frostwork 0.2.1__tar.gz → 0.2.2__tar.gz

{frostwork-0.2.1 → frostwork-0.2.2}/CHANGELOG.md

@@ -3,6 +3,13 @@
 All notable user-facing changes will be recorded here. Frostwork will follow semantic versioning after its
 first public release.
 
+## 0.2.2 (2026-11-05)
+
+### Performance
+
+- `navigation()` no longer takes quadratic time on a document with many unclosed `<meta` tags, e.g.
+  `"<meta " * 80000`, which took over 30 seconds in 0.2.1.
+
 ## 0.2.1 (2026-11-05)
 
 ### Correctness

{frostwork-0.2.1 → frostwork-0.2.2}/Cargo.lock

@@ -50,7 +50,7 @@ dependencies = [
 
 [[package]]
 name = "frostwork"
-version = "0.2.1"
+version = "0.2.2"
 dependencies = [
  "chardetng",
  "encoding_rs",

{frostwork-0.2.1 → frostwork-0.2.2}/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "frostwork"
-version = "0.2.1"
+version = "0.2.2"
 edition = "2021"
 description = "Treeless, one-pass HTML extraction — CSS/XPath selectors matched on a streaming scan, no DOM."
 license = "Apache-2.0"

{frostwork-0.2.1 → frostwork-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: frostwork
-Version: 0.2.1
+Version: 0.2.2
 Classifier: Development Status :: 4 - Beta
 Classifier: Programming Language :: Rust
 Classifier: Programming Language :: Python :: 3 :: Only

{frostwork-0.2.1 → frostwork-0.2.2}/pyproject.toml

@@ -5,7 +5,7 @@ build-backend = "maturin"
 
 [project]
 name = "frostwork"
-version = "0.2.1"
+version = "0.2.2"
 description = "Treeless, one-pass HTML extraction — CSS/XPath selectors matched on a streaming scan, no DOM."
 readme = "README.md"
 license = "Apache-2.0"
@@ -89,7 +89,7 @@ module-name = "frostwork._frostwork"
 python-source = "python"
 
 [tool.bumpversion]
-current_version = "0.2.1"
+current_version = "0.2.2"
 commit = true
 tag = true
 tag_name = "{new_version}"

{frostwork-0.2.1 → frostwork-0.2.2}/src/lib.rs

@@ -750,24 +750,24 @@ impl Plan {
 
 /// Whether `bytes` can hold a `<base>`, when `base`, or a `<meta>` with an `http-equiv` attribute,
 /// which every meta refresh needs, when `refresh`. Tags are matched ASCII-case-insensitively wherever
-/// `<name` is followed by the end of a tag name, a superset of the start tags named `name`.
+/// `<name` is followed by the end of a tag name, a superset of the start tags named `name`, and
+/// `http-equiv` anywhere after the first `<meta`, a superset of its attributes. Linear in `bytes`.
 fn may_change(bytes: &[u8], base: bool, refresh: bool) -> bool {
-    let mut attrs = Vec::new();
-    memchr::memchr_iter(b'<', bytes).any(|i| {
-        let tag_end = |name: &[u8]| {
-            let end = i + 1 + name.len();
-            (bytes.get(i + 1..end).is_some_and(|candidate| candidate.eq_ignore_ascii_case(name))
-                && bytes.get(end).is_some_and(|&c| !tokenizer::is_name_char(c)))
-            .then_some(end)
-        };
-        (base && tag_end(b"base").is_some())
-            || (refresh
-                && tag_end(b"meta").is_some_and(|end| {
-                    attrs.clear();
-                    tokenizer::scan_attrs(bytes, end, &mut attrs);
-                    attrs.iter().any(|(name, _)| name.eq_ignore_ascii_case(b"http-equiv"))
-                }))
-    })
+    let tag_at = |i: usize, name: &[u8]| {
+        let end = i + 1 + name.len();
+        bytes.get(i + 1..end).is_some_and(|candidate| candidate.eq_ignore_ascii_case(name))
+            && bytes.get(end).is_some_and(|&c| !tokenizer::is_name_char(c))
+    };
+    let tags = || memchr::memchr_iter(b'<', bytes);
+    (base && tags().any(|i| tag_at(i, b"base")))
+        || (refresh
+            && tags().find(|&i| tag_at(i, b"meta")).is_some_and(|i| {
+                // `q` is the rarest byte of `http-equiv` in markup, so few candidates reach the comparison.
+                let rest = &bytes[i..];
+                memchr::memchr2_iter(b'q', b'Q', rest).any(|q| {
+                    q >= 6 && rest.get(q - 6..q + 4).is_some_and(|w| w.eq_ignore_ascii_case(b"http-equiv"))
+                })
+            }))
 }
 
 /// The base URL and meta refresh of `html`, as a browser reads them; see

{frostwork-0.2.1 → frostwork-0.2.2}/src/matcher/navigation.rs

@@ -702,6 +702,7 @@ mod tests {
         let utf16: Vec<u8> = "\u{feff}<meta http-equiv=refresh content=3>".encode_utf16().flat_map(u16::to_le_bytes).collect();
         assert_eq!(refresh(&utf16), to(3, DOC));
         assert_eq!(refresh(br#"<meta a=">" http-equiv=refresh content=5>"#), to(5, DOC));
+        assert_eq!(refresh(b"<script>'<meta a=\"</script><meta http-equiv=refresh content=6>\"'</script>"), to(6, DOC));
         assert_eq!(refresh(b"<meta name=refresh content=4><p http-equiv=refresh>http-equiv</p>"), None);
         let base = |html: &[u8]| nav(html, None, true).base_url;
         assert_eq!(base(b"<BASE\thref=/a/>"), Some("http://h/a/".into()));
@@ -709,6 +710,16 @@ mod tests {
         assert_eq!(base(b"<basefont href=/c/>"), Some(DOC.into()));
     }
 
+    #[test]
+    fn many_unclosed_meta_tags_take_linear_time() {
+        let evil = "<meta ".repeat(80000);
+        let start = std::time::Instant::now();
+        assert_eq!(nav(evil.as_bytes(), None, true).meta_refresh, None);
+        let good = evil + r#"<meta http-equiv="refresh" content="3;url=/next">"#;
+        assert_eq!(nav(good.as_bytes(), None, true).meta_refresh, Some((3, "http://h/next".into())));
+        assert!(start.elapsed().as_secs() < 2, "{:?}", start.elapsed());
+    }
+
     #[test]
     fn delay_is_whole_seconds() {
         let delay = |content: &str| nav(refresh(content).as_bytes(), None, true).meta_refresh.map(|(d, _)| d);

{frostwork-0.2.1 → frostwork-0.2.2}/src/tokenizer.rs

@@ -477,7 +477,7 @@ fn handle_markup<'a, S: TokenSink<'a>>(
 /// After `</img\nsrc="http:>`, whose value is unterminated and so runs on to the next quote 300 bytes
 /// later, that reading keeps an `</a>`, two `</div>`s and a whole `<div id='HTML3'>` start tag.
 /// Reporting an element the document does not contain is the FALSE-POSITIVE direction.
-pub(crate) fn scan_attrs<'a>(
+fn scan_attrs<'a>(
     b: &'a [u8],
     mut i: usize,
     attr_buf: &mut Vec<(&'a [u8], Option<&'a [u8]>)>,