@geml/geml 1.4.2 → 1.4.3

package/codemap/adapters/crg.mjs

@@ -1,120 +1,120 @@
-// geml-code-graph adapter: code-review-graph SQLite (`graph.db`, tree-sitter based) →
-// the exchange format of docs/DESIGN-geml-code-graph.md §3 (symbols + edges).
-//
-// Everything this adapter emits is syntax-level extraction, so per §3.3 every
-// symbol and edge carries resolution:"heuristic"; resolved targets get
-// confidence:"medium", unresolved ones become `to_text` rows (confidence:"low").
-// CONTAINS is not an edge — containment is document structure (GEP-0002).
-//
-// Beyond the core schema, symbols may carry optional navigation flags the
-// engine knows about: `entry` (a `main` function) and `flow_crit` (max
-// criticality of execution flows entered at this symbol, when >= 0.6).
-import { DatabaseSync } from "node:sqlite";
-
-const EDGE_KIND = {
-  CALLS: "calls",
-  IMPORTS_FROM: "imports",
-  INHERITS: "inherits",
-  TESTED_BY: "tested-by",
-  REFERENCES: "references",
-};
-
-export function extract({ db: dbPath, root }) {
-  const db = new DatabaseSync(dbPath);
-  try {
-    return extractFrom(db, root);
-  } finally {
-    // Always release the handle: an open DatabaseSync keeps graph.db locked for
-    // the process lifetime, so on Windows the caller cannot delete or replace
-    // it (EPERM). try/finally closes it even if extraction throws.
-    db.close();
-  }
-}
-
-function extractFrom(db, root) {
-  const rootFs = root.replace(/\\/g, "/").replace(/\/?$/, "/");
-  const rel = (p) => {
-    p = String(p).replace(/\\/g, "/");
-    return p.startsWith(rootFs) ? p.slice(rootFs.length) : p;
-  };
-
-  const rows = db.prepare(
-    "SELECT id, kind, name, qualified_name, file_path, line_start, line_end, language, is_test FROM nodes",
-  ).all();
-
-  // anchor = "<lang>:<relfile>#<name>", File symbols just "<lang>:<relfile>".
-  // Same-file same-name collisions get ~2, ~3 … ordered by line_start so the
-  // numbering is stable across rebuilds (§4.2 / risk 3).
-  const byKey = new Map();
-  for (const r of rows) {
-    const key = `${r.language ?? "unknown"}:${rel(r.file_path)}#${r.name}`;
-    if (!byKey.has(key)) byKey.set(key, []);
-    byKey.get(key).push(r);
-  }
-  const anchorOf = new Map(); // node rowid -> anchor
-  for (const [key, list] of byKey) {
-    list.sort((a, b) => (a.line_start ?? 0) - (b.line_start ?? 0) || a.id - b.id);
-    list.forEach((r, i) => {
-      const base = r.kind === "File" ? key.slice(0, key.lastIndexOf("#")) : key;
-      anchorOf.set(r.id, i === 0 ? base : `${base}~${i + 1}`);
-    });
-  }
-
-  // Navigation flags from engine-specific tables.
