adler_core/

registry.rs

//! Site registry — loading, validation, filtering.
//!
//! The default registry is embedded into the binary at compile time via
//! [`include_str!`]. Callers can override it with a file at runtime through
//! [`Registry::load_from_path`].

use std::collections::{BTreeMap, HashMap, HashSet};
use std::path::Path;

use serde::Deserialize;

use crate::error::{Error, Result};
use crate::site::{Engine, Site};

const EMBEDDED_REGISTRY: &str = include_str!("../data/sites.json");

/// Supplementary registry derived from the `WhatsMyName` project
/// (`WebBreacher/WhatsMyName`, CC BY-SA 4.0). Kept as a separate
/// constant because its data license is incompatible with the
/// MIT-only [`EMBEDDED_REGISTRY`] above; callers opt in explicitly
/// via [`Registry::default_embedded_with_wmn`] to keep the default
/// MIT-clean for downstream redistribution.
const EMBEDDED_WMN_REGISTRY: &str = include_str!("../data/sites_wmn.json");

/// A loaded, validated collection of site definitions.
///
/// Engines (shared signature templates referenced by [`Site::engine`])
/// are resolved into sites at load time — by the time you call
/// [`Registry::sites`] every entry already has its inherited
/// `signals` / `request_headers` / `regex_check` materialised. The original
/// [`Engine`] objects are kept on the registry for re-export and
/// inspection via [`Registry::engines`].
#[derive(Debug, Clone, Deserialize)]
pub struct Registry {
    #[serde(default)]
    engines: BTreeMap<String, Engine>,
    sites: Vec<Site>,
}

impl Registry {
    /// Load the default site list embedded into the crate at build time.
    pub fn default_embedded() -> Result<Self> {
        Self::from_json_str(EMBEDDED_REGISTRY)
    }

    /// Load the default site list *plus* the `WhatsMyName`-derived
    /// supplementary set. `WhatsMyName` data is licensed CC BY-SA 4.0
    /// (see `LICENSE-CC-BY-SA-4.0` at the repo root); enabling this
    /// path means downstream redistribution of the merged scan data
    /// must respect the `ShareAlike` obligation. Sites contributed by
    /// the `WhatsMyName` tranche carry the `source:wmn` tag for
    /// provenance.
    ///
    /// Engines from the WMN tranche merge with the MIT tranche;
    /// case-insensitive site-name collisions resolve in favour of the
    /// MIT-tranche entry (the hand-curated Sherlock/Maigret-derived
    /// signature wins; the WMN duplicate is dropped). Returns an
    /// error only if either tranche fails its own validation —
    /// engine references are checked across the merged set.
    pub fn default_embedded_with_wmn() -> Result<Self> {
        let mut base = Self::default_embedded()?;
        let wmn: Self = serde_json::from_str(EMBEDDED_WMN_REGISTRY)?;
        let existing_names: HashSet<String> =
            base.sites.iter().map(|s| s.name.to_lowercase()).collect();
        // URL-claim only counts enabled base entries — the dedup pattern
        // keeps disabled siblings at the canonical's URL, and a WMN
        // entry colliding with one of *those* is no worse than colliding
        // with the canonical.
        let claimed_urls: HashSet<String> = base
            .sites
            .iter()
            .filter(|s| !s.disabled)
            .map(|s| s.url.as_str().to_owned())
            .collect();
        for (name, engine) in wmn.engines {
            base.engines.entry(name).or_insert(engine);
        }
        for site in wmn.sites {
            if existing_names.contains(&site.name.to_lowercase()) {
                continue;
            }
            if !site.disabled && claimed_urls.contains(site.url.as_str()) {
                // Base already has an enabled site at this URL; WMN's
                // version would just produce a doubled probe, and
                // validate() would refuse the merged registry. Drop the
                // WMN entry; base canonical wins (same precedence rule
                // we apply for name collisions).
                continue;
            }
            base.sites.push(site);
        }
        base.resolve_engines()?;
        base.validate()?;
        Ok(base)
    }

