xff - Test Plan

Companion to implementation-plan.md. Project rule: tests at every level; no one-shots - any manual verification must be codified here or as a test before a PR is reviewable.
Status: Draft · 2026-06-07

Levels

1. parser - pure unit tests (no fixtures)

• Table-driven argv[] → AST (globals, roots, expression tree) and argv[] → diagnostic (message + source span).
• Cover: cluster parsing (-Hg+x), suffix-sign toggles (bare/+/-), --long/--no-X, -- boundary, paths beginning with -/+, GNU vs BSD globals, -exec … \; / … + greedy consumption, operator precedence (! > -a > -o, ,), arity errors.

2. find-compatibility conformance (Phase-1 gate)

• A fixture tree (varied names incl. dotfiles, sizes, mtimes, perms, symlinks, nested dirs).
• An expression matrix run through both real find and xff; assert identical output + exit code.
• Run on both dialects: Linux GNU-find and macOS BSD-find.

3. engine - integration

• Predicate/action correctness over fixture trees; -exec/-delete/-prune effects; -xdev, depth bounds.
• Parallel-traversal determinism under --sort; short-circuit evaluation order observable via side-effecting actions.

4. regex

• Per-engine conformance (RE2 vs PCRE2); literal-prefilter must not change match results (equivalence test).

5. render - golden tests

• One golden per format (plain/NUL/JSONL/CSV/aligned/markdown/tree).
• Unicode-width alignment (CJK/emoji/combining marks line up); non-UTF-8 path handling in JSON (escape policy).

6. config

• Cascade precedence (CLI > local dir > user > default); provenance (unset ≠ off); promotion-to-modern defaults.

CI

• bazel test //... on macOS + Linux for every PR; buildifier lint.