mirror of
https://github.com/elicpeter/nyx.git
synced 2026-06-06 19:35:13 +02:00
576 lines
22 KiB
Rust
576 lines
22 KiB
Rust
//! Parity tests: same fixture, same mode, indexed vs non-indexed scan paths
|
|
//! MUST produce identical diagnostics.
|
|
//!
|
|
//! This invariant is release-critical. A scanner whose output depends on
|
|
//! whether indexing is enabled is not reliable for CI gates, suppressions,
|
|
//! or triage workflows. The tests here lock parity as a hard contract.
|
|
//!
|
|
//! ## What counts as "identical"
|
|
//!
|
|
//! We compare diagnostics as an unordered multiset of fingerprints:
|
|
//!
|
|
//! ```text
|
|
//! (path_relative_to_fixture, line, col, severity_str, rule_id)
|
|
//! ```
|
|
//!
|
|
//! Path-dependent fields (absolute path, rank_score derived from ordering,
|
|
//! evidence snippets that cite absolute paths) are excluded from the
|
|
//! fingerprint because they are not expected to diverge in meaning, only
|
|
//! in representation.
|
|
//!
|
|
//! If an engine change is justified in making indexed and non-indexed diverge,
|
|
//! the diff must be *explicit* in the test, not papered over by a loose
|
|
//! comparison. There are currently no such documented exceptions.
|
|
|
|
#[allow(dead_code)]
|
|
mod common;
|
|
|
|
use common::test_config;
|
|
use nyx_scanner::commands::index::build_index;
|
|
use nyx_scanner::commands::scan::{Diag, scan_with_index_parallel};
|
|
use nyx_scanner::database::index::Indexer;
|
|
use nyx_scanner::utils::config::AnalysisMode;
|
|
use std::collections::BTreeMap;
|
|
use std::path::{Path, PathBuf};
|
|
use std::sync::{Arc, OnceLock};
|
|
|
|
// Indexed scan helpers each open a fresh SQLite r2d2 pool (default max_size
|
|
// ~ ncpus+4) plus rayon-driven tree-sitter parsing across the fixture tree.
|
|
// Cargo runs every `#[test]` in this binary in parallel by default, so 30+
|
|
// indexed scans race to acquire file descriptors at once. Each pooled
|
|
// SQLite WAL connection costs ~3 fds (db + -wal + -shm); on sandboxes with
|
|
// a low per-process fd limit this exhausts EMFILE and surfaces as
|
|
// `Os { code: 24, … "Too many open files" }` panics from `build_index` /
|
|
// `scan_with_index_parallel`.
|
|
//
|
|
// Cap the pool to a small number of connections via `NYX_INDEX_POOL_MAX` so
|
|
// each parallel test holds far fewer fds. The cap is set once before any
|
|
// indexed scan runs, which keeps the suite embarrassingly parallel
|
|
// (previous workaround serialised every indexed scan via a process-wide
|
|
// mutex; that doubled wall-clock time on multi-core hosts).
|
|
fn ensure_index_pool_cap() {
|
|
static SET: OnceLock<()> = OnceLock::new();
|
|
SET.get_or_init(|| {
|
|
if std::env::var_os("NYX_INDEX_POOL_MAX").is_none() {
|
|
// SAFETY: We set the env var exactly once, inside `OnceLock`'s
|
|
// single-init barrier. No other thread in this test binary
|
|
// has called `Indexer::init` yet (the helpers call
|
|
// `ensure_index_pool_cap()` before any `init`), so no reader
|
|
// is concurrently observing this env value.
|
|
unsafe {
|
|
std::env::set_var("NYX_INDEX_POOL_MAX", "2");
|
|
}
|
|
}
|
|
});
|
|
}
|
|
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
// Fingerprint
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
|
|
/// Stable cross-path fingerprint for a single diagnostic.
|
|
///
|
|
/// Intentionally **does not** include:
|
|
/// - `path` (absolute): normalized to fixture-relative form instead.
|
|
/// - `rank_score` / `rank_reason`: derived from post-processing ordering.
|
|
/// - `evidence` snippets: contain absolute paths and formatting variations
|
|
/// that are representation-level, not analysis-level.
|
|
/// - `confidence`: derived deterministically from the fields we *do* compare;
|
|
/// if those match, confidence matches.
|
|
///
|
|
/// **Does** include:
|
|
/// - `(line, col)`, where the finding is reported.
|
|
/// - `severity`, the analyst-visible triage axis.
