Date: Fri, 28 Aug 2026 11:14:25 -0300
Subject: [PATCH 04/29] test: cover HTML detection review regressions
---
tests/html.rs | 36 ++++++++++++++++++++++++++++++++++++
1 file changed, 36 insertions(+)
diff --git a/tests/html.rs b/tests/html.rs
index 4b02c628..44b88e58 100644
--- a/tests/html.rs
+++ b/tests/html.rs
@@ -12,6 +12,42 @@ fn html_doctype_is_detected_from_content() {
assert_eq!(Format::from_bytes(html), Some(Format::Html));
}
+#[test]
+fn html_doctype_allows_html5_ascii_whitespace() {
+ assert_eq!(Format::from_bytes(b""), Some(Format::Html));
+ assert_eq!(Format::from_bytes(b""), Some(Format::Html));
+}
+
+#[test]
+fn html_prefix_wins_over_embedded_pdf_marker() {
+ let html = b"%PDF-1.7 is text here";
+ assert_eq!(Format::from_bytes(html), Some(Format::Html));
+}
+
+#[test]
+fn utf16_html_is_detected_from_content() {
+ let source = "hello";
+
+ let mut le = vec![0xFF, 0xFE];
+ for unit in source.encode_utf16() {
+ le.extend_from_slice(&unit.to_le_bytes());
+ }
+ assert_eq!(Format::from_bytes(&le), Some(Format::Html));
+
+ let mut be = vec![0xFE, 0xFF];
+ for unit in source.encode_utf16() {
+ be.extend_from_slice(&unit.to_be_bytes());
+ }
+ assert_eq!(Format::from_bytes(&be), Some(Format::Html));
+}
+
+#[test]
+fn unrelated_charset_attribute_does_not_change_decoding() {
+ let html = "café
".as_bytes();
+ let markdown = to_markdown_bytes(html, None).unwrap();
+ assert_eq!(markdown, "café\n");
+}
+
#[test]
fn malformed_html5_is_repaired_before_conversion() {
let html = br#"Hello first
second "#;
From 50b9629952fb718d7716fa38988c6f17bfde2dc3 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Fri, 28 Aug 2026 18:14:35 +0000
Subject: [PATCH 05/29] fix: address HTML review regressions
---
Cargo.lock | 1 +
Cargo.toml | 3 +-
src/formats/detect.rs | 27 +++--
src/formats/html.rs | 269 +++++++++++++++++++++++-------------------
tests/html.rs | 47 +++++++-
5 files changed, 214 insertions(+), 133 deletions(-)
diff --git a/Cargo.lock b/Cargo.lock
index 24767b2c..03039ba6 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -95,6 +95,7 @@ dependencies = [
"csv",
"encoding_rs",
"flate2",
+ "html5ever",
"insta",
"log",
"pdf-inspector",
diff --git a/Cargo.toml b/Cargo.toml
index b8b1f3f2..e24a9208 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -26,6 +26,7 @@ cfb = "0.14.0"
csv = "1.4.0"
flate2 = "1"
encoding_rs = "0.8.35"
+html5ever = "0.39.0"
log = "0.4"
pdf-inspector = "1.14.2"
quick-xml = "0.41.0"
@@ -34,4 +35,4 @@ zip = { version = "8.6.0", default-features = false, features = ["deflate"] }
[profile.release]
lto = "thin"
-strip = "symbols"
+strip = "symbols"
\ No newline at end of file
diff --git a/src/formats/detect.rs b/src/formats/detect.rs
index 4f647492..04bc849b 100644
--- a/src/formats/detect.rs
+++ b/src/formats/detect.rs
@@ -64,18 +64,21 @@ fn looks_like_html(bytes: &[u8]) -> bool {
}
fn looks_like_utf16_html(bytes: &[u8], little_endian: bool) -> bool {
- let units: Vec = bytes
- .chunks_exact(2)
- .take(256)
- .map(|pair| {
- if little_endian {
- u16::from_le_bytes([pair[0], pair[1]])
- } else {
- u16::from_be_bytes([pair[0], pair[1]])
- }
- })
- .collect();
- let decoded = String::from_utf16_lossy(&units);
+ let mut units = bytes.chunks_exact(2).map(|pair| {