    /// Parse and validate a registry from a JSON string. Engine
    /// references on each site are resolved before validation;
    /// a site that names an engine which doesn't exist in the
    /// `engines` block fails loading with [`Error::InvalidSite`].
    pub fn from_json_str(json: &str) -> Result<Self> {
        let mut registry: Self = serde_json::from_str(json)?;
        registry.resolve_engines()?;
        registry.apply_tag_derived_policy();
        registry.validate()?;
        Ok(registry)
    }

    /// Inheritable engine templates, keyed by name. Useful for
    /// introspection and for serialising the registry back out;
    /// detection paths read the resolved fields off the sites
    /// directly and don't need to consult this map.
    pub fn engines(&self) -> &BTreeMap<String, Engine> {
        &self.engines
    }

    /// Walk every site's tags for `region:XX` markers and fill
    /// [`AccessPolicy::prefer_geo`](crate::AccessPolicy::prefer_geo)
    /// with the matching country codes. **Soft** routing only — a
    /// site declaring `region:ru` *prefers* a Russian egress when one
    /// is configured but still works from anywhere else; the router
    /// falls back to the default egress on no match rather than
    /// reporting `Uncertain(GeoUnavailable)`.
    ///
    /// Skips sites that already declare a hard
    /// [`AccessPolicy::geo`](crate::AccessPolicy::geo) — explicit
    /// policy wins on conflict, same convention as engine inheritance.
    /// Existing `prefer_geo` entries are also preserved: tag-derived
    /// codes are *added*, not replaced, so a hand-tuned policy can
    /// stack on top of the tag. Invalid country codes (`region:xx` is
    /// not exactly two ASCII letters) are silently skipped — they
    /// shouldn't exist in the registry, and a parse error here would
    /// break the load for a tag the scanner already ignores in
    /// every other context.
    fn apply_tag_derived_policy(&mut self) {
        for site in &mut self.sites {
            if !site.access.geo.is_empty() {
                continue;
            }
            for tag in &site.tags {
                let Some(rest) = tag.strip_prefix("region:") else {
                    continue;
                };
                let Some(cc) = crate::access::CountryCode::new(rest) else {
                    continue;
                };
                if !site.access.prefer_geo.contains(&cc) {
                    site.access.prefer_geo.push(cc);
                }
            }
        }
    }

    /// Merge each engine's inheritable fields into the sites that
    /// reference it. After this call every site's `signals`,
    /// `request_headers` and `regex_check` reflect the effective
    /// values used by the scanner.
    ///
    /// Per-site fields are authoritative: anything declared
    /// explicitly on a site wins on conflict; only empty / unset
    /// fields are filled from the engine.
    fn resolve_engines(&mut self) -> Result<()> {
        for (name, engine) in &self.engines {
            engine.validate(name)?;
        }
        for site in &mut self.sites {
            let Some(name) = &site.engine else {
                continue;
            };
            let Some(engine) = self.engines.get(name) else {
                return Err(Error::InvalidSite {
                    reason: format!(
                        "site {:?}: references engine {name:?} which is not defined",
                        site.name
                    ),
                });
            };
            engine.merge_into(site);
        }
        Ok(())
    }

    /// Read a registry from a JSON file.
    pub fn load_from_path(path: impl AsRef<Path>) -> Result<Self> {
        let bytes = std::fs::read(path)?;
        let json = std::str::from_utf8(&bytes).map_err(|e| Error::InvalidSite {
            reason: format!("registry file is not valid UTF-8: {e}"),
        })?;
        Self::from_json_str(json)
    }

    /// Borrow all sites in load order.
    pub fn sites(&self) -> &[Site] {
        &self.sites
    }

    /// Number of sites.
    pub fn len(&self) -> usize {
        self.sites.len()
    }

    /// True if the registry has no sites (always false for a valid load,
    /// since we'd already have rejected it).
    pub fn is_empty(&self) -> bool {
        self.sites.is_empty()
    }