-  const mains = new Set(
-    db.prepare("SELECT id FROM nodes WHERE kind='Function' AND name='main'").all().map((r) => r.id),
-  );
-  const flowCrit = new Map();
-  try {
-    for (const r of db.prepare(
-      "SELECT entry_point_id id, max(criticality) c FROM flows GROUP BY 1 HAVING c >= 0.6",
-    ).all()) flowCrit.set(r.id, r.c);
-  } catch { /* no flows table */ }
-
-  const symbols = rows.map((r) => {
-    const s = {
-      anchor: anchorOf.get(r.id),
-      lang: r.language ?? "unknown",
-      kind: r.kind,
-      name: r.name,
-      file: rel(r.file_path),
-      line_start: r.line_start ?? undefined,
-      line_end: r.line_end ?? undefined,
-      is_test: r.is_test ? true : undefined,
-      entry: mains.has(r.id) ? true : undefined,
-      flow_crit: flowCrit.get(r.id),
-      resolution: "heuristic",
-    };
-    return s;
-  });
-
-  const idByQual = new Map(rows.map((r) => [r.qualified_name, r.id]));
-  const edges = [];
-  for (const e of db.prepare(
-    "SELECT kind, source_qualified, target_qualified, file_path, line FROM edges",
-  ).all()) {
-    const kind = EDGE_KIND[e.kind];
-    if (!kind) continue; // CONTAINS and anything unknown
-    const fromId = idByQual.get(e.source_qualified);
-    if (fromId === undefined) continue; // dangling source: nothing to attach to
-    const toId = idByQual.get(e.target_qualified);
-    const edge = {
-      kind,
-      from: anchorOf.get(fromId),
-      resolution: "heuristic",
-      site: { file: rel(e.file_path), line: e.line ?? 0 },
-    };
-    if (toId !== undefined) {
-      edge.to = anchorOf.get(toId);
-      edge.confidence = "medium";
-    } else {
-      // Keep only a readable short name for the unresolved target (§5.2's
-      // calls-unresolved line) — qualified names here are often paths.
-      edge.to_text = String(e.target_qualified).replace(/\\/g, "/").split("/").pop();
-      edge.confidence = "low";
-    }
-    edges.push(edge);
-  }
-
-  return { symbols, edges };
-}
+// geml-code-graph adapter: code-review-graph SQLite (`graph.db`, tree-sitter based) →
+// the exchange format of docs/DESIGN-geml-code-graph.md §3 (symbols + edges).
+//
+// Everything this adapter emits is syntax-level extraction, so per §3.3 every
+// symbol and edge carries resolution:"heuristic"; resolved targets get
+// confidence:"medium", unresolved ones become `to_text` rows (confidence:"low").
+// CONTAINS is not an edge — containment is document structure (GEP-0002).
+//
+// Beyond the core schema, symbols may carry optional navigation flags the
+// engine knows about: `entry` (a `main` function) and `flow_crit` (max
+// criticality of execution flows entered at this symbol, when >= 0.6).
+import { DatabaseSync } from "node:sqlite";
+
+const EDGE_KIND = {
+  CALLS: "calls",
+  IMPORTS_FROM: "imports",
+  INHERITS: "inherits",
+  TESTED_BY: "tested-by",
+  REFERENCES: "references",
+};
+
+export function extract({ db: dbPath, root }) {
+  const db = new DatabaseSync(dbPath);
+  try {
+    return extractFrom(db, root);
+  } finally {
+    // Always release the handle: an open DatabaseSync keeps graph.db locked for
+    // the process lifetime, so on Windows the caller cannot delete or replace
+    // it (EPERM). try/finally closes it even if extraction throws.
+    db.close();
+  }
+}
+
+function extractFrom(db, root) {
+  const rootFs = root.replace(/\\/g, "/").replace(/\/?$/, "/");
+  const rel = (p) => {
+    p = String(p).replace(/\\/g, "/");
+    return p.startsWith(rootFs) ? p.slice(rootFs.length) : p;
+  };
+
+  const rows = db.prepare(
+    "SELECT id, kind, name, qualified_name, file_path, line_start, line_end, language, is_test FROM nodes",
+  ).all();
+
+  // anchor = "<lang>:<relfile>#<name>", File symbols just "<lang>:<relfile>".
+  // Same-file same-name collisions get ~2, ~3 … ordered by line_start so the
+  // numbering is stable across rebuilds (§4.2 / risk 3).