+ if little_endian {
+ u16::from_le_bytes([pair[0], pair[1]])
+ } else {
+ u16::from_be_bytes([pair[0], pair[1]])
+ }
+ });
+ let mut prefix = Vec::with_capacity(64);
+ if let Some(first) =
+ units.find(|unit| !matches!(*unit, 0x0009 | 0x000A | 0x000C | 0x000D | 0x0020))
+ {
+ prefix.push(first);
+ prefix.extend(units.take(63));
+ }
+ let decoded = String::from_utf16_lossy(&prefix);
looks_like_ascii_html(decoded.as_bytes())
}
diff --git a/src/formats/html.rs b/src/formats/html.rs
index 58bc3429..721d3d34 100644
--- a/src/formats/html.rs
+++ b/src/formats/html.rs
@@ -8,7 +8,14 @@ use crate::package::xml::{Attr, Element, Node};
use crate::shared::html::{HtmlCtx, Stylesheet};
use crate::shared::uri::is_absolute_uri;
use encoding_rs::{Encoding, UTF_16BE, UTF_16LE, WINDOWS_1252};
+use html5ever::LocalName;
+use html5ever::tendril::StrTendril;
+use html5ever::tokenizer::states::{Rawtext, Rcdata, ScriptData};
+use html5ever::tokenizer::{
+ BufferQueue, EndTag, StartTag, TagToken, Token, TokenSink, TokenSinkResult, Tokenizer,
+};
use scraper::{ElementRef, Html, Node as HtmlNode};
+use std::cell::{Cell, RefCell};
use std::rc::Rc;
/// Parse a standalone HTML document into anydoc's document model.
@@ -29,6 +36,8 @@ pub fn parse(bytes: &[u8]) -> Result {
}
let text = decode_html(bytes);
+ preflight_html_complexity(&text)?;
+
let parsed = Html::parse_document(&text);
let root = parsed.root_element();
@@ -70,163 +79,180 @@ fn decode_html(bytes: &[u8]) -> String {
fn sniff_meta_charset(bytes: &[u8]) -> Option<&'static Encoding> {
const SNIFF_BYTES: usize = 1024;
- let mut prefix = bytes[..bytes.len().min(SNIFF_BYTES)].to_vec();
- prefix.make_ascii_lowercase();
-
- let mut offset = 0usize;
- while let Some(found) = find_bytes(&prefix[offset..], b"'))
- {
- offset = start + 1;
- continue;
- }
-
- let Some(end_rel) = find_tag_end(&prefix[attrs_start..]) else {
- break;
- };
- let end = attrs_start + end_rel;
- let attrs = &prefix[attrs_start..end];
+ let prefix = String::from_utf8_lossy(&bytes[..bytes.len().min(SNIFF_BYTES)]);
+ let parsed = Html::parse_document(prefix.as_ref());
+ let root = parsed.root_element();
- if let Some(label) = html_attr(attrs, b"charset")
- && let Some(encoding) = Encoding::for_label(label)
+ for meta in root.descendent_elements().filter(|element| element.value().name() == "meta") {
+ if let Some(label) = meta.value().attr("charset")
+ && let Some(encoding) = Encoding::for_label(label.trim().as_bytes())
{
return Some(encoding);
}
- let is_content_type = html_attr(attrs, b"http-equiv")
- .is_some_and(|value| value.eq_ignore_ascii_case(b"content-type"));
+ let is_content_type = meta
+ .value()
+ .attr("http-equiv")
+ .is_some_and(|value| value.trim().eq_ignore_ascii_case("content-type"));
if is_content_type
- && let Some(content) = html_attr(attrs, b"content")
+ && let Some(content) = meta.value().attr("content")
&& let Some(label) = content_type_charset(content)
- && let Some(encoding) = Encoding::for_label(label)
+ && let Some(encoding) = Encoding::for_label(label.as_bytes())
{
return Some(encoding);
}
-
- offset = end.saturating_add(1);
}
None
}
-fn find_tag_end(bytes: &[u8]) -> Option {
- let mut quote = None;
- for (index, &byte) in bytes.iter().enumerate() {
