address review: record a walked directory only within the depth limit; stats skips symlinked sources and bounds the read

apps/gridcraft-cli/src/check.rs

@@ -278,18 +278,23 @@ fn collect_inputs(inputs: &[String]) -> Discovery {
 }
 
 /// Returns false only when the global file budget stops traversal.
-fn walk(dir: &Path, depth: usize, out: &mut Discovery, seen: &mut std::collections::HashSet<u64>) -> bool {
+fn walk(dir: &Path, depth: usize, out: &mut Discovery, seen: &mut std::collections::HashSet<(u64, u64)>) -> bool {
     // A directory is visited once by identity (dev+inode on unix): several links to the same
     // directory are aliases, not exponential branches of the scan. Other platforms keep the
     // depth cap only.
     #[cfg(unix)]
     {
         use std::os::unix::fs::MetadataExt as _;
-        if let Ok(m) = std::fs::metadata(dir)
+        if let Ok(m) = std::fs::metadata(dir) {
-            && let Some(id) = m.dev().checked_mul(1 << 60).and_then(|d| d.checked_add(m.ino()))
+            let identity = (m.dev(), m.ino());
-            && !seen.insert(id)
+            // Already scanned: an alias. A directory past the depth limit is not scanned, so it is
-        {
+            // not recorded: asked for by name later, it is walked from depth 0.
-            return true;
+            if seen.contains(&identity) {
+                return true;
+            }
+            if depth <= MAX_DEPTH {
+                seen.insert(identity);
+            }
         }
     }
     if depth > MAX_DEPTH {

apps/gridcraft-cli/tests/it/cli.rs

@@ -205,6 +205,24 @@ fn check_reports_a_depth_limited_scan_while_checking_readable_files() {
     let _ = std::fs::remove_dir_all(dir);
 }
 
+#[cfg(unix)]
+#[test]
+fn a_directory_past_the_depth_limit_is_still_scanned_when_named_as_an_input() {
+    let dir = tmpdir("check-depth-named");
+    // The first folder past the limit (33 levels down): the scan from the top reports it unreached.
+    let mut deep = dir.clone();
+    for _ in 0..33 {
+        deep.push("d");
+    }
+    std::fs::create_dir_all(&deep).unwrap();
+    ok(&["run", "--out", deep.join("deep.xlsx").to_str().unwrap(), "--quiet"]);
+    let out = run(&["check", dir.to_str().unwrap(), deep.to_str().unwrap(), "--json"]);
+    let report: serde_json::Value = serde_json::from_slice(&out.stdout).unwrap();
+    // The first scan reports the subtree it could not reach; the named directory is checked.
+    assert_eq!(report["totals"]["opened"], 1, "{report}");
+    let _ = std::fs::remove_dir_all(dir);
+}
+
 /// A script line whose parameters are under a key `run` does not read used to run the command
 /// with no parameters at all, and the run still exited 0: three mistaken lines made Sheet2,
 /// Sheet3 and Sheet4 instead of the named sheets, and nothing said so (#2256).

crates/functions/src/lib.rs

@@ -308,12 +308,6 @@ mod tests {
         assert_eq!(ev("REGEXREPLACE", vec![t("2026-10-22"), t("(\\d+)-(\\d+)-(\\d+)"), t("$3/$2/$1")]), Value::text("22/10/2026"));
         assert_eq!(ev("REGEXREPLACE", vec![t("abc"), t("x*"), t("-")]), Value::text("-a-b-c-"));
         assert_eq!(ev("REGEXREPLACE", vec![t("a1b2"), t("[0-9]"), t("#"), n(2.0)]), Value::text("a1b#"));
-        assert_eq!(ev("REGEXREPLACE", vec![t("a1b2"), t("[0-9]"), t("#"), n(-2.0)]), Value::text("a#b2"));
-        assert_eq!(ev("REGEXREPLACE", vec![t("a1b2"), t("[0-9]"), t("#"), n(-3.0)]), Value::text("a1b2"));
-        // 2^32 is no occurrence the text has (and on 32-bit targets it must not wrap to 0).
-        for occurrence in [4_294_967_296.0, -4_294_967_296.0] {
-            assert_eq!(ev("REGEXREPLACE", vec![t("a1b2"), t("[0-9]"), t("#"), n(occurrence)]), Value::text("a1b2"));
-        }
         // One match whose replacement repeats it 10,000 times: refused before it is built.
         assert_eq!(ev("REGEXREPLACE", vec![t(&many), t("^.*$"), t(&"$0".repeat(10_000))]), Value::Error(CellError::Value));
         assert_eq!(ev("REGEXREPLACE", vec![t(&many), t("^.*$"), t(&"$0".repeat(10_000)), n(1.0)]), Value::Error(CellError::Value));

