Ogg, xtask: check packet ranges against the source first; stats skips directory links

crates/ogg/src/lib.rs

@@ -415,6 +415,11 @@ impl CaptureScanner {
 }
 
 fn read_parts(src: &(impl ByteSource + ?Sized), parts: &[(u64, u32)]) -> Result<Vec<u8>> {
+    // Ranges come from a scan of some source; check them against this one before sizing a buffer.
+    let len = src.len();
+    if parts.iter().any(|&(o, n)| o.checked_add(u64::from(n)).is_none_or(|end| end > len)) {
+        return Err(Error::Invalid("packet range lies outside the source".into()));
+    }
     let mut v = Vec::with_capacity(parts.iter().map(|p| p.1 as usize).sum());
     for &(o, n) in parts {
         let a = v.len();

crates/ogg/src/tests.rs

@@ -274,3 +274,13 @@ fn a_budget_too_small_to_search_is_spent() {
     let f = opus_file(6, 312, 0);
     assert_eq!(scanner.find(&f, 0, f.len() as u64).unwrap(), None);
 }
+
+#[test]
+fn read_packet_checks_its_ranges_against_the_source_before_allocating() {
+    let f = opus_file(10, 312, 500);
+    let o = open(&f).unwrap();
+    let s = o.stream_of(Codec::Opus).unwrap();
+    // The last packet lies past the end of a source cut short: an error, not a read of missing bytes.
+    let short = &f[..f.len() / 2];
+    assert!(matches!(o.read_packet(&short, s, 9), Err(Error::Invalid(_))));
+}

xtask/src/assets.rs

@@ -61,23 +61,16 @@ pub fn missing(files: &[String], assets_md: &str) -> Vec<String> {
     files.iter().filter(|f| is_asset(f) && !patterns.iter().any(|p| glob(p, f))).cloned().collect()
 }
 
-/// `*` matches any run of characters except `/`; everything else literally. Matched with a
-/// table over the pattern, so a pattern with many `*` costs pattern length times name length
-/// rather than exploring every way to split the name.
+/// `*` matches any run of characters except `/`; everything else literally.
 fn glob(pat: &str, s: &str) -> bool {
-    let t: Vec<char> = s.chars().collect();
-    // `row[k]`: the pattern so far matches the first k characters of the name.
-    let mut row = vec![false; t.len() + 1];
-    row[0] = true;
-    for pc in pat.chars() {
-        let mut next = vec![false; t.len() + 1];
-        next[0] = pc == '*' && row[0];
-        for k in 1..=t.len() {
-            next[k] = if pc == '*' { row[k] || (next[k - 1] && t[k - 1] != '/') } else { row[k - 1] && t[k - 1] == pc };
+    match pat.split_once('*') {
+        None => pat == s,
+        Some((head, tail)) => {
+            let Some(rest) = s.strip_prefix(head) else { return false };
+            let ends = rest.char_indices().map(|(i, _)| i).chain(std::iter::once(rest.len()));
+            ends.take_while(|&i| !rest[..i].contains('/')).any(|i| glob(tail, &rest[i..]))
         }
-        row = next;
     }
-    row[t.len()]
 }
 
 /// Splits `git ls-files -z` output into real file names (no display quoting of Unicode or special characters).
@@ -127,15 +120,6 @@ mod tests {
         assert_eq!(missing(&["assets/é.png".into()], "`assets/*.svg`"), vec!["assets/é.png".to_string()]);
     }
 
-    #[test]
-    fn many_wildcards_on_a_failing_name_finish_at_once() {
-        let pat = format!("{}b", "a*".repeat(40));
-        assert!(!glob(&pat, &"a".repeat(60)));
-        assert!(glob(&pat, &format!("{}b", "a".repeat(60))));
-        assert!(!glob("a*b", "a/b"));
-        assert!(glob("a*b", "ab"));
-    }
-
     #[test]
     fn ls_files_output_is_nul_separated_and_unquoted() {
         let names = parse_ls_files("assets/zażółć.png\0docs/images/a b\nc.png\0".as_bytes());

xtask/src/stats.rs

@@ -18,7 +18,13 @@ fn walk(dir: &Path, c: &mut Counts) {
         let p = e.path();
         let name = e.file_name();
         let name = name.to_string_lossy();
-        if p.is_dir() {
+        // `file_type` does not follow links, so a directory link is not walked (and aliases are
+        // not counted twice).
+        let Ok(kind) = e.file_type() else { continue };
+        if kind.is_symlink() {
+            continue;
+        }
+        if kind.is_dir() {
             // Nested fuzz workspaces and build output are not part of the crate.
             if name != "target" && name != "fuzz" && !name.starts_with('.') {
                 walk(&p, c);