    /// Apply include/exclude name filters and a tag filter.
    ///
    /// - If `include` is non-empty, only sites whose name contains at least
    ///   one include term (case-insensitive substring) are kept.
    /// - Sites whose name contains any exclude term are dropped.
    /// - If `tags` is non-empty, only sites carrying at least one of the
    ///   requested tags are kept (case-insensitive). A site with no tags is
    ///   therefore dropped by a tag filter — asking for `--tag social` means
    ///   "only social-tagged sites".
    /// - Sites carrying any tag in `exclude_tags` are dropped (e.g.
    ///   `--exclude-tag bot-protected` for a fast clean run).
    /// - **NSFW sites are auto-excluded** (the `nsfw` tag) unless
    ///   `include_nsfw` is `true` or `tags` explicitly asks for `nsfw`.
    ///   This matches Sherlock's `--nsfw` opt-in pattern and prevents
    ///   the default `adler <username>` from surfacing adult-site URLs
    ///   the user didn't ask for.
    /// - Sites are returned by value (cloned) so the result is independent
    ///   of the registry's lifetime — convenient for handing to the executor.
    pub fn filter(
        &self,
        include: &[String],
        exclude: &[String],
        tags: &[String],
        exclude_tags: &[String],
        include_nsfw: bool,
    ) -> Vec<Site> {
        let include: Vec<String> = include.iter().map(|s| s.to_lowercase()).collect();
        let exclude: Vec<String> = exclude.iter().map(|s| s.to_lowercase()).collect();
        let want_tags: Vec<String> = tags.iter().map(|s| s.to_lowercase()).collect();
        let mut drop_tags: Vec<String> = exclude_tags.iter().map(|s| s.to_lowercase()).collect();

        // NSFW gate: auto-exclude unless the caller explicitly opted in,
        // either via `include_nsfw` or by asking for the `nsfw` tag.
        let nsfw_tag = "nsfw".to_owned();
        let asking_for_nsfw = want_tags.contains(&nsfw_tag);
        if !include_nsfw && !asking_for_nsfw && !drop_tags.contains(&nsfw_tag) {
            drop_tags.push(nsfw_tag);
        }

        self.sites
            .iter()
            .filter(|site| {
                // Disabled sites are skipped unconditionally — the bool
                // is meant for parking known-broken entries with a
                // reason comment instead of deleting them, so they
                // never get probed even with a fresh include filter.
                if site.disabled {
                    return false;
                }
                let name = site.name.to_lowercase();
                let included = include.is_empty() || include.iter().any(|i| name.contains(i));
                let excluded = exclude.iter().any(|x| name.contains(x));
                let lower_tags: Vec<String> = site.tags.iter().map(|t| t.to_lowercase()).collect();
                let tagged =
                    want_tags.is_empty() || lower_tags.iter().any(|t| want_tags.contains(t));
                let tag_excluded = lower_tags.iter().any(|t| drop_tags.contains(t));
                included && !excluded && tagged && !tag_excluded
            })
            .cloned()
            .collect()
    }

    /// Distinct tags across all sites, sorted, with the count of sites
    /// carrying each. Powers `--list-tags`.
    pub fn tag_counts(&self) -> Vec<(String, usize)> {
        let mut counts: std::collections::BTreeMap<String, usize> =
            std::collections::BTreeMap::new();
        for site in &self.sites {
            for tag in &site.tags {
                *counts.entry(tag.clone()).or_insert(0) += 1;
            }
        }
        counts.into_iter().collect()
    }