+  const byKey = new Map();
+  for (const r of rows) {
+    const key = `${r.language ?? "unknown"}:${rel(r.file_path)}#${r.name}`;
+    if (!byKey.has(key)) byKey.set(key, []);
+    byKey.get(key).push(r);
+  }
+  const anchorOf = new Map(); // node rowid -> anchor
+  for (const [key, list] of byKey) {
+    list.sort((a, b) => (a.line_start ?? 0) - (b.line_start ?? 0) || a.id - b.id);
+    list.forEach((r, i) => {
+      const base = r.kind === "File" ? key.slice(0, key.lastIndexOf("#")) : key;
+      anchorOf.set(r.id, i === 0 ? base : `${base}~${i + 1}`);
+    });
+  }
+
+  // Navigation flags from engine-specific tables.
+  const mains = new Set(
+    db.prepare("SELECT id FROM nodes WHERE kind='Function' AND name='main'").all().map((r) => r.id),
+  );
+  const flowCrit = new Map();
+  try {
+    for (const r of db.prepare(
+      "SELECT entry_point_id id, max(criticality) c FROM flows GROUP BY 1 HAVING c >= 0.6",
+    ).all()) flowCrit.set(r.id, r.c);
+  } catch { /* no flows table */ }
+
+  const symbols = rows.map((r) => {
+    const s = {
+      anchor: anchorOf.get(r.id),
+      lang: r.language ?? "unknown",
+      kind: r.kind,
+      name: r.name,
+      file: rel(r.file_path),
+      line_start: r.line_start ?? undefined,
+      line_end: r.line_end ?? undefined,
+      is_test: r.is_test ? true : undefined,
+      entry: mains.has(r.id) ? true : undefined,
+      flow_crit: flowCrit.get(r.id),
+      resolution: "heuristic",
+    };
+    return s;
+  });
+
+  const idByQual = new Map(rows.map((r) => [r.qualified_name, r.id]));
+  const edges = [];
+  for (const e of db.prepare(
+    "SELECT kind, source_qualified, target_qualified, file_path, line FROM edges",
+  ).all()) {
+    const kind = EDGE_KIND[e.kind];
+    if (!kind) continue; // CONTAINS and anything unknown
+    const fromId = idByQual.get(e.source_qualified);
+    if (fromId === undefined) continue; // dangling source: nothing to attach to
+    const toId = idByQual.get(e.target_qualified);
+    const edge = {
+      kind,
+      from: anchorOf.get(fromId),
+      resolution: "heuristic",
+      site: { file: rel(e.file_path), line: e.line ?? 0 },
+    };
+    if (toId !== undefined) {
+      edge.to = anchorOf.get(toId);
+      edge.confidence = "medium";
+    } else {
+      // Keep only a readable short name for the unresolved target (§5.2's
+      // calls-unresolved line) — qualified names here are often paths.
+      edge.to_text = String(e.target_qualified).replace(/\\/g, "/").split("/").pop();
+      edge.confidence = "low";
+    }
+    edges.push(edge);
+  }
+
+  return { symbols, edges };
+}

package/codemap/adapters/joern.mjs

@@ -1,131 +1,131 @@
-// geml-code-graph adapter: Joern CPG export → exchange format (DESIGN §3.2/3.4).
-//
-// Consumes the raw JSONL written by joern-export.sc (methods.jsonl +
-// calls.jsonl) — so the node side never touches the JVM. Everything here is
-// resolution:"cpg". Confidence mapping (§3.4):
-//   exactly one internal callee            → high
-//   several internal callees (dispatch)    → first as `to` + the rest as
-//                                            `candidates`, medium
-//   no internal callee (external/pointer)  → to_text, low (unresolved)
-import { readFileSync } from "node:fs";
-import { join } from "node:path";
-
-const LANG_BY_EXT = {
-  c: "c", h: "c", cc: "cpp", cpp: "cpp", cxx: "cpp", hpp: "cpp", hh: "cpp",
-  java: "java", js: "javascript", mjs: "javascript", ts: "typescript",
-  py: "python", kt: "kotlin", go: "go", rb: "ruby", swift: "swift",
-  cs: "csharp", php: "php",
-};
-const langOf = (file) => LANG_BY_EXT[file.split(".").pop()?.toLowerCase()] ?? "unknown";
-
-const readJsonl = (p) => readFileSync(p, "utf8").split("\n").filter(Boolean).map((l) => JSON.parse(l));
-
-export function extract({ raw, root }) {
-  const rootFs = root ? root.replace(/\\/g, "/").replace(/\/?$/, "/") : "";
-  const rel = (p) => {
-    p = String(p).replace(/\\/g, "/");
-    return rootFs && p.startsWith(rootFs) ? p.slice(rootFs.length) : p.replace(/^\//, "");
-  };
-
-  const methods = readJsonl(join(raw, "methods.jsonl"));
-  const calls = readJsonl(join(raw, "calls.jsonl"));
-
-  // Identity: fullName|signature|file — the same tuple joern-export.sc keys on.