|
|
/// - `rule_id`, which detector fired.
|
|
/// - `path_validated`, semantic axis used by triage UIs.
|
|
///
|
|
/// If any of these differ between paths, the engine has genuinely produced
|
|
/// different *findings*, not just different metadata.
|
|
#[derive(Debug, Clone, PartialEq, Eq, PartialOrd, Ord)]
|
|
struct Fingerprint {
|
|
rel_path: String,
|
|
line: usize,
|
|
col: usize,
|
|
severity: &'static str,
|
|
rule_id: String,
|
|
path_validated: bool,
|
|
}
|
|
|
|
fn fingerprint(diag: &Diag, fixture_root: &Path) -> Fingerprint {
|
|
let abs = Path::new(&diag.path);
|
|
let rel = abs
|
|
.strip_prefix(fixture_root)
|
|
.map(|p| p.to_string_lossy().into_owned())
|
|
.unwrap_or_else(|_| diag.path.clone());
|
|
// Normalize Windows-style separators on UNIX for deterministic compare.
|
|
let rel = rel.replace('\\', "/");
|
|
Fingerprint {
|
|
rel_path: rel,
|
|
line: diag.line,
|
|
col: diag.col,
|
|
severity: diag.severity.as_db_str(),
|
|
rule_id: diag.id.clone(),
|
|
path_validated: diag.path_validated,
|
|
}
|
|
}
|
|
|
|
fn fingerprints(diags: &[Diag], fixture_root: &Path) -> Vec<Fingerprint> {
|
|
let mut v: Vec<Fingerprint> = diags.iter().map(|d| fingerprint(d, fixture_root)).collect();
|
|
v.sort();
|
|
v
|
|
}
|
|
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
// Scan helpers
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
|
|
fn scan_no_index(fixture_root: &Path, mode: AnalysisMode) -> Vec<Diag> {
|
|
let cfg = test_config(mode);
|
|
nyx_scanner::scan_no_index(fixture_root, &cfg).expect("scan_no_index should succeed")
|
|
}
|
|
|
|
/// Cold indexed scan: fresh DB, build index, then run indexed scan.
|
|
fn scan_indexed_cold(fixture_root: &Path, mode: AnalysisMode) -> (Vec<Diag>, PathBuf) {
|
|
ensure_index_pool_cap();
|
|
let cfg = test_config(mode);
|
|
let td = tempfile::tempdir().expect("tempdir");
|
|
let db_path = td.path().join("parity.sqlite");
|
|
|
|
build_index("parity", fixture_root, &db_path, &cfg, false).expect("build_index");
|
|
let pool = Indexer::init(&db_path).expect("init pool");
|
|
let diags = scan_with_index_parallel("parity", Arc::clone(&pool), &cfg, false, fixture_root)
|
|
.expect("indexed scan");
|
|
|
|
// Keep tempdir alive by returning the db_path; actually return ownership of td.
|
|
// We leak by forgetting the tempdir since the caller only needs the diags.
|
|
// (Leaving tempdir scope drops it; we want it cleaned up, so we *don't* forget.)
|
|
// The tempdir drops here and removes the file, diags are already owned.
|
|
std::mem::drop(td);
|
|
(diags, db_path)
|
|
}
|
|
|
|
/// Warm indexed scan: build index once, then run indexed scan **twice** on the
|
|
/// same pool. The second scan tests that cached artefacts don't perturb
|
|
/// output.
|
|
fn scan_indexed_warm(fixture_root: &Path, mode: AnalysisMode) -> Vec<Diag> {
|
|
ensure_index_pool_cap();
|
|
let cfg = test_config(mode);
|
|
let td = tempfile::tempdir().expect("tempdir");
|
|
let db_path = td.path().join("parity.sqlite");
|
|
|
|
build_index("parity", fixture_root, &db_path, &cfg, false).expect("build_index");
|
|
let pool = Indexer::init(&db_path).expect("init pool");
|
|
let _cold = scan_with_index_parallel("parity", Arc::clone(&pool), &cfg, false, fixture_root)
|
|
.expect("cold indexed scan");
|
|
let warm = scan_with_index_parallel("parity", Arc::clone(&pool), &cfg, false, fixture_root)
|
|
.expect("warm indexed scan");
|
|
std::mem::drop(td);
|
|
warm
|
|
}
|
|
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
// Diff reporting
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
|
|
fn format_fingerprint_set_diff(
|
|
label_a: &str,
|
|
a: &[Fingerprint],
|
|
label_b: &str,
|
|
b: &[Fingerprint],
|
|
) -> String {
|
|
// Count multiplicity of each fingerprint, divergence can be a changed
|
|
// *count* even when both sides contain the same key.
