Merge release/v5-prerelease (#281, one lockfile model per ecosystem) into v5/one-hosted-engine

#281's format-registry changes to the hosted flow land where this branch
moved that code: core hosted::engine derives REDIRECT_CANDIDATE_FILES
and file_ecosystem from formats::registry, hosted::guidance re-exports
the pnpm lock-version sniffs from formats::pnpm, and the in-memory
root detection reads registry::root_marker. The CLI hosted_memory
redirect.rs #281 edited is already gone on this branch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01KHZ8uzdXfkG2zH8ZYDG8ju

Author: claude

crates/socket-patch-cli/src/commands/get.rs

@@ -11,6 +11,7 @@ use socket_patch_core::api::types::{
 };
 use socket_patch_core::crawlers::fuzzy_match::fuzzy_match_packages;
 use socket_patch_core::crawlers::{CrawlerOptions, Ecosystem};
+use socket_patch_core::formats::pnpm::PnpmLock;
 use socket_patch_core::manifest::operations::{read_manifest, write_manifest};
 pub(crate) use socket_patch_core::manifest::records::record_from_patch_response;
 use socket_patch_core::manifest::records::{build_patch_record, files_for_manifest};
@@ -1425,47 +1426,6 @@ fn purl_has_version(purl: &str) -> bool {
         })
 }
 
-/// Does the raw pnpm-lock text RESOLVE `name@version`? Boundary-anchored
-/// probes over the three lock grammars — a plain `contains` collides on
-/// version prefixes (`left-pad@1.3.0` matches inside
-/// `left-pad@1.3.0-beta.1`), name suffixes (`pad@1.3.0` inside
-/// `left-pad@1.3.0`), and unscoped-inside-scoped names (`name@1.0.0` inside
-/// `@scope/name@1.0.0`). The needles cover v6/v9's `name@version` and v5's
-/// `/name/version` key spellings; a match counts only when the preceding
-/// char cannot extend the name (start/whitespace/quote, or a `/` delimiter
-/// itself preceded by such a boundary) and the following char cannot extend
-/// the version (so `:`, `'`, `(`, and v5's `_peer` suffix all accept).
-/// Heuristic by design: a false negative degrades to a calm skip, a false
-/// positive costs one grant request the rewriter's per-dep confirmation
-/// then ignores.
-fn pnpm_lock_resolves(text: &str, name: &str, version: &str) -> bool {
-    let version_boundary = |c: char| !(c.is_ascii_alphanumeric() || matches!(c, '.' | '-' | '+'));
-    let name_boundary = |c: char| matches!(c, ' ' | '\t' | '\n' | '\r' | '\'' | '"');
-    for needle in [format!("{name}@{version}"), format!("/{name}/{version}")] {
-        for (pos, _) in text.match_indices(needle.as_str()) {
-            let before_ok = match text[..pos].chars().next_back() {
-                None => true,
-                // v5/v6's leading key delimiter — legitimate only when the
-                // char before it is itself a boundary (otherwise this is a
-                // scoped `@scope/<name>` tail: a DIFFERENT package).
-                Some('/') => text[..pos - 1]
-                    .chars()
-                    .next_back()
-                    .is_none_or(name_boundary),
-                Some(c) => name_boundary(c),
-            };
-            let after_ok = text[pos + needle.len()..]
-                .chars()
-                .next()
-                .is_none_or(version_boundary);
-            if before_ok && after_ok {
-                return true;
-            }
-        }
-    }
-    false
-}
-
 /// Outcome of the coarse installed-VERSION narrowing over a CVE/GHSA/PURL
 /// search fan-out (see [`filter_to_installed_purls`]).
 struct InstalledNarrowing {
@@ -1504,7 +1464,7 @@ struct InstalledNarrowing {
 /// `yarn_pnp_unsupported`, not a false "not installed"). pnpm PnP skips
 /// carry `pnpm_pnp_unsupported` in agent/vendored modes; hosted mode — the
 /// refusal's own remedy — keeps the versions the raw pnpm-lock.yaml text
-/// resolves ([`pnpm_lock_resolves`]), labels a judged miss
+/// resolves ([`PnpmLock::resolves`]), labels a judged miss
 /// `package_not_installed` like any other mode, and reserves the layout
 /// code for an unreadable lock (no judgment possible).
 ///
@@ -1576,6 +1536,7 @@ async fn filter_to_installed_purls(
     let pnpm_pnp_lock_text: Option<String> = (pnp_pnpm && mode == super::scan::ScanMode::Hosted)
         .then(|| std::fs::read_to_string(common.cwd.join("pnpm-lock.yaml")).ok())
         .flatten();
+    let pnpm_pnp_lock = pnpm_pnp_lock_text.as_deref().map(PnpmLock::parse);
 
