UPSTREAM PR #2471: feat: add WASI (wasm32-wasip2) platform support

[loci-dev]

Note

Source pull request: GitoxideLabs/gitoxide#2471

Summary

Add compilation support for the wasm32-wasip2 target across four crates, enabling gitoxide to compile under WASI sandboxed environments such as Wasmtime.

Changes

• gix-path: Enable #![feature(wasip2)] via cfg_attr for WASI builds, required for std::os::wasi path handling
• gix-fs/symlink: Split #[cfg(not(windows))] into separate #[cfg(unix)] and #[cfg(target_os = "wasi")] blocks — WASI uses std::fs::soft_link since std::os::unix::fs::symlink is unavailable
• gix-fs/capabilities: Add a Default impl for Capabilities on WASI (no executable bit, symlinks enabled, no case-folding or unicode precomposition)
• gix-index/fs: Replace #[cfg(not(windows))] / #[cfg(windows)] with #[cfg(unix)] / #[cfg(not(unix))], grouping WASI with Windows for metadata access via std::fs::Metadata instead of rustix::fs::Stat

Motivation

We're compiling the GitButler CLI (but) to wasm32-wasip2 to run git operations inside sandboxed WASI agents. These four changes were the minimum needed to get gitoxide compiling for this target.

Notes

• The #![feature(wasip2)] in gix-path requires nightly or RUSTC_BOOTSTRAP=1 on stable — this is a temporary requirement until std::os::wasi stabilizes
• rustix doesn't support WASI, so gix-index falls back to std::fs::Metadata (same as Windows)
• No behavioral changes for existing unix/windows/macos targets — all existing #[cfg] paths are preserved

Test plan

• Verified cargo build --target wasm32-wasip2 succeeds for a downstream consumer (GitButler but CLI)
• Verified no regressions on native Linux build
• CI should confirm no regressions on Windows/macOS

[loci-review]

Overview

Analysis of WASI platform support addition (commit 43bf94d) across two binaries shows negligible performance impact. Total functions: 34,636 (6,159 modified, 8,250 new, 8,282 removed).

Power consumption:

• target.aarch64-unknown-linux-gnu.release.gix: 804,258.03 nJ → 803,778.78 nJ (-0.06%)
• target.aarch64-unknown-linux-gnu.release.ein: 299,406.49 nJ → 301,228.08 nJ (+0.61%)

Function Analysis

Standard library I/O improvements (performance-critical):

• std::io::Read::read_buf_exact (gix): Response time -1.72% (-163ns), used extensively in pack file reading
• std::io::BufRead::skip_until (gix): Response time -1.97% (-62ns), benefits protocol handling
• gix_negotiate::consecutive::Algorithm::add_tip (gix): Response time -3.52% (-223ns), improves fetch negotiation

Regressions in non-critical paths:

• human_format::to_scaled_value (ein): Response time +1,125,522% (+68,312ns) — external dependency used only for CLI progress formatting
• std::io::Read::read_exact (gix): Response time +22.06% (+548ns) — added protocol-aware functionality (header handling, content-type validation)
• bstr::BString::fmt (gix): Response time +40.99% (+983ns) — external dependency for debug formatting
• hyper::client::conn::http1::upgrades::UpgradeableConnection::poll (gix): Response time +2.50% (+64µs), throughput time +3,897% — enhanced state machine in HTTP transport, masked by network latency

Compiler-generated changes: Multiple OUTLINED_FUNCTION_* instances show large percentage increases (up to 506,496%) but minimal absolute impact (<70µs), affecting only cold paths (error handling, cleanup, initialization).

Source code context: Single commit added WASI platform support to gix-fs, gix-path, and gix-config-value with compile-time conditionals. Changes don't affect aarch64-linux target directly. Observed performance differences stem from compiler optimizations, standard library updates, and external dependency changes rather than application code modifications.

Additional Findings

Performance-critical Git operations (pack handling, object database access, diff algorithms) remain unaffected. Core I/O primitives show improvements through better lock acquisition (-2.6% in lock_slow) and optimized read operations. The WASI platform support successfully adds cross-platform compatibility without compromising performance on existing targets.

🔎 Full breakdown: Loci Inspector
💬 Questions? Tag @loci-dev