|
|
let mut count_a: BTreeMap<&Fingerprint, usize> = BTreeMap::new();
|
|
let mut count_b: BTreeMap<&Fingerprint, usize> = BTreeMap::new();
|
|
for fp in a {
|
|
*count_a.entry(fp).or_default() += 1;
|
|
}
|
|
for fp in b {
|
|
*count_b.entry(fp).or_default() += 1;
|
|
}
|
|
|
|
let all_keys: std::collections::BTreeSet<&Fingerprint> =
|
|
count_a.keys().chain(count_b.keys()).copied().collect();
|
|
|
|
let mut only_a = Vec::new();
|
|
let mut only_b = Vec::new();
|
|
let mut diff_counts = Vec::new();
|
|
for k in all_keys {
|
|
let ca = *count_a.get(k).unwrap_or(&0);
|
|
let cb = *count_b.get(k).unwrap_or(&0);
|
|
match (ca, cb) {
|
|
(a, 0) if a > 0 => only_a.push((k, a)),
|
|
(0, b) if b > 0 => only_b.push((k, b)),
|
|
(a, b) if a != b => diff_counts.push((k, a, b)),
|
|
_ => {}
|
|
}
|
|
}
|
|
|
|
let mut out = String::new();
|
|
out.push_str(&format!(
|
|
"\n=== Parity diff ({label_a} vs {label_b}) ===\n\
|
|
{label_a}: {} findings, {label_b}: {} findings\n",
|
|
a.len(),
|
|
b.len()
|
|
));
|
|
if !only_a.is_empty() {
|
|
out.push_str(&format!("\nOnly in {label_a} ({}):\n", only_a.len()));
|
|
for (fp, n) in only_a {
|
|
out.push_str(&format!(
|
|
" {} x{n} {}:{}:{} [{}] {}{}\n",
|
|
if n > 1 { "" } else { " " },
|
|
fp.rel_path,
|
|
fp.line,
|
|
fp.col,
|
|
fp.severity,
|
|
fp.rule_id,
|
|
if fp.path_validated {
|
|
" (validated)"
|
|
} else {
|
|
""
|
|
}
|
|
));
|
|
}
|
|
}
|
|
if !only_b.is_empty() {
|
|
out.push_str(&format!("\nOnly in {label_b} ({}):\n", only_b.len()));
|
|
for (fp, n) in only_b {
|
|
out.push_str(&format!(
|
|
" {} x{n} {}:{}:{} [{}] {}{}\n",
|
|
if n > 1 { "" } else { " " },
|
|
fp.rel_path,
|
|
fp.line,
|
|
fp.col,
|
|
fp.severity,
|
|
fp.rule_id,
|
|
if fp.path_validated {
|
|
" (validated)"
|
|
} else {
|
|
""
|
|
}
|
|
));
|
|
}
|
|
}
|
|
if !diff_counts.is_empty() {
|
|
out.push_str(&format!("\nCount mismatch ({}):\n", diff_counts.len()));
|
|
for (fp, na, nb) in diff_counts {
|
|
out.push_str(&format!(
|
|
" {label_a}={na} {label_b}={nb} {}:{}:{} [{}] {}\n",
|
|
fp.rel_path, fp.line, fp.col, fp.severity, fp.rule_id
|
|
));
|
|
}
|
|
}
|
|
out
|
|
}
|
|
|
|
fn assert_parity(
|
|
label_a: &str,
|
|
a: &[Fingerprint],
|
|
label_b: &str,
|
|
b: &[Fingerprint],
|
|
fixture_name: &str,
|
|
) {
|
|
if a == b {
|
|
return;
|
|
}
|
|
panic!(
|
|
"[{fixture_name}] Parity violation between {label_a} and {label_b}:{}",
|
|
format_fingerprint_set_diff(label_a, a, label_b, b)
|
|
);
|
|
}
|
|
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
// Parity test driver
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
|
|
fn fixture_path(name: &str) -> PathBuf {
|
|
PathBuf::from(env!("CARGO_MANIFEST_DIR"))
|
|
.join("tests")
|
|
.join("fixtures")
|
|
.join(name)
|
|
}
|
|
|
|
fn run_parity(fixture_name: &str, mode: AnalysisMode) {
|
|
let dir = fixture_path(fixture_name);
|
|
let no_index_diags = scan_no_index(&dir, mode);
|
|
let (indexed_diags, _db) = scan_indexed_cold(&dir, mode);
|
|
|
|
let a = fingerprints(&no_index_diags, &dir);
|
|
let b = fingerprints(&indexed_diags, &dir);
|
|
assert_parity("no-index", &a, "indexed-cold", &b, fixture_name);
|
|
}
|
|
|
|
fn run_parity_warm(fixture_name: &str, mode: AnalysisMode) {
|
|
let dir = fixture_path(fixture_name);
|
|
let no_index_diags = scan_no_index(&dir, mode);
|
|
let warm_diags = scan_indexed_warm(&dir, mode);
|
|
|
|
let a = fingerprints(&no_index_diags, &dir);
|
|
let b = fingerprints(&warm_diags, &dir);
|
|
assert_parity("no-index", &a, "indexed-warm", &b, fixture_name);
|
|
}
|
|
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
// Fixtures under parity contract, Full mode
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
//
|
|
// Representative mix covering all 10 supported languages plus cross-file
|
|
// resolution, framework-specific rules, and auth analysis. Every fixture
|
|