     let mut out = InstalledNarrowing {
         kept: Vec::new(),
@@ -1605,8 +1566,8 @@ async fn filter_to_installed_purls(
             // The pnpm PnP refusal's own remedy is the hosted lockfile
             // rewrite — but only for versions the lock ACTUALLY resolves:
             // keeping the whole fan-out would request grants for every
-            // version ever patched. Anchored probe over the raw lock text
-            // (see `pnpm_lock_resolves`); a hit is kept (the rewriter's
+            // version ever patched. The lock model's key probe
+            // (`PnpmLock::resolves`); a hit is kept (the rewriter's
             // per-dep confirmation still decides). A judged MISS is a
             // genuine "version not resolved" verdict — the layout blocked
             // nothing — so it carries the same `package_not_installed` code
@@ -1615,9 +1576,9 @@ async fn filter_to_installed_purls(
             let decoded = canon(&result.purl);
             let coord = decoded.strip_prefix("pkg:npm/").unwrap_or(&decoded);
             if mode == super::scan::ScanMode::Hosted {
-                match (pnpm_pnp_lock_text.as_deref(), coord.rsplit_once('@')) {
-                    (Some(text), Some((name, version))) => {
-                        if pnpm_lock_resolves(text, name, version) {
+                match (&pnpm_pnp_lock, coord.rsplit_once('@')) {
+                    (Some(lock), Some((name, version))) => {
+                        if lock.resolves(name, version) {
                             out.kept.push(result.clone());
                             continue;
                         }
@@ -3914,77 +3875,6 @@ pub(crate) fn base64_decode(input: &str) -> Result<Vec<u8>, String> {
 mod tests {
     use super::*;
 
-    /// The pnpm-PnP hosted lock probe must be boundary-anchored: plain
-    /// substring matching collides on version prefixes, name suffixes, and
-    /// unscoped-inside-scoped names.
-    #[test]
-    fn pnpm_lock_resolves_is_boundary_anchored() {
-        // v9/v6/v5 key spellings all resolve.
-        assert!(pnpm_lock_resolves(
-            "lockfileVersion: '9.0'\n\nsnapshots:\n\n  left-pad@1.3.0:\n",
-            "left-pad",
-            "1.3.0"
-        ));
-        assert!(pnpm_lock_resolves(
-            "  /left-pad@1.3.0:\n    resolution: {}\n",
-            "left-pad",
-            "1.3.0"
-        ));
-        assert!(pnpm_lock_resolves(
-            "  /left-pad/1.3.0:\n    resolution: {}\n",
-            "left-pad",
-            "1.3.0"
-        ));
-        // Peer-qualified keys still resolve: v9 `(peer)` and v5 `_peer`.
-        assert!(pnpm_lock_resolves(
-            "  'left-pad@1.3.0(react@18.0.0)':\n",
-            "left-pad",
-            "1.3.0"
-        ));
-        assert!(pnpm_lock_resolves(
-            "  /left-pad/1.3.0_react@18.0.0:\n",
-            "left-pad",
-            "1.3.0"
-        ));
-        // Scoped names resolve in both quoted-v9 and v6 spellings.
-        assert!(pnpm_lock_resolves(
-            "  '@scope/name@1.0.0':\n",
-            "@scope/name",
-            "1.0.0"
-        ));
-        assert!(pnpm_lock_resolves(
-            "  /@scope/name@1.0.0:\n",
-            "@scope/name",
-            "1.0.0"
-        ));
-
-        // Version-prefix collision: 1.3.0 must NOT match 1.3.0-beta.1.
-        assert!(!pnpm_lock_resolves(
-            "  left-pad@1.3.0-beta.1:\n",
-            "left-pad",
-            "1.3.0"
-        ));
-        // Name-suffix collision: `pad` must NOT match inside `left-pad`.
-        assert!(!pnpm_lock_resolves("  left-pad@1.3.0:\n", "pad", "1.3.0"));
-        assert!(!pnpm_lock_resolves("  /left-pad/1.3.0:\n", "pad", "1.3.0"));
-        // Unscoped-inside-scoped: `name` must NOT match `@scope/name`.
-        assert!(!pnpm_lock_resolves(
-            "  '@scope/name@1.0.0':\n",
-            "name",
-            "1.0.0"
-        ));
-        assert!(!pnpm_lock_resolves(
-            "  /@scope/name@1.0.0:\n",
-            "name",
-            "1.0.0"
-        ));
-        // Absent version: never resolves.
-        assert!(!pnpm_lock_resolves(
-            "  left-pad@1.3.0:\n",
-            "left-pad",
-            "2.0.0"
-        ));
-    }
     use socket_patch_core::api::types::{PatchFileResponse, VulnerabilityResponse};
     use std::collections::HashMap;
 