    fn validate(&self) -> Result<()> {
        if self.sites.is_empty() {
            return Err(Error::InvalidSite {
                reason: "registry has no sites".into(),
            });
        }
        for site in &self.sites {
            site.validate()?;
        }
        let mut seen: HashSet<String> = HashSet::new();
        for site in &self.sites {
            let key = site.name.to_lowercase();
            if !seen.insert(key) {
                return Err(Error::InvalidSite {
                    reason: format!("duplicate site name: {:?}", site.name),
                });
            }
        }
        // (URL, signals) uniqueness among ENABLED sites: each
        // (URL template, signal-set) pair should back exactly one live
        // entry. Disabled entries can legitimately share URLs with
        // their canonicals — that's how the `duplicate of <canonical>`
        // dedup pattern works. A second enabled hit at the same URL
        // *and* the same signal array is almost always an importer
        // re-introducing a known duplicate
        // (Sherlock/Maigret/WhatsMyName each name the same site
        // slightly differently); the doctor would otherwise
        // double-probe the URL for an identical verdict.
        //
        // Same URL with *distinct* signals is the legitimate-alias
        // shape — WordPress.com (Public/Private/Deleted) hit the same
        // API endpoint and disambiguate via their `body_present`
        // marker, which the doctor reads as three independent verdicts.
        let mut seen_url_sig: HashMap<(String, String), &str> = HashMap::new();
        for site in &self.sites {
            if site.disabled {
                continue;
            }
            // `serde_json` for the signal key gives a canonical
            // serialisation that doesn't depend on field-order or
            // `Debug` formatting, both of which could shift between
            // Rust releases or after a `#[derive(Debug)]` rearrange.
            // serde_json is already a workspace dep; the cost is
            // ~one allocation per enabled site at load time.
            let sigs_key = serde_json::to_string(&site.signals)
                .expect("Signal derives Serialize and contains no Map<_, _> with non-string keys");
            let key = (site.url.as_str().to_owned(), sigs_key);
            if let Some(prev) = seen_url_sig.insert(key, site.name.as_str()) {
                return Err(Error::InvalidSite {
                    reason: format!(
                        "duplicate (URL, signals) among enabled sites: {:?} and {:?} both back \
                         {:?} with identical signals. Mark one `disabled: true` with \
                         `disabled_reason: \"duplicate of {prev}\"` (or, if the two entries are \
                         supposed to disambiguate via different markers, give each a distinct \
                         signal set).",
                        prev,
                        site.name,
                        site.url.as_str(),
                    ),
                });
            }
        }
        Ok(())
    }
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn embedded_registry_loads_and_validates() {
        let registry = Registry::default_embedded().expect("embedded registry must load");
        // The registry is imported from Sherlock (~450 sites); a floor well
        // above the old hand-written 15 guards against accidental truncation.
        assert!(
            registry.len() >= 100,
            "imported registry should have ≥100 sites, got {}",
            registry.len()
        );
        // Spot-check a couple of well-known entries. (HackerNews used
        // to be here but was pruned 2026-05-26 — its Sherlock-side
        // known_present went stale and the imported signature
        // doctor-failed; can be restored via OVERRIDES in
        // import_sherlock.py with a working account.)
        let names: Vec<&str> = registry.sites().iter().map(|s| s.name.as_str()).collect();
        assert!(names.contains(&"GitHub"));
        assert!(names.contains(&"Reddit"));
        assert!(names.contains(&"Telegram"));
    }

    #[test]
    fn wmn_embedded_registry_loads_and_supersets_default() {
        let base = Registry::default_embedded().unwrap();
        let merged = Registry::default_embedded_with_wmn().expect("WMN-merged registry must load");
        assert!(
            merged.len() > base.len(),
            "WMN merge must add sites: base={} merged={}",
            base.len(),
            merged.len()
        );
        // Every base-tranche name survives the merge; case-insensitive
        // collisions resolve in favour of the MIT-tranche entry.
        let merged_names: HashSet<String> = merged
            .sites()
            .iter()
            .map(|s| s.name.to_lowercase())
            .collect();
        for s in base.sites() {
            assert!(
                merged_names.contains(&s.name.to_lowercase()),
                "merge dropped base-tranche site {:?}",
                s.name
            );
        }
        // At least one WMN-only site carries the provenance tag.
        let has_wmn_tag = merged
            .sites()
            .iter()
            .any(|s| s.tags.iter().any(|t| t == "source:wmn"));
        assert!(has_wmn_tag, "no site carries the source:wmn tag");
    }

    #[test]
    fn rejects_empty_registry() {
        let err = Registry::from_json_str(r#"{ "sites": [] }"#).unwrap_err();
        assert!(matches!(err, Error::InvalidSite { .. }));
    }

    #[test]
    fn rejects_duplicate_site_names() {
        let json = r#"{
            "sites": [
                { "name": "GitHub", "url": "https://github.com/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }] },
                { "name": "github", "url": "https://github.com/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }] }
            ]
        }"#;
        let err = Registry::from_json_str(json).unwrap_err();
        assert!(matches!(err, Error::InvalidSite { .. }));
        assert!(err.to_string().contains("duplicate"));
    }