// listed here is a hard parity invariant: a regression must either be fixed
|
|
// or explicitly documented (see bottom of this file).
|
|
|
|
const FULL_MODE_PARITY_FIXTURES: &[&str] = &[
|
|
// Cross-file taint resolution across languages
|
|
"cross_file_js_sqli",
|
|
"cross_file_py_const_passthrough",
|
|
"cross_file_go_handler_exec",
|
|
"cross_file_java_sqli",
|
|
"cross_file_taint",
|
|
"cross_file_ssa_propagation",
|
|
"cross_file_ssa_sanitizer",
|
|
"cross_file_scc_convergence",
|
|
// Single-language cross-file + framework
|
|
"rust_web_app",
|
|
"rust_framework_rules",
|
|
"rust_module_path_resolution",
|
|
"express_app",
|
|
"flask_app",
|
|
"go_server",
|
|
"java_service",
|
|
// Auth analysis
|
|
"auth_analysis_integration",
|
|
"auth_analysis_frameworks_integration",
|
|
// AST / pattern heavy
|
|
"patterns",
|
|
// Termination + state
|
|
"taint_termination",
|
|
"state",
|
|
// Noise-reduction / suppression
|
|
"route_registration_noise",
|
|
];
|
|
|
|
#[test]
|
|
fn parity_full_cross_file_js_sqli() {
|
|
run_parity("cross_file_js_sqli", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_cross_file_py_const_passthrough() {
|
|
run_parity("cross_file_py_const_passthrough", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_cross_file_go_handler_exec() {
|
|
run_parity("cross_file_go_handler_exec", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_cross_file_java_sqli() {
|
|
run_parity("cross_file_java_sqli", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_cross_file_taint() {
|
|
run_parity("cross_file_taint", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_cross_file_ssa_propagation() {
|
|
run_parity("cross_file_ssa_propagation", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_cross_file_ssa_sanitizer() {
|
|
run_parity("cross_file_ssa_sanitizer", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_cross_file_scc_convergence() {
|
|
run_parity("cross_file_scc_convergence", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_rust_web_app() {
|
|
run_parity("rust_web_app", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_rust_framework_rules() {
|
|
run_parity("rust_framework_rules", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_rust_module_path_resolution() {
|
|
run_parity("rust_module_path_resolution", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_express_app() {
|
|
run_parity("express_app", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_flask_app() {
|
|
run_parity("flask_app", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_go_server() {
|
|
run_parity("go_server", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_java_service() {
|
|
run_parity("java_service", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_auth_analysis_integration() {
|
|
run_parity("auth_analysis_integration", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_auth_analysis_frameworks_integration() {
|
|
run_parity("auth_analysis_frameworks_integration", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_patterns() {
|
|
run_parity("patterns", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_taint_termination() {
|
|
run_parity("taint_termination", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_state() {
|
|
run_parity("state", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_full_route_registration_noise() {
|
|
run_parity("route_registration_noise", AnalysisMode::Full);
|
|
}
|
|
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
// Non-Full analysis modes, the Taint-mode filter divergence lives here
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
//
|
|
// Taint mode is the narrowest CFG-capable mode. Historically the indexed
|
|
// path filtered output to `taint*`/`cfg-*` rule ids while the non-indexed
|
|
// path did not, silently dropping state-model and auth-analysis findings
|
|
// from indexed scans. This test pins the fix.