- match (quote, byte) {
- (Some(q), b) if b == q => quote = None,
- (Some(_), _) => {}
- (None, b'\'' | b'"') => quote = Some(byte),
- (None, b'>') => return Some(index),
- (None, _) => {}
+fn content_type_charset(content: &str) -> Option<&str> {
+ for parameter in content.split(';') {
+ let Some((name, value)) = parameter.split_once('=') else {
+ continue;
+ };
+ if name.trim().eq_ignore_ascii_case("charset") {
+ let label = value.trim().trim_matches(|c| c == '\'' || c == '"').trim();
+ if !label.is_empty() {
+ return Some(label);
+ }
}
}
None
}
-fn html_attr<'a>(attrs: &'a [u8], wanted: &[u8]) -> Option<&'a [u8]> {
- let mut pos = 0usize;
- while pos < attrs.len() {
- while attrs.get(pos).is_some_and(|b| b.is_ascii_whitespace() || *b == b'/') {
- pos += 1;
- }
- if pos >= attrs.len() {
- break;
- }
+#[derive(Default)]
+struct HtmlComplexitySink {
+ node_count: Cell,
+ open_elements: RefCell>,
+ node_limit_exceeded: Cell,
+ depth_limit_exceeded: Cell,
+}
- let name_start = pos;
- while attrs
- .get(pos)
- .is_some_and(|b| !b.is_ascii_whitespace() && !matches!(b, b'=' | b'/' | b'>'))
- {
- pos += 1;
+impl HtmlComplexitySink {
+ fn bump_node(&self) {
+ let count = self.node_count.get().saturating_add(1);
+ self.node_count.set(count);
+ if count > limits::MAX_XML_NODES {
+ self.node_limit_exceeded.set(true);
}
- if pos == name_start {
- pos += 1;
- continue;
- }
- let name = &attrs[name_start..pos];
+ }
- while attrs.get(pos).is_some_and(u8::is_ascii_whitespace) {
- pos += 1;
+ fn push_element(&self, name: &LocalName) {
+ let mut open = self.open_elements.borrow_mut();
+ open.push(name.clone());
+ if open.len() > limits::MAX_XML_DEPTH {
+ self.depth_limit_exceeded.set(true);
}
+ }
- let mut value = &attrs[pos..pos];
- if attrs.get(pos) == Some(&b'=') {
- pos += 1;
- while attrs.get(pos).is_some_and(u8::is_ascii_whitespace) {
- pos += 1;
- }
+ fn close_element(&self, name: &LocalName) {
+ let mut open = self.open_elements.borrow_mut();
+ if let Some(position) = open.iter().rposition(|candidate| candidate == name) {
+ open.truncate(position);
+ }
+ }
+}
- if let Some("e @ (b'\'' | b'"')) = attrs.get(pos) {
- pos += 1;
- let value_start = pos;
- while attrs.get(pos).is_some_and(|b| *b != quote) {
- pos += 1;
+impl TokenSink for HtmlComplexitySink {
+ type Handle = ();
+
+ fn process_token(&self, token: Token, _line_number: u64) -> TokenSinkResult {
+ match token {
+ TagToken(tag) => match tag.kind {
+ StartTag => {
+ self.bump_node();
+ let name = tag.name.as_ref();
+ if !tag.self_closing && !is_void_html_element(name) {
+ self.push_element(&tag.name);
+ }
+ match name {
+ "title" | "textarea" => TokenSinkResult::RawData(Rcdata),
+ "style" | "xmp" | "iframe" | "noembed" | "noframes" => {
+ TokenSinkResult::RawData(Rawtext)
+ }
+ "script" => TokenSinkResult::RawData(ScriptData),
+ "plaintext" => TokenSinkResult::Plaintext,
+ _ => TokenSinkResult::Continue,
+ }
}
- value = &attrs[value_start..pos];
- if attrs.get(pos) == Some("e) {
- pos += 1;
+ EndTag => {
+ self.close_element(&tag.name);
+ TokenSinkResult::Continue
}
- } else {
- let value_start = pos;
- while attrs
- .get(pos)
- .is_some_and(|b| !b.is_ascii_whitespace() && !matches!(b, b'/' | b'>'))
