@geml/geml - npm Package Compare versions

Comparing version 1.7.0 to 1.7.1

codemap/build.mjs

@@ -39,3 +39,3 @@ #!/usr/bin/env node
 import { writeFileSync, mkdirSync, existsSync, readFileSync, statSync } from "node:fs";
-import { join, resolve, basename, dirname, relative } from "node:path";
+import { join, resolve, basename, dirname, relative, sep } from "node:path";
 import { fileURLToPath } from "node:url";
@@ -104,3 +104,3 @@ import { execFileSync, spawnSync } from "node:child_process";
     root: rootAbs, globs: excludeGlobs0, gitignore: !args.includes("--no-gitignore"),
-    files: [...files, ...manifests, ...pkgs], exec: execFileSync,
+    files: [...files, ...manifests, ...pkgs], run: execFileSync,
   });
@@ -141,3 +141,6 @@ const jobs = detectLanguages(rootAbs, { files, manifests, pkgs, excluder });
   // spaced launcher PATH is a full path, so quoting it keeps %~dp0 correct.
-  const q = (s) => (/[\s"]/.test(String(s)) ? `"${String(s).replace(/"/g, '\\"')}"` : String(s));
+  // WHEN it does quote, it defers to shq below: escaping only `"` left a
+  // trailing backslash (a path ending `...\`) escaping our own closing quote,
+  // which merges the next token into this one. Same CRT rules, one source.
+  const q = (s) => (/[\s"]/.test(String(s)) ? shq(s) : String(s));
   // Hardened quote for command ARGUMENTS on win32. Node does NOT escape args
@@ -384,3 +387,3 @@ // under shell:true — it only concatenates them (Node DEP0190) — so an
   files: [...new Set(symbols.map((s) => s.file))],
-  exec: execFileSync,
+  run: execFileSync,
 });
@@ -398,3 +401,3 @@ const kept = symbols.filter((s) => !excluder(s.file));
     root: rootAbs, globs: excludeGlobs, gitignore: !args.includes("--no-gitignore"),
-    files: [...c.files, ...c.manifests, ...c.pkgs], exec: execFileSync,
+    files: [...c.files, ...c.manifests, ...c.pkgs], run: execFileSync,
   });
@@ -428,3 +431,14 @@ entryHints = detectEntries(rootAbs, {
     files: scanFiles,
-    readText: (rel) => { try { return readFileSync(join(rootAbs, ...rel.split("/")), "utf8"); } catch { return null; } },
+    // `rel` is indexer OUTPUT, not a path this build authored, so a `..`
+    // segment must not turn a source read into an escape from the scanned
+    // root. Same gate the document resolver uses: resolve first, then require
+    // the result to BE the root or sit under it — compared with the separator,
+    // so a sibling named `<root>-evil` is not mistaken for a child.
+    readText: (rel) => {
+      try {
+        const p = resolve(rootAbs, ...String(rel).split("/"));
+        if (p !== rootAbs && !p.startsWith(rootAbs + sep)) return null;
+        return readFileSync(p, "utf8");
+      } catch { return null; }
+    },
   });
@@ -431,0 +445,0 @@ for (const e of httpEdges) edges.push(e);