|
|
|
|
#[test]
|
|
fn parity_taint_cross_file_js_sqli() {
|
|
run_parity("cross_file_js_sqli", AnalysisMode::Taint);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_taint_cross_file_py_const_passthrough() {
|
|
run_parity("cross_file_py_const_passthrough", AnalysisMode::Taint);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_taint_auth_analysis_integration() {
|
|
// This fixture exercises auth_analysis rules, which were previously
|
|
// dropped by the indexed Taint-mode filter.
|
|
run_parity("auth_analysis_integration", AnalysisMode::Taint);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_cfg_mode_cross_file_js_sqli() {
|
|
run_parity("cross_file_js_sqli", AnalysisMode::Cfg);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_ast_mode_patterns() {
|
|
run_parity("patterns", AnalysisMode::Ast);
|
|
}
|
|
|
|
/// The `state/` fixture is dense with state-model findings (`rs.resource.*`,
|
|
/// `auth.*`). These are produced by `run_cfg_analyses` under *any* CFG-
|
|
/// capable mode, including Taint-only. A historical filter in the indexed
|
|
/// path dropped everything that wasn't `taint*`/`cfg-*` from Taint-mode
|
|
/// output, silently swallowing state findings, this test pins that fix.
|
|
#[test]
|
|
fn parity_taint_state_fixture() {
|
|
run_parity("state", AnalysisMode::Taint);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_cfg_state_fixture() {
|
|
run_parity("state", AnalysisMode::Cfg);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_ast_state_fixture() {
|
|
run_parity("state", AnalysisMode::Ast);
|
|
}
|
|
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
// Warm-scan parity, detects caching bugs in the indexed path
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
|
|
#[test]
|
|
fn parity_warm_cross_file_js_sqli() {
|
|
run_parity_warm("cross_file_js_sqli", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_warm_auth_analysis_integration() {
|
|
run_parity_warm("auth_analysis_integration", AnalysisMode::Full);
|
|
}
|
|
|
|
#[test]
|
|
fn parity_warm_patterns_ast_mode() {
|
|
run_parity_warm("patterns", AnalysisMode::Ast);
|
|
}
|
|
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
// Sweep: every fixture in FULL_MODE_PARITY_FIXTURES must pass Full-mode parity
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
//
|
|
// The explicit per-fixture tests above give quick signal on what broke; this
|
|
// sweep locks the list itself so adding a fixture to the "release-critical"
|
|
// set is a deliberate choice. Run with: `cargo test parity_full_sweep -- --nocapture`.
|
|
|
|
#[test]
|
|
fn parity_full_sweep_all_fixtures() {
|
|
let mut failures: Vec<(String, String)> = Vec::new();
|
|
for name in FULL_MODE_PARITY_FIXTURES {
|
|
let dir = fixture_path(name);
|
|
let a = fingerprints(&scan_no_index(&dir, AnalysisMode::Full), &dir);
|
|
let (indexed, _db) = scan_indexed_cold(&dir, AnalysisMode::Full);
|
|
let b = fingerprints(&indexed, &dir);
|
|
if a != b {
|
|
failures.push((
|
|
(*name).to_string(),
|
|
format_fingerprint_set_diff("no-index", &a, "indexed-cold", &b),
|
|
));
|
|
}
|
|
}
|
|
|
|
if !failures.is_empty() {
|
|
let mut msg = format!(
|
|
"parity sweep failed for {} / {} fixtures:\n",
|
|
failures.len(),
|
|
FULL_MODE_PARITY_FIXTURES.len()
|
|
);
|
|
for (fixture, diff) in &failures {
|
|
msg.push_str(&format!("\n── {fixture} ──{diff}\n"));
|
|
}
|
|
panic!("{msg}");
|
|
}
|
|
}
|
|
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
// Principled divergences (documented exceptions)
|
|
// ─────────────────────────────────────────────────────────────────────────────
|
|
//
|
|
// None. Release-critical modes (Full, Taint, Cfg, Ast) must match bit-for-bit
|
|
// on the finding fingerprint. If you think you need to add an exception,
|
|
// the test above should be the primary gate, don't loosen parity without
|
|
// writing a test that demonstrates *why* the divergence is acceptable.
|