- {
- pos += 1;
+ },
+ Token::CharacterTokens(text) => {
+ if !text.is_empty() {
+ self.bump_node();
}
- value = &attrs[value_start..pos];
+ TokenSinkResult::Continue
}
- }
-
- if name.eq_ignore_ascii_case(wanted) {
- return Some(value);
+ Token::CommentToken(_) | Token::DoctypeToken(_) | Token::NullCharacterToken => {
+ self.bump_node();
+ TokenSinkResult::Continue
+ }
+ Token::EOFToken | Token::ParseError(_) => TokenSinkResult::Continue,
}
}
- None
}
-fn content_type_charset(content: &[u8]) -> Option<&[u8]> {
- let found = find_bytes(content, b"charset")?;
- let mut pos = found + b"charset".len();
- while content.get(pos).is_some_and(u8::is_ascii_whitespace) {
- pos += 1;
- }
- if content.get(pos) != Some(&b'=') {
- return None;
- }
- pos += 1;
- while content.get(pos).is_some_and(u8::is_ascii_whitespace) {
- pos += 1;
- }
+fn is_void_html_element(name: &str) -> bool {
+ matches!(
+ name,
+ "area"
+ | "base"
+ | "br"
+ | "col"
+ | "embed"
+ | "hr"
+ | "img"
+ | "input"
+ | "link"
+ | "meta"
+ | "param"
+ | "source"
+ | "track"
+ | "wbr"
+ )
+}
- let quote = match content.get(pos) {
- Some(b'\'') | Some(b'"') => {
- let quote = content[pos];
- pos += 1;
- Some(quote)
- }
- _ => None,
- };
- let start = pos;
- while let Some(&byte) = content.get(pos) {
- let stop = quote.map_or_else(
- || byte.is_ascii_whitespace() || matches!(byte, b';' | b'\'' | b'"'),
- |q| byte == q,
- );
- if stop {
- break;
+fn preflight_html_complexity(text: &str) -> Result<(), ConvertError> {
+ const CHUNK_BYTES: usize = 64 * 1024;
+ let tokenizer = Tokenizer::new(HtmlComplexitySink::default(), Default::default());
+ let input = BufferQueue::default();
+ let mut offset = 0usize;
+
+ while offset < text.len() {
+ let mut end = offset.saturating_add(CHUNK_BYTES).min(text.len());
+ while end > offset && !text.is_char_boundary(end) {
+ end -= 1;
}
- pos += 1;
+ input.push_back(StrTendril::from(&text[offset..end]));
+ let _ = tokenizer.feed(&input);
+ check_preflight_limits(&tokenizer.sink)?;
+ offset = end;
}
- (pos > start).then_some(&content[start..pos])
+
+ tokenizer.end();
+ check_preflight_limits(&tokenizer.sink)
}
-fn find_bytes(haystack: &[u8], needle: &[u8]) -> Option {
- haystack.windows(needle.len()).position(|window| window == needle)
+fn check_preflight_limits(sink: &HtmlComplexitySink) -> Result<(), ConvertError> {
+ if sink.node_limit_exceeded.get() {
+ return Err(ConvertError::ResourceLimit {
+ limit: "max_xml_nodes",
+ detail: format!("HTML token stream has more than {} nodes", limits::MAX_XML_NODES),
+ });
+ }
+ if sink.depth_limit_exceeded.get() {
+ return Err(ConvertError::ResourceLimit {
+ limit: "max_xml_depth",
+ detail: format!(
+ "HTML source nesting depth exceeds {} before DOM construction",
+ limits::MAX_XML_DEPTH
+ ),
+ });
+ }
+ Ok(())
}
fn adapt_element(
@@ -349,5 +375,10 @@ mod tests {
#[test]
fn charset_sniff_ignores_non_meta_text_and_attributes() {
assert_eq!(sniff_meta_charset(b"utf-8
"), None);
+ assert_eq!(sniff_meta_charset(b"utf-8
"), None);
+ assert_eq!(
+ sniff_meta_charset(b""),
+ None
+ );
}
}
diff --git a/tests/html.rs b/tests/html.rs
index 44b88e58..de27cc91 100644
--- a/tests/html.rs
+++ b/tests/html.rs
@@ -1,4 +1,4 @@