@@ -4963,36 +4853,6 @@ mod tests {
         );
     }
 
-    // --- pnpm_lock_resolves: needle at byte 0 ------------------------------
-    // The boundary probe reads the char BEFORE the match; a match at the very
-    // start of the text has none (`None => true`). A regression that indexes
-    // `text[..pos - 1]` unconditionally would underflow/panic here.
-
-    #[test]
-    fn pnpm_lock_resolves_needle_at_start_of_text() {
-        // pos == 0, plain v9 spelling: no preceding char is a valid boundary.
-        assert!(pnpm_lock_resolves("left-pad@1.3.0:\n", "left-pad", "1.3.0"));
-        // pos == 0, v5/v6 `/name/version` and `/name@version` spellings: the
-        // leading `/` delimiter itself has nothing before it.
-        assert!(pnpm_lock_resolves(
-            "/left-pad/1.3.0:\n",
-            "left-pad",
-            "1.3.0"
-        ));
-        assert!(pnpm_lock_resolves(
-            "/left-pad@1.3.0:\n",
-            "left-pad",
-            "1.3.0"
-        ));
-        // Still boundary-checked at the start of text: a scoped tail whose
-        // name begins mid-token must NOT match.
-        assert!(!pnpm_lock_resolves(
-            "@scope/left-pad@1.3.0:\n",
-            "left-pad",
-            "1.3.0"
-        ));
-    }
-
     // --- write_all_patch_blobs ---------------------------------------------
     // The per-patch fan-out over write_blob_entry: the FIRST bad entry must
     // fail the whole patch (Err(())) and leave nothing outside the blobs

crates/socket-patch-cli/src/commands/scan/hosted.rs

@@ -2092,7 +2092,6 @@ mod tests {
         wrap_tokens, wrap_words, TAKEOVER_INFO_CODES,
     };
     use super::{wheel_metadata_concurrency, WHEEL_METADATA_CONCURRENCY};
-    use socket_patch_core::constants::npm_family;
     use socket_patch_core::hosted::engine::REDIRECT_CANDIDATE_FILES;
     use socket_patch_core::patch::redirect::DepOverride;
     use socket_patch_core::utils::concurrent::API_CONCURRENCY_ENV;
@@ -3316,25 +3315,54 @@ mod tests {
     }
 
