascade from missing target/debug/codex-linux-sandbox.

rebroad · 5 months ago

Additional update with a concrete mitigation result:

I cherry-picked commit 25bf59fe68b70e56a58069c04e8156e80c9a18e8 (fix(core): route apply_patch self-invocation through codex binary) and reran:

cd codex-rs && cargo test -p codex-core --test all

Impact on my machine:

Before cherry-pick: 602 passed; 181 failed; 16 ignored
After cherry-pick: 715 passed; 68 failed; 16 ignored

So this single commit removes 113 failures in this environment and appears to resolve the large apply_patch/cascading bucket.

I still have 68 failures remaining (now concentrated in other areas such as js_repl/code_mode/view_image/rmcp_client-related suites).