codemap/exclude.mjs

@@ -18,13 +18,13 @@ // Source exclusion for the codemap build.
 export function globToRegExp(glob) {
+  // The glob comes from a `--exclude` argument, so nothing in it may reach the
+  // compiled pattern as *syntax*. Split on the wildcards, keeping them (the
+  // capture group), which leaves the array strictly alternating: even indices
+  // are literal text, odd indices are `*`, `**` or `**/`. Literals go through a
+  // total regex-metacharacter escape; wildcards map to fixed patterns. Neither
+  // path can carry an unescaped metacharacter through.
+  const parts = String(glob).split(/(\*\*\/?|\*)/);
   let re = "";
-  for (let i = 0; i < glob.length; i++) {
-    const c = glob[i];
-    if (c === "*") {
-      if (glob[i + 1] === "*") { re += ".*"; i++; if (glob[i + 1] === "/") i++; }
-      else re += "[^/]*";
-    } else if ("\\^$+?.()|{}[]".includes(c)) {
-      re += "\\" + c;
-    } else {
-      re += c;
-    }
+  for (let i = 0; i < parts.length; i++) {
+    if (i % 2 === 0) re += parts[i].replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+    else re += parts[i] === "*" ? "[^/]*" : ".*"; // `**` and `**/` span separators
   }
@@ -37,6 +37,10 @@ return new RegExp("^" + re + "$");
 // the dir is not a repo — both mean "ignore nothing", not a build failure.
-export function gitIgnored(root, files, exec = _execFileSync) {
+// The injected runner is named `run`, not `exec`: it is always an execFile-shaped
+// (program, args[]) call that spawns NO shell, whereas a callback named `exec`
+// reads — to a human skimming, and to a static analyser — as the shell-string
+// child_process API. The name should not imply the dangerous one.
+export function gitIgnored(root, files, run = _execFileSync) {
   if (!files.length) return new Set();
   try {
-    const out = exec("git", ["-C", root, "check-ignore", "--stdin"], { input: files.join("\n"), encoding: "utf8" });
+    const out = run("git", ["-C", root, "check-ignore", "--stdin"], { input: files.join("\n"), encoding: "utf8" });
     return new Set(out.split(/\r?\n/).filter(Boolean));
@@ -50,6 +54,6 @@ } catch (e) {
 // Build a predicate (file) => shouldExclude.
-export function makeExcluder({ root, globs = [], gitignore = true, files = [], exec } = {}) {
+export function makeExcluder({ root, globs = [], gitignore = true, files = [], run } = {}) {
   const res = globs.map(globToRegExp);
-  const ignored = gitignore ? gitIgnored(root, files, exec) : new Set();
+  const ignored = gitignore ? gitIgnored(root, files, run) : new Set();
   return (file) => ignored.has(file) || res.some((r) => r.test(file));
 }

codemap/refresh.mjs

@@ -98,3 +98,6 @@ #!/usr/bin/env node
 //     metachar inside a dir-name argument is therefore inert.
-const q = (s) => (/[\s"]/.test(String(s)) ? `"${String(s).replace(/"/g, '\\"')}"` : String(s));
+// q quotes only when it must; WHEN it does it defers to shq, so a program path
+// ending in a backslash cannot escape our own closing quote and swallow the
+// next token. One set of CRT rules, in one place.
+const q = (s) => (/[\s"]/.test(String(s)) ? shq(s) : String(s));
 const shq = (s) => `"${String(s).replace(/(\\*)"/g, '$1$1\\"').replace(/(\\+)$/, '$1$1')}"`;
@@ -101,0 +104,0 @@ // Human-readable render of a step for the log / refusal message — DISPLAY ONLY,

codemap/verify.mjs

@@ -40,3 +40,6 @@ #!/usr/bin/env node
 // CRT rules for embedded " / trailing \.
-const q = (s) => (/[\s"]/.test(String(s)) ? `"${String(s).replace(/"/g, '\\"')}"` : String(s));
+// q quotes only when it must; WHEN it does it defers to shq, so a program path
+// ending in a backslash cannot escape our own closing quote and swallow the
+// next token. One set of CRT rules, in one place.
+const q = (s) => (/[\s"]/.test(String(s)) ? shq(s) : String(s));
 const shq = (s) => `"${String(s).replace(/(\\*)"/g, '$1$1\\"').replace(/(\\+)$/, '$1$1')}"`;
@@ -43,0 +46,0 @@ // A codemap document's `src=` routes are written relative to the indexed

