Code Review Process

Code review during migration has unique concerns beyond standard Rust review. Reviewers must verify semantic preservation, check for unsafe code correctness, and validate performance characteristics of transpiled code.

Review Checklist

General (All Code)

Code compiles with zero warnings (cargo clippy -- -D warnings)
Tests pass and cover the new code (>= 95%)
No unnecessary unwrap() or expect() in production code
Error types are meaningful and actionable
Documentation exists for public API

Migration-Specific

Transpiled output matches original behavior (parity tests present)
No semantic drift from the source language
Dependencies mapped correctly (e.g., numpy operations use trueno)
Performance benchmarks show no regression vs original

Unsafe Code Policy

Unsafe code requires elevated review. Any PR containing unsafe must:

Document why safe alternatives are insufficient
Include a // SAFETY: comment explaining the invariants
Be reviewed by at least two team members
Have dedicated tests exercising the unsafe boundary

#![allow(unused)]
fn main() {
// SAFETY: `data` is guaranteed to be aligned to 32 bytes by the allocator,
// and `len` is bounds-checked by the caller. The pointer is valid for the
// lifetime of the slice.
unsafe {
    std::arch::x86_64::_mm256_load_ps(data.as_ptr())
}
}

Performance Review

For code on the hot path, verify:

Check	How to Verify
No accidental allocations in loops	Run DHAT or review for `Vec::new()`, `format!()`, `to_string()`
SIMD where applicable	Check trueno usage for data-parallel operations
Correct backend selection	Verify the 5x PCIe rule for GPU paths
Buffer reuse	Look for `clear()` + reuse patterns instead of `new()`

Using PMAT in Review

Reviewers can use pmat to quickly assess code quality:

# Check complexity of changed functions
pmat analyze complexity ./src/changed_module.rs

# Find fault patterns (unwrap, panic, unsafe)
pmat query "changed_function" --faults --include-source

Review Workflow

Author runs make tier2 before submitting (pre-commit checks)
CI runs make tier4 automatically on the PR
Reviewer checks pmat analysis and CI results
Reviewer verifies parity tests exist for migrated code
Two approvals required for unsafe code, one for safe code
Merge only after quality gate passes (batuta stack gate)

Common Review Feedback

Issue	Feedback Template
Missing error context	“Add `.context()` with a descriptive message”
Bare unwrap	“Replace with `?` or handle the error explicitly”
Missing parity test	“Add a test comparing output to the Python original”
Allocation in hot loop	“Consider pre-allocating this buffer outside the loop”
Undocumented unsafe	“Add a `// SAFETY:` comment explaining the invariants”

Navigate: Table of Contents

Keyboard shortcuts

The Batuta Book