     #[test]
-    fn redirect_candidates_match_the_shared_npm_family_table() {
-        // Drift guard, both directions, without classifying the non-npm
-        // rows: every table row flagged redirect_candidate must be in the
-        // candidate list, and no npm-family row NOT so flagged may appear
-        // (binary candidates are read separately).
-        for name in npm_family::names_with(|r| r.redirect_candidate) {
-            assert!(
-                REDIRECT_CANDIDATE_FILES.contains(&name),
-                "{name} is flagged redirect_candidate but missing from \
-                 REDIRECT_CANDIDATE_FILES"
-            );
-        }
-        for name in npm_family::names_with(|r| !r.redirect_candidate) {
-            assert!(
-                !REDIRECT_CANDIDATE_FILES.contains(&name),
-                "{name} is deliberately NOT a redirect candidate (see the \
-                 npm_family table) but appears in REDIRECT_CANDIDATE_FILES"
-            );
-        }
+    fn redirect_candidates_are_pinned_by_value() {
+        // Hardcoded on purpose: the candidate list is derived from the
+        // format registry, so a row dropped (or a HOSTED flag lost) there
+        // must fail here instead of silently shrinking what hosted reads.
+        assert_eq!(
+            *REDIRECT_CANDIDATE_FILES,
+            [
+                "package-lock.json",
+                "npm-shrinkwrap.json",
+                "pnpm-lock.yaml",
+                "shrinkwrap.yaml",
+                "node_modules/.modules.yaml",
+                "yarn.lock",
+                ".yarnrc.yml",
+                "bun.lock",
+                "bun.lockb",
+                "vlt-lock.json",
+                "vlt.json",
+                "node_modules/.vlt-lock.json",
+                "requirements.txt",
+                "uv.lock",
+                "poetry.lock",
+                "pdm.lock",
+                "Pipfile.lock",
+                "pyproject.toml",
+                "hatch.toml",
+                "Cargo.toml",
+                "Cargo.lock",
+                ".cargo/config.toml",
+                ".cargo/config",
+                "composer.lock",
+                "nuget.config",
+                "packages.lock.json",
+                "Gemfile",
+                "Gemfile.lock",
+                "gems.rb",
+                "gems.locked",
+                "go.mod",
+                "go.sum",
+                "pom.xml",
+                ".mvn/maven.config",
+                ".mvn/checksums/checksums.sha256",
+                "settings.gradle",
+                "settings.gradle.kts",
+                "build.gradle",
+                "build.gradle.kts",
+            ]
+        );
     }
     // ── Human-output formatting ────────────────────────────────────────────
 

crates/socket-patch-cli/src/commands/vendored_backend/repair.rs

@@ -28,6 +28,7 @@ use std::path::{Path, PathBuf};
 
 use socket_patch_core::api::client::{get_api_client_with_overrides, ApiClient};
 use socket_patch_core::constants::SOCKET_DIR;
+use socket_patch_core::formats::registry;
 use socket_patch_core::manifest::schema::{PatchManifest, PatchRecord};
 use socket_patch_core::patch::copy_tree::remove_tree;
 use socket_patch_core::utils::fs::{
@@ -57,36 +58,6 @@ struct Candidate {
     reason: &'static str,
 }
 
-/// Files the vendor backends rewire — the search space for
-/// `.socket/vendor/<eco>/<uuid>/<leaf>` references. The Python locks the
-/// root LISTS (`pylock*.toml`, `*.py.lock` + script) and the requirements
-/// `-r` include tree are appended at scan time.
-const WIRING_FILES: &[&str] = &[
-    "vlt-lock.json",
-    "package-lock.json",
-    "npm-shrinkwrap.json",
-    "pnpm-lock.yaml",
-    "yarn.lock",
-    "bun.lock",
-    "package.json",
-    "Cargo.toml",
-    "Cargo.lock",
-    // Pre-v5 vendored cargo wiring (migrated into Cargo.toml on re-run).
-    ".cargo/config.toml",
-    ".cargo/config",
-    "go.mod",
-    "composer.json",
-    "composer.lock",
-    "Gemfile",
-    "Gemfile.lock",
-    "uv.lock",
-    "pyproject.toml",
-    "poetry.lock",
-    "pdm.lock",
-    "Pipfile.lock",
-    "requirements.txt",
-];
-
 /// Scan the wiring-bearing files for vendored-artifact references,
 /// returning deduped `(ecosystem, uuid, artifact relpath)` triples. Pure
 /// text scan plus native binary Bun resolution records and the canonical
@@ -148,13 +119,14 @@ pub(crate) async fn scan_vendor_references(project_root: &Path) -> Vec<(String,
 }
 
 /// Every wiring-bearing file name the vendor backends may rewrite, relative
-/// to `project_root`: [`WIRING_FILES`], vlt importer manifests, the Python
+/// to `project_root`: the registry's vendored wiring files
+/// ([`registry::VENDORED`]), vlt importer manifests, the Python
 /// locks the root lists (and their scripts) and the requirements `-r`
 /// include tree. Sorted and deduplicated; entries need not exist.
 async fn wiring_files(project_root: &Path) -> Vec<String> {
-    let mut files: Vec<String> = WIRING_FILES
-        .iter()
-        .map(|file| (*file).to_string())
+    let mut files: Vec<String> = registry::paths_with(registry::VENDORED)
+        .into_iter()
+        .map(str::to_string)
         .collect();
     files.extend(vendor::vlt_lock::vlt_importer_package_jsons(project_root).await);
     if let Ok(paths) = socket_patch_core::utils::python_lock::python_lock_paths(project_root) {