fix: run Git for Windows sh shim in POSIX mode

[Sebastian Thiel (Byron)]

Tasks

• refackiew

Everything below this line is Codex

---

What changed

• retain whether gix_path::env::shell() selected Git for Windows' bin/sh.exe shim
• add gix_path::env::shell_command() to configure that shim with --posix
• use the configured default shell command from gix-command while leaving with_shell_program() unchanged
• add a Windows regression test that verifies Bash has POSIX mode enabled

Why

Git for Windows' bin/sh.exe shim delegates to bash.exe. Unlike the non-shim
usr/bin/sh.exe, that starts Bash without POSIX mode, so shell commands run by
gix-command behave differently from shell commands run by Git.

The POSIX option is applied only when the known Git for Windows shim was selected.
Fallback shells and caller-provided shell programs do not receive the Bash-specific
option.

Fixes #1868.

Validation

• cargo test -p gix-command
• cargo test -p gix-path -- --test-threads=1
• cargo clippy -p gix-path -p gix-command --lib -- -D warnings
• cargo fmt --all -- --check
• cargo doc -p gix-path --no-deps

[Eliah Kagan (EliahKagan)]

I'm trying to remember why we didn't just go with passing --posix before (under the relevant conditions). I think there was some reason discussed for not moving forward with it, but I don't know if that reason still applies, and maybe it was just that I didn't have time to make the change at that time. If nothing about that can be found on the issue and its comments then this approach is probably okay. I can look at this in a couple days, but I don't think there's any need to wait if the comment discussion has been examined.

[Sebastian Thiel (Byron)]

Thanks for sharing! While somehow I lost the context of the agent that performed this work, I just asked again to see if the comments in the issue materially affect this implementation: short answer is No.

Codex

No material correctness issue found. The issue comments support the branch’s approach:

• It keeps the Git-for-Windows shim and its environment setup, as recommended in the discussion.
• It adds --posix only to the detected shim and leaves custom shells untouched: gix-path/src/env/mod.rs, gix-command/src/lib.rs.
• It avoids parent-process environment mutation, matching the stated library-safety concern.

One non-material residual edge remains: fallback sh.exe instances that cannot be identified as the Git-for-Windows shim receive no --posix. Applying it indiscriminately would break non-Bash shells, so the branch’s conservative behavior is correct.

Relevant crate tests, formatting, and git diff --check pass. The Windows-only regression test exists but could not run on this macOS host. No changes made.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6f27275e3a

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 (@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 (@codex) address that feedback".