    #[test]
    fn rejects_duplicate_enabled_urls() {
        // Two enabled sites at the same URL is almost always an importer
        // re-introducing a known duplicate. Reject at load time with a
        // message naming both entries.
        let json = r#"{
            "sites": [
                { "name": "Hub Code", "url": "https://example.com/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }] },
                { "name": "HubCode", "url": "https://example.com/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }] }
            ]
        }"#;
        let err = Registry::from_json_str(json).unwrap_err();
        assert!(matches!(err, Error::InvalidSite { .. }));
        let msg = err.to_string();
        assert!(msg.contains("duplicate (URL, signals)"), "msg: {msg}");
        assert!(msg.contains("Hub Code"), "msg: {msg}");
        assert!(msg.contains("HubCode"), "msg: {msg}");
    }

    #[test]
    fn allows_duplicate_urls_with_distinct_signals() {
        // Same URL, distinct signal sets — this is the legitimate-alias
        // shape (e.g. WordPress.com (Public/Private/Deleted) hit one
        // endpoint and disambiguate via the body marker). Must NOT
        // trigger the URL-uniqueness rule.
        let json = r#"{
            "sites": [
                { "name": "Site Public", "url": "https://example.com/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }] },
                { "name": "Site Private", "url": "https://example.com/{username}",
                  "signals": [{ "kind": "status_found", "codes": [403] }] }
            ]
        }"#;
        let registry = Registry::from_json_str(json).expect("distinct-signal alias must validate");
        assert_eq!(registry.len(), 2);
    }

    #[test]
    fn allows_duplicate_urls_when_one_side_is_disabled() {
        // The dedup pattern that the v0.14 hygiene pass established:
        // canonical stays enabled, the surplus entry gets
        // `disabled: true` + `disabled_reason: "duplicate of <canonical>"`.
        // This shape must continue loading cleanly.
        let json = r#"{
            "sites": [
                { "name": "Hub Code", "url": "https://example.com/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }] },
                { "name": "HubCode", "url": "https://example.com/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "disabled": true,
                  "disabled_reason": "duplicate of Hub Code" }
            ]
        }"#;
        let registry = Registry::from_json_str(json).expect("dedup pattern must validate");
        assert_eq!(registry.len(), 2);
    }

    #[test]
    fn rejects_invalid_site_definition() {
        // Missing {username} placeholder.
        let json = r#"{
            "sites": [
                { "name": "Bad", "url": "https://example.com/",
                  "signals": [{ "kind": "status_found", "codes": [200] }] }
            ]
        }"#;
        assert!(Registry::from_json_str(json).is_err());
    }

    #[test]
    fn rejects_malformed_json() {
        let err = Registry::from_json_str("{").unwrap_err();
        assert!(matches!(err, Error::Json(_)));
    }

    #[test]
    fn filter_include_is_case_insensitive_substring() {
        let registry = Registry::default_embedded().unwrap();
        let only_github = registry.filter(&["github".into()], &[], &[], &[], false);
        assert_eq!(only_github.len(), 1);
        assert_eq!(only_github[0].name, "GitHub");

        let many = registry.filter(&["e".into()], &[], &[], &[], false); // matches anything with "e"
        assert!(many.len() > 1);
    }

    #[test]
    fn filter_exclude_drops_matches() {
        let registry = Registry::default_embedded().unwrap();
        // Include NSFW to keep the test focused on the name-exclude
        // path; the NSFW auto-exclusion is exercised separately.
        let baseline = registry.filter(&[], &[], &[], &[], true);
        let without_github = registry.filter(&[], &["github".into()], &[], &[], true);
        assert!(without_github.iter().all(|s| s.name != "GitHub"));
        // Asserting against the baseline (filtered) count rather than
        // `registry.len()` so this test is robust to changes in the
        // disabled-site count — `len()` includes disabled entries,
        // `filter()` does not.
        assert_eq!(without_github.len(), baseline.len() - 1);
    }