dist/from-md.js

@@ -58,3 +58,8 @@ // GEML reference parser — Markdown → GEML conversion.
     const bareSafe = /^[^\s"]+$/.test(v);
-    return bareSafe ? `${m[1]}=${v}` : `${m[1]}="${v.replace(/"/g, '\\"')}"`;
+    // No backslash escape here — a `data` block value is read by `coerce`, which
+    // simply strips the outer quotes (§4); GEML defines no `\"` escape at all.
+    // Emitting one wrote the backslash into the parsed value, so `a"b` came back
+    // as `a\"b`. Quoting the value verbatim round-trips instead: coerce keeps
+    // everything between the first and last quote, embedded quotes included.
+    return bareSafe ? `${m[1]}=${v}` : `${m[1]}="${v}"`;
 }
@@ -61,0 +66,0 @@ // Rewrite Markdown autolinks `<https://…>` / `<mailto:…>` into GEML links

dist/to-md.js

@@ -51,6 +51,28 @@ // GEML -> Markdown projection (the inverse direction of from-md.ts).
 }
+// Escape a `|` so GFM keeps it inside the cell instead of splitting the row.
+// GFM resolves backslash escapes in a row BEFORE it splits on `|`, so a
+// backslash run sitting right in front of our escape would eat it: a code span
+// holding `a\|b` became `a\\|b`, which reads as a literal backslash followed by
+// an UNescaped pipe — a spurious cell break. Double any such run first, then
+// escape the pipe. Runs already produced by escText (`\\` for a literal
+// backslash) survive this unchanged, so pre-rendered Markdown stays intact.
+//
+// The backslash run is matched as `\\+\|?` — one ATOMIC token, run and pipe
+// together — not as `(\\*)\|`. The latter is quadratic: on a cell holding a
+// long run of backslashes and no pipe, the engine matches the run from every
+// index in it and fails at the required `|` each time. Here the greedy `\\+`
+// takes the whole run in one match and the trailing `\|?` is optional, so
+// nothing backtracks and each character is visited once.
+function escPipe(s) {
+    return s.replace(/\\+\|?|\|/g, (m) => {
+        if (m.charAt(m.length - 1) !== "|")
+            return m; // a run with no pipe after it
+        const bs = m.slice(0, -1); // the run that would otherwise eat our escape
+        return bs + bs + "\\|";
+    });
+}
 // Inline text for a table cell: render inlines, then neutralise the two bytes
 // that would break a GFM cell.
 function cellText(c) {
-    return seq(c.inlines).replace(/\|/g, "\\|").replace(/\n/g, " ");
+    return escPipe(seq(c.inlines)).replace(/\n/g, " ");
 }
@@ -76,3 +98,3 @@ // ---------------------------------------------------------------------------
         lines.push(`*${t.caption}*`, "");
-    lines.push(`| ${cols.map((c) => c.replace(/\|/g, "\\|")).join(" | ")} |`);
+    lines.push(`| ${cols.map(escPipe).join(" | ")} |`);
     lines.push(`| ${cols.map((_, i) => sep(t.align[i])).join(" | ")} |`);
@@ -79,0 +101,0 @@ const pad = (cells) => {

package.json

 {
   "name": "@geml/geml",
-  "version": "1.7.0",
+  "version": "1.7.1",
   "mcpName": "io.github.geml-spec/geml",
@@ -5,0 +5,0 @@ "publishConfig": {

New alerts

AI-detected potential code anomaly

Supply chain risk

AI has identified unusual behaviors that may pose a security risk.

Found 2 instances

URL strings

Supply chain risk

Package contains fragments of external URLs or IP addresses, which the package may be accessing at runtime.

Fixed alerts

AI-detected potential code anomaly

Supply chain risk

AI has identified unusual behaviors that may pose a security risk.

Found 2 instances

URL strings

Supply chain risk

Package contains fragments of external URLs or IP addresses, which the package may be accessing at runtime.

Improved metrics

Total package byte prevSize

882497

0.7%

Lines of code

16317

0.55%

No dependency changes