test(e2e): allow external step programs#560
Conversation
Co-authored-by: GPT-5 Codex <codex@openai.com>
d2161c1 to
c1af19f
Compare
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: c1af19f3cc
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| env::var_os(exe_env.as_str()).unwrap_or_else(|| panic!("{exe_env} not set")); | ||
| CommandBuilder::new(resolved) | ||
| } else { | ||
| CommandBuilder::new(program) |
There was a problem hiding this comment.
On Windows, external tools provisioned under packages/tools/node_modules/.bin (such as vite) are pnpm-generated .cmd/.ps1 shims rather than native executables. portable_pty::CommandBuilder launches the program directly, so CommandBuilder::new("vite") cannot execute those shims; setting PATHEXT later only configures the child environment and does not route the initial launch through a command interpreter. Consequently, ignored external-tool E2Es still fail to spawn on Windows unless the harness resolves and invokes the appropriate shim explicitly.
AGENTS.md reference: AGENTS.md:L147-L152
Useful? React with 👍 / 👎.
1 participant
Motivation
The E2E snapshot harness resolves
vtandvttthrough Cargo-provided binary paths so the default suite remains self-contained and portable. That hard restriction also prevents ignored fixtures from opting into external tools that the repository explicitly provisions inpackages/tools.What changed
CARGO_BIN_EXE_*resolution forvtandvtt.PATH.The full local
just readygate passes on this branch.