    #[test]
    fn filter_include_and_exclude_compose() {
        let registry = Registry::default_embedded().unwrap();
        // Include "git", then exclude "lab" → keep GitHub, drop GitLab.
        let filtered = registry.filter(&["git".into()], &["lab".into()], &[], &[], false);
        let names: Vec<&str> = filtered.iter().map(|s| s.name.as_str()).collect();
        assert!(names.contains(&"GitHub"));
        assert!(!names.contains(&"GitLab"));
        // Exclude wins over include for sites containing both terms (none here).
    }

    #[test]
    fn filter_with_no_matches_returns_empty() {
        let registry = Registry::default_embedded().unwrap();
        let filtered = registry.filter(&["does-not-exist-xyz".into()], &[], &[], &[], false);
        assert!(filtered.is_empty());
    }

    #[test]
    fn disabled_sites_are_skipped_by_filter() {
        let json = r#"{
            "sites": [
                { "name": "Alive", "url": "https://alive.example/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }] },
                { "name": "Parked", "url": "https://parked.example/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "disabled": true }
            ]
        }"#;
        let registry = Registry::from_json_str(json).unwrap();
        // sites() returns everything including disabled — it's the
        // serialisation view. filter() is the scan view and drops
        // disabled entries.
        assert_eq!(registry.sites().len(), 2);
        let scanned = registry.filter(&[], &[], &[], &[], false);
        let names: Vec<&str> = scanned.iter().map(|s| s.name.as_str()).collect();
        assert_eq!(names, vec!["Alive"]);
    }

    #[test]
    fn source_field_round_trips() {
        let json = r#"{
            "sites": [
                { "name": "Nitter", "url": "https://nitter.example/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "source": "Twitter" }
            ]
        }"#;
        let registry = Registry::from_json_str(json).unwrap();
        assert_eq!(registry.sites()[0].source.as_deref(), Some("Twitter"));
    }

    fn tagged_registry() -> Registry {
        let json = r#"{
            "sites": [
                { "name": "Soc", "url": "https://soc.example/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "tags": ["social", "region:ru"] },
                { "name": "Dev", "url": "https://dev.example/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "tags": ["dev"] },
                { "name": "Plain", "url": "https://plain.example/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }] }
            ]
        }"#;
        Registry::from_json_str(json).unwrap()
    }

    #[test]
    fn tag_filter_keeps_only_matching_tags_and_drops_untagged() {
        let r = tagged_registry();
        let social = r.filter(&[], &[], &["social".into()], &[], false);
        let names: Vec<&str> = social.iter().map(|s| s.name.as_str()).collect();
        assert_eq!(names, ["Soc"], "tag filter should keep only tagged matches");
    }

    #[test]
    fn tag_filter_is_or_within_requested_tags_and_case_insensitive() {
        let r = tagged_registry();
        let either = r.filter(&[], &[], &["DEV".into(), "social".into()], &[], false);
        let names: Vec<&str> = either.iter().map(|s| s.name.as_str()).collect();
        assert_eq!(names, ["Soc", "Dev"]);
    }

    #[test]
    fn no_tag_filter_includes_untagged_sites() {
        let r = tagged_registry();
        assert_eq!(r.filter(&[], &[], &[], &[], false).len(), 3);
    }

    #[test]
    fn exclude_tag_drops_matching_sites() {
        let r = tagged_registry();
        let kept = r.filter(&[], &[], &[], &["social".into()], false);
        let names: Vec<&str> = kept.iter().map(|s| s.name.as_str()).collect();
        // Soc carries "social" → dropped; Dev and untagged Plain remain.
        assert_eq!(names, ["Dev", "Plain"], "{names:?}");
    }

    fn nsfw_registry() -> Registry {
        let json = r#"{
            "sites": [
                { "name": "Family", "url": "https://family.example/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "tags": ["social"] },
                { "name": "Adult", "url": "https://adult.example/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "tags": ["nsfw"] }
            ]
        }"#;
        Registry::from_json_str(json).unwrap()
    }

    #[test]
    fn nsfw_sites_excluded_by_default() {
        let r = nsfw_registry();
        let kept = r.filter(&[], &[], &[], &[], false);
        let names: Vec<&str> = kept.iter().map(|s| s.name.as_str()).collect();
        assert_eq!(names, ["Family"], "nsfw site must be excluded by default");
    }