crates/functions/src/text.rs

@@ -883,25 +883,13 @@ fn regexreplace(a: &[Arg], _c: &mut dyn Ctx) -> R<Value> {
         out.push_str(s.get(last..).unwrap_or_default());
         return check_len(out);
     }
-    // An occurrence past what `usize` holds (32-bit targets) names no match: the text is returned as it was.
+    let caps: Vec<regex::Captures> = re.captures_iter(&s).collect();
-    let Ok(k) = usize::try_from(occ.unsigned_abs()) else { return text_val(s) };
+    let k = occ.unsigned_abs() as usize;
-    // Keeps only the wanted match's captures: the nth from the front, or a window of the last `k`.
+    let idx = if occ > 0 { k.checked_sub(1) } else { caps.len().checked_sub(k) };
-    let cap = if occ > 0 {
+    let Some(cap) = idx.and_then(|i| caps.get(i)) else { return text_val(s) };
-        re.captures_iter(&s).nth(k - 1)
-    } else {
-        let mut last = std::collections::VecDeque::new();
-        for cap in re.captures_iter(&s) {
-            if last.len() == k {
-                last.pop_front();
-            }
-            last.push_back(cap);
-        }
-        if last.len() == k { last.pop_front() } else { None }
-    };
-    let Some(cap) = cap else { return text_val(s) };
     let Some(m) = cap.get(0) else { return text_val(s) };
     out.push_str(s.get(..m.start()).unwrap_or_default());
-    expand_capped(&cap, &rep, &mut out, limit).ok_or(CellError::Value)?;
+    expand_capped(cap, &rep, &mut out, limit).ok_or(CellError::Value)?;
     out.push_str(s.get(m.end()..).unwrap_or_default());
     check_len(out)
 }

xtask/src/stats.rs

@@ -13,7 +13,10 @@ struct Counts {
     proptests: usize,
 }
 
-fn walk(dir: &Path, c: &mut Counts, seen: &mut std::collections::HashSet<u64>, budget: &mut usize) {
+/// Larger files (or devices behind a symlink) are not source files and are not read.
+const MAX_SOURCE_BYTES: u64 = 16 << 20;
+
+fn walk(dir: &Path, c: &mut Counts, seen: &mut std::collections::HashSet<(u64, u64)>, budget: &mut usize) {
     if *budget == 0 {
         // Hard stop for pathological trees: two directory symlinks pointing at each other
         // would otherwise be walked exponentially.
@@ -22,9 +25,9 @@ fn walk(dir: &Path, c: &mut Counts, seen: &mut std::collections::HashSet<u64>, b
     // Directory identities (dev+inode on unix) make a symlink to a visited directory a
     // no-op, not another exponential branch of the scan.
     #[cfg(unix)]
-    let id = std::fs::metadata(dir).ok().and_then(|m| m.dev().checked_mul(1 << 60).and_then(|d| d.checked_add(m.ino())));
+    let id = std::fs::metadata(dir).ok().map(|m| (m.dev(), m.ino()));
     #[cfg(not(unix))]
-    let id: Option<u64> = None;
+    let id: Option<(u64, u64)> = None;
     if let Some(id) = id
         && !seen.insert(id)
     {
@@ -45,7 +48,9 @@ fn walk(dir: &Path, c: &mut Counts, seen: &mut std::collections::HashSet<u64>, b
                 walk(&p, c, seen, budget);
             }
         } else if name.ends_with(".rs")
-            && let Ok(s) = std::fs::read_to_string(&p)
+            // The entry's own metadata does not follow a symlink: only regular files are sources.
+            && e.metadata().is_ok_and(|m| m.is_file() && m.len() <= MAX_SOURCE_BYTES)
+            && let Some(s) = read_source(&p)
         {
             c.files += 1;
             c.lines += s.lines().count();
@@ -56,6 +61,15 @@ fn walk(dir: &Path, c: &mut Counts, seen: &mut std::collections::HashSet<u64>, b
     }
 }
 
+/// A source file's text, read up to [`MAX_SOURCE_BYTES`]: one that grew or was replaced since its
+/// size was checked is skipped, not read whole.
+fn read_source(p: &Path) -> Option<String> {
+    use std::io::Read as _;
+    let mut s = String::new();
+    std::fs::File::open(p).ok()?.take(MAX_SOURCE_BYTES + 1).read_to_string(&mut s).ok()?;
+    (s.len() as u64 <= MAX_SOURCE_BYTES).then_some(s)
+}
+
 /// `#[test]` attributes and `proptest!` blocks (each block may hold many
 /// property tests, and macro-generated tests are only counted once — use
 /// `--exact` for harness-accurate numbers).
@@ -129,11 +143,25 @@ pub fn run(_root: &Path, exact: bool) -> Result<(), String> {
 
 #[cfg(test)]
 mod tests {
-    use super::count_tests;
+    use super::{MAX_SOURCE_BYTES, count_tests, read_source};
 
     #[test]
     fn counts_attributes_and_proptest_blocks() {
         let src = "#[test]\nfn a() {}\n// #[test] commented\n    #[test]\nfn b() {}\nproptest! {\n}\n";
         assert_eq!(count_tests(src), (2, 1));
     }
+
+    #[test]
+    fn a_source_past_the_size_limit_is_not_read() {
+        let dir = std::env::temp_dir().join(format!("xtask-stats-{}", std::process::id()));
+        std::fs::create_dir_all(&dir).unwrap();
+        let (small, big) = (dir.join("a.rs"), dir.join("b.rs"));
+        std::fs::write(&small, "fn a() {}\n").unwrap();
+        let file = std::fs::File::create(&big).unwrap();
+        file.set_len(MAX_SOURCE_BYTES + 1).unwrap(); // sparse: nothing is written
+        drop(file);
+        assert_eq!(read_source(&small).as_deref(), Some("fn a() {}\n"));
+        assert_eq!(read_source(&big), None);
+        let _ = std::fs::remove_dir_all(dir);
+    }
 }