-  const keyOf = (m) => `${m.fullName}|${m.signature}|${rel(m.file)}`;
-
-  // Some frontends (javasrc2cpg: constructors, overload/bridge pairs) emit
-  // several method records with an IDENTICAL fullName|signature|file — one
-  // logical method must become one symbol, so dedupe by key first, keeping the
-  // record with the widest line span (the real body over a synthetic stub).
-  const byKey = new Map();
-  for (const m of methods) {
-    const k = keyOf(m);
-    const prev = byKey.get(k);
-    if (!prev || (m.lineEnd ?? 0) - (m.lineStart ?? 0) > (prev.lineEnd ?? 0) - (prev.lineStart ?? 0)) {
-      byKey.set(k, m);
-    }
-  }
-  const uniqueMethods = [...byKey.values()];
-
-  // Anchors: "<lang>:<relfile>#<name>(<sig>)"; same-file same-name-and-sig
-  // duplicates that remain distinct (different fullName) get ~2, ~3 by line order.
-  const byAnchorBase = new Map();
-  for (const m of uniqueMethods) {
-    const base = `${langOf(m.file)}:${rel(m.file)}#${m.name}(${m.signature})`;
-    if (!byAnchorBase.has(base)) byAnchorBase.set(base, []);
-    byAnchorBase.get(base).push(m);
-  }
-  const anchorByKey = new Map();
-  for (const [base, list] of byAnchorBase) {
-    list.sort((a, b) => (a.lineStart ?? 0) - (b.lineStart ?? 0));
-    list.forEach((m, i) => anchorByKey.set(keyOf(m), i === 0 ? base : `${base}~${i + 1}`));
-  }
-
-  // Class-qualified display names: in a class language a method's natural
-  // short name is `Cls.method` — bare names ("invoke", "execute") repeat in
-  // every other class and read as noise. Constructors are `Cls.new`, static
-  // initialisers `Cls.static{}`. C (no classes: fullName has no dot) keeps
-  // plain names. The anchor keeps the raw identity — presentation only.
-  const displayName = (m) => {
-    const head = String(m.fullName ?? "").split(":")[0]; // pkg.Outer$Inner.method
-    const di = head.lastIndexOf(".");
-    const owner = di > 0 ? head.slice(0, di) : "";
-    const simple = (owner.split(".").pop() || "").split("$").pop() || "";
-    if (m.name === "<init>") return simple ? `${simple}.new` : "new";
-    if (m.name === "<clinit>") return simple ? `${simple}.static{}` : "static{}";
-    return simple ? `${simple}.${m.name}` : m.name;
-  };
-
-  const symbols = [];
-  const seenFiles = new Set();
-  for (const m of uniqueMethods) {
-    const file = rel(m.file);
-    symbols.push({
-      anchor: anchorByKey.get(keyOf(m)),
-      lang: langOf(file),
-      kind: "Function",
-      name: displayName(m),
-      file,
-      line_start: m.lineStart ?? undefined,
-      line_end: m.lineEnd ?? undefined,
-      signature: m.signature || undefined,
-      entry: m.name === "main" ? true : undefined,
-      resolution: "cpg",
-    });
-    // Derive one File symbol per source file so emit gets stable heading ids.