    #[test]
    fn nsfw_sites_included_when_flag_set() {
        let r = nsfw_registry();
        let kept = r.filter(&[], &[], &[], &[], true);
        assert_eq!(kept.len(), 2, "both sites present with include_nsfw=true");
    }

    #[test]
    fn nsfw_sites_included_when_tag_asked_for_explicitly() {
        // `--tag nsfw` is an explicit opt-in; should bypass the default
        // auto-exclusion even with include_nsfw=false.
        let r = nsfw_registry();
        let kept = r.filter(&[], &[], &["nsfw".into()], &[], false);
        let names: Vec<&str> = kept.iter().map(|s| s.name.as_str()).collect();
        assert_eq!(names, ["Adult"]);
    }

    #[test]
    fn tag_counts_are_sorted_with_per_tag_totals() {
        let r = tagged_registry();
        assert_eq!(
            r.tag_counts(),
            vec![
                ("dev".to_owned(), 1),
                ("region:ru".to_owned(), 1),
                ("social".to_owned(), 1),
            ]
        );
    }

    #[test]
    fn engine_inheritance_fills_empty_site_signals() {
        // Site has no `signals` block — should inherit the engine's.
        let json = r#"{
            "engines": {
                "Discourse": {
                    "signals": [
                        { "kind": "status_found", "codes": [200] },
                        { "kind": "body_absent", "text": "Oops! That page doesn't exist" }
                    ]
                }
            },
            "sites": [
                { "name": "Mozilla Forum", "url": "https://discourse.mozilla.org/u/{username}",
                  "engine": "Discourse" }
            ]
        }"#;
        let r = Registry::from_json_str(json).unwrap();
        let site = &r.sites()[0];
        assert_eq!(site.signals.len(), 2);
        assert_eq!(site.engine.as_deref(), Some("Discourse"));
        // engines map preserved
        assert!(r.engines().contains_key("Discourse"));
    }

    #[test]
    fn site_overrides_engine_signals_on_conflict() {
        // Site declares its own `signals` — engine's must NOT replace them.
        let json = r#"{
            "engines": {
                "Discourse": {
                    "signals": [{ "kind": "status_found", "codes": [200] }]
                }
            },
            "sites": [
                { "name": "Custom", "url": "https://example.com/{username}",
                  "engine": "Discourse",
                  "signals": [
                    { "kind": "status_found", "codes": [200] },
                    { "kind": "status_not_found", "codes": [404] }
                  ] }
            ]
        }"#;
        let r = Registry::from_json_str(json).unwrap();
        // The site-declared 2 signals win over the engine's 1 signal.
        assert_eq!(r.sites()[0].signals.len(), 2);
    }

    #[test]
    fn engine_headers_merge_with_site_headers_per_key() {
        // Engine declares one header; site declares another. Resolved
        // site should carry both. On per-key conflict the site wins.
        let json = r#"{
            "engines": {
                "Foo": {
                    "signals": [{ "kind": "status_found", "codes": [200] }],
                    "request_headers": {
                        "X-Engine": "engine-value",
                        "User-Agent": "engine-ua"
                    }
                }
            },
            "sites": [
                { "name": "S", "url": "https://example.com/{username}",
                  "engine": "Foo",
                  "request_headers": { "User-Agent": "site-ua" } }
            ]
        }"#;
        let r = Registry::from_json_str(json).unwrap();
        let h = &r.sites()[0].request_headers;
        assert_eq!(h.get("X-Engine").map(String::as_str), Some("engine-value"));
        assert_eq!(h.get("User-Agent").map(String::as_str), Some("site-ua"));
    }

    #[test]
    fn missing_engine_reference_fails_load() {
        let json = r#"{
            "engines": {},
            "sites": [
                { "name": "Mock", "url": "https://example.com/{username}",
                  "engine": "DoesNotExist" }
            ]
        }"#;
        let err = Registry::from_json_str(json).unwrap_err();
        assert!(
            err.to_string()
                .contains("references engine \"DoesNotExist\""),
            "expected missing-engine error, got: {err}"
        );
    }