-    if (!seenFiles.has(file)) {
-      seenFiles.add(file);
-      symbols.push({
-        anchor: `${langOf(file)}:${file}`,
-        lang: langOf(file),
-        kind: "File",
-        name: file.split("/").pop(),
-        file,
-        resolution: "cpg",
-      });
-    }
-  }
-
-  const edges = [];
-  for (const c of calls) {
-    const from = anchorByKey.get(`${c.callerFullName}|${c.callerSignature}|${rel(c.callerFile)}`);
-    if (!from) continue;
-    const site = { file: rel(c.callerFile), line: c.line ?? 0 };
-    const targets = (c.callees ?? [])
-      .map((t) => anchorByKey.get(`${t.fullName}|${t.signature}|${rel(t.file)}`))
-      .filter(Boolean);
-    if (targets.length === 0) {
-      edges.push({ kind: "calls", from, to_text: c.name, resolution: "cpg", confidence: "low", site });
-    } else if (targets.length === 1) {
-      edges.push({ kind: "calls", from, to: targets[0], resolution: "cpg", confidence: "high", site });
-    } else {
-      const [first, ...rest] = [...new Set(targets)].sort();
-      edges.push({
-        kind: "calls", from, to: first, resolution: "cpg", confidence: "medium",
-        note: `dispatch, ${targets.length} candidates`, candidates: rest, site,
-      });
-    }
-  }
-
-  return { symbols, edges };
-}
+// geml-code-graph adapter: Joern CPG export → exchange format (DESIGN §3.2/3.4).
+//
+// Consumes the raw JSONL written by joern-export.sc (methods.jsonl +
+// calls.jsonl) — so the node side never touches the JVM. Everything here is
+// resolution:"cpg". Confidence mapping (§3.4):
+//   exactly one internal callee            → high
+//   several internal callees (dispatch)    → first as `to` + the rest as
+//                                            `candidates`, medium
+//   no internal callee (external/pointer)  → to_text, low (unresolved)
+import { readFileSync } from "node:fs";
+import { join } from "node:path";
+
+const LANG_BY_EXT = {
+  c: "c", h: "c", cc: "cpp", cpp: "cpp", cxx: "cpp", hpp: "cpp", hh: "cpp",
+  java: "java", js: "javascript", mjs: "javascript", ts: "typescript",
+  py: "python", kt: "kotlin", go: "go", rb: "ruby", swift: "swift",
+  cs: "csharp", php: "php",
+};
+const langOf = (file) => LANG_BY_EXT[file.split(".").pop()?.toLowerCase()] ?? "unknown";
+
+const readJsonl = (p) => readFileSync(p, "utf8").split("\n").filter(Boolean).map((l) => JSON.parse(l));
+
+export function extract({ raw, root }) {
+  const rootFs = root ? root.replace(/\\/g, "/").replace(/\/?$/, "/") : "";
+  const rel = (p) => {
+    p = String(p).replace(/\\/g, "/");
+    return rootFs && p.startsWith(rootFs) ? p.slice(rootFs.length) : p.replace(/^\//, "");
+  };
+
+  const methods = readJsonl(join(raw, "methods.jsonl"));
+  const calls = readJsonl(join(raw, "calls.jsonl"));
+
+  // Identity: fullName|signature|file — the same tuple joern-export.sc keys on.
+  const keyOf = (m) => `${m.fullName}|${m.signature}|${rel(m.file)}`;
+
+  // Some frontends (javasrc2cpg: constructors, overload/bridge pairs) emit
+  // several method records with an IDENTICAL fullName|signature|file — one
+  // logical method must become one symbol, so dedupe by key first, keeping the
+  // record with the widest line span (the real body over a synthetic stub).