    #[test]
    fn engine_regex_check_inherited_when_site_has_none() {
        let json = r#"{
            "engines": {
                "Bounded": {
                    "signals": [{ "kind": "status_found", "codes": [200] }],
                    "regex_check": "^[a-z]{3,16}$"
                }
            },
            "sites": [
                { "name": "S", "url": "https://example.com/{username}",
                  "engine": "Bounded" }
            ]
        }"#;
        let r = Registry::from_json_str(json).unwrap();
        assert_eq!(r.sites()[0].regex_check.as_deref(), Some("^[a-z]{3,16}$"));
    }

    #[test]
    fn region_tag_auto_populates_prefer_geo() {
        let json = r#"{
            "sites": [
                { "name": "vk.com", "url": "https://vk.com/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "tags": ["region:ru", "social"] }
            ]
        }"#;
        let r = Registry::from_json_str(json).unwrap();
        let prefer = &r.sites()[0].access.prefer_geo;
        assert_eq!(prefer.len(), 1);
        assert_eq!(prefer[0].as_str(), "ru");
        // Hard geo stays empty — the tag is soft.
        assert!(r.sites()[0].access.geo.is_empty());
    }

    #[test]
    fn multiple_region_tags_stack() {
        let json = r#"{
            "sites": [
                { "name": "Pan-Slavic", "url": "https://example.test/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "tags": ["region:ru", "region:by", "region:ua"] }
            ]
        }"#;
        let r = Registry::from_json_str(json).unwrap();
        let codes: Vec<&str> = r.sites()[0]
            .access
            .prefer_geo
            .iter()
            .map(super::super::access::CountryCode::as_str)
            .collect();
        assert_eq!(codes, vec!["ru", "by", "ua"]);
    }

    #[test]
    fn explicit_hard_geo_suppresses_tag_derived_soft() {
        // A site with hard `access.geo = ["pl"]` AND a `region:ru` tag:
        // the explicit hard policy wins, prefer_geo stays empty.
        // Otherwise tag-derived soft would silently re-route a probe
        // that the maintainer deliberately pinned to PL.
        let json = r#"{
            "sites": [
                { "name": "PL-only", "url": "https://example.test/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "tags": ["region:ru"],
                  "access": { "geo": ["pl"] } }
            ]
        }"#;
        let r = Registry::from_json_str(json).unwrap();
        assert_eq!(r.sites()[0].access.geo[0].as_str(), "pl");
        assert!(r.sites()[0].access.prefer_geo.is_empty());
    }

    #[test]
    fn malformed_region_tag_is_ignored() {
        // `region:` followed by something that isn't a 2-letter code:
        // skip it silently rather than reject the whole load. The tag
        // already had no routing semantics in older versions.
        let json = r#"{
            "sites": [
                { "name": "Weird", "url": "https://example.test/{username}",
                  "signals": [{ "kind": "status_found", "codes": [200] }],
                  "tags": ["region:eurasia", "region:r", "region:RU"] }
            ]
        }"#;
        let r = Registry::from_json_str(json).unwrap();
        // Only the valid 2-letter "RU" survives (lowercased to "ru").
        let codes: Vec<&str> = r.sites()[0]
            .access
            .prefer_geo
            .iter()
            .map(super::super::access::CountryCode::as_str)
            .collect();
        assert_eq!(codes, vec!["ru"]);
    }

    #[test]
    fn load_from_path_round_trips_via_tempfile() {
        let mut path = std::env::temp_dir();
        path.push(format!("adler-test-registry-{}.json", std::process::id()));
        std::fs::write(
            &path,
            r#"{
                "sites": [
                    { "name": "Mock", "url": "https://example.com/{username}",
                      "signals": [{ "kind": "status_found", "codes": [200] }] }
                ]
            }"#,
        )
        .unwrap();
        let result = Registry::load_from_path(&path);
        let _ = std::fs::remove_file(&path);
        let registry = result.unwrap();
        assert_eq!(registry.len(), 1);
        assert_eq!(registry.sites()[0].name, "Mock");
    }
}