+  const byKey = new Map();
+  for (const m of methods) {
+    const k = keyOf(m);
+    const prev = byKey.get(k);
+    if (!prev || (m.lineEnd ?? 0) - (m.lineStart ?? 0) > (prev.lineEnd ?? 0) - (prev.lineStart ?? 0)) {
+      byKey.set(k, m);
+    }
+  }
+  const uniqueMethods = [...byKey.values()];
+
+  // Anchors: "<lang>:<relfile>#<name>(<sig>)"; same-file same-name-and-sig
+  // duplicates that remain distinct (different fullName) get ~2, ~3 by line order.
+  const byAnchorBase = new Map();
+  for (const m of uniqueMethods) {
+    const base = `${langOf(m.file)}:${rel(m.file)}#${m.name}(${m.signature})`;
+    if (!byAnchorBase.has(base)) byAnchorBase.set(base, []);
+    byAnchorBase.get(base).push(m);
+  }
+  const anchorByKey = new Map();
+  for (const [base, list] of byAnchorBase) {
+    list.sort((a, b) => (a.lineStart ?? 0) - (b.lineStart ?? 0));
+    list.forEach((m, i) => anchorByKey.set(keyOf(m), i === 0 ? base : `${base}~${i + 1}`));
+  }
+
+  // Class-qualified display names: in a class language a method's natural
+  // short name is `Cls.method` — bare names ("invoke", "execute") repeat in
+  // every other class and read as noise. Constructors are `Cls.new`, static
+  // initialisers `Cls.static{}`. C (no classes: fullName has no dot) keeps
+  // plain names. The anchor keeps the raw identity — presentation only.
+  const displayName = (m) => {
+    const head = String(m.fullName ?? "").split(":")[0]; // pkg.Outer$Inner.method
+    const di = head.lastIndexOf(".");
+    const owner = di > 0 ? head.slice(0, di) : "";
+    const simple = (owner.split(".").pop() || "").split("$").pop() || "";
+    if (m.name === "<init>") return simple ? `${simple}.new` : "new";
+    if (m.name === "<clinit>") return simple ? `${simple}.static{}` : "static{}";
+    return simple ? `${simple}.${m.name}` : m.name;
+  };
+
+  const symbols = [];
+  const seenFiles = new Set();
+  for (const m of uniqueMethods) {
+    const file = rel(m.file);
+    symbols.push({
+      anchor: anchorByKey.get(keyOf(m)),
+      lang: langOf(file),
+      kind: "Function",
+      name: displayName(m),
+      file,
+      line_start: m.lineStart ?? undefined,
+      line_end: m.lineEnd ?? undefined,
+      signature: m.signature || undefined,
+      entry: m.name === "main" ? true : undefined,
+      resolution: "cpg",
+    });
+    // Derive one File symbol per source file so emit gets stable heading ids.
+    if (!seenFiles.has(file)) {
+      seenFiles.add(file);
+      symbols.push({
+        anchor: `${langOf(file)}:${file}`,
+        lang: langOf(file),
+        kind: "File",
+        name: file.split("/").pop(),
+        file,
+        resolution: "cpg",
+      });
+    }
+  }
+
+  const edges = [];
+  for (const c of calls) {
+    const from = anchorByKey.get(`${c.callerFullName}|${c.callerSignature}|${rel(c.callerFile)}`);
+    if (!from) continue;
+    const site = { file: rel(c.callerFile), line: c.line ?? 0 };
+    const targets = (c.callees ?? [])
+      .map((t) => anchorByKey.get(`${t.fullName}|${t.signature}|${rel(t.file)}`))
+      .filter(Boolean);
+    if (targets.length === 0) {
+      edges.push({ kind: "calls", from, to_text: c.name, resolution: "cpg", confidence: "low", site });
+    } else if (targets.length === 1) {
+      edges.push({ kind: "calls", from, to: targets[0], resolution: "cpg", confidence: "high", site });
+    } else {
+      const [first, ...rest] = [...new Set(targets)].sort();
+      edges.push({
+        kind: "calls", from, to: first, resolution: "cpg", confidence: "medium",
+        note: `dispatch, ${targets.length} candidates`, candidates: rest, site,
+      });
+    }
+  }
+
+  return { symbols, edges };
+}