-use anydoc::{Format, to_markdown_bytes};
+use anydoc::{ConvertError, Format, to_markdown_bytes};
#[test]
fn html_extensions_are_named() {
@@ -41,6 +41,23 @@ fn utf16_html_is_detected_from_content() {
assert_eq!(Format::from_bytes(&be), Some(Format::Html));
}
+#[test]
+fn utf16_html_detection_allows_long_leading_whitespace() {
+ let source = format!("{}hello", " ".repeat(300));
+
+ let mut le = vec![0xFF, 0xFE];
+ for unit in source.encode_utf16() {
+ le.extend_from_slice(&unit.to_le_bytes());
+ }
+ assert_eq!(Format::from_bytes(&le), Some(Format::Html));
+
+ let mut be = vec![0xFE, 0xFF];
+ for unit in source.encode_utf16() {
+ be.extend_from_slice(&unit.to_be_bytes());
+ }
+ assert_eq!(Format::from_bytes(&be), Some(Format::Html));
+}
+
#[test]
fn unrelated_charset_attribute_does_not_change_decoding() {
let html = "café
".as_bytes();
@@ -48,6 +65,34 @@ fn unrelated_charset_attribute_does_not_change_decoding() {
assert_eq!(markdown, "café\n");
}
+#[test]
+fn meta_looking_text_in_comment_does_not_change_decoding() {
+ let html = "café
".as_bytes();
+ let markdown = to_markdown_bytes(html, None).unwrap();
+ assert_eq!(markdown, "café\n");
+}
+
+#[test]
+fn meta_looking_text_in_script_does_not_change_decoding() {
+ let html =
+ "café
"
+ .as_bytes();
+ let markdown = to_markdown_bytes(html, None).unwrap();
+ assert_eq!(markdown, "café\n");
+}
+
+#[test]
+fn html_node_limit_covers_nodes_outside_body_before_dom_materialization() {
+ let mut html = String::from("");
+ for _ in 0..2_000_001 {
+ html.push_str("");
+ }
+ html.push_str("ok");
+
+ let error = to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap_err();
+ assert!(matches!(error, ConvertError::ResourceLimit { limit: "max_xml_nodes", .. }));
+}
+
#[test]
fn malformed_html5_is_repaired_before_conversion() {
let html = br#"Hello first
second "#;
From 9920d30ba6466b682fe1eb226239b1f5360edac9 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Fri, 28 Aug 2026 18:42:10 -0300
Subject: [PATCH 06/29] fix: address second HTML review regressions
---
src/formats/detect.rs | 9 +++----
src/formats/html.rs | 55 ++++++++++++++++++++++++++++++++--------
tests/html.rs | 58 ++++++++++++++++++++++++++++++++++++++++++-
3 files changed, 104 insertions(+), 18 deletions(-)
diff --git a/src/formats/detect.rs b/src/formats/detect.rs
index 04bc849b..77072f45 100644
--- a/src/formats/detect.rs
+++ b/src/formats/detect.rs
@@ -64,12 +64,9 @@ fn looks_like_html(bytes: &[u8]) -> bool {
}
fn looks_like_utf16_html(bytes: &[u8], little_endian: bool) -> bool {
- let mut units = bytes.chunks_exact(2).map(|pair| {
- if little_endian {
- u16::from_le_bytes([pair[0], pair[1]])
- } else {
- u16::from_be_bytes([pair[0], pair[1]])
- }
+ let (pairs, _) = bytes.as_chunks::<2>();
+ let mut units = pairs.iter().map(|pair| {
+ if little_endian { u16::from_le_bytes(*pair) } else { u16::from_be_bytes(*pair) }
});
let mut prefix = Vec::with_capacity(64);
if let Some(first) =
diff --git a/src/formats/html.rs b/src/formats/html.rs
index 721d3d34..bef02a07 100644
--- a/src/formats/html.rs
+++ b/src/formats/html.rs
@@ -78,7 +78,7 @@ fn decode_html(bytes: &[u8]) -> String {
}
fn sniff_meta_charset(bytes: &[u8]) -> Option<&'static Encoding> {
- const SNIFF_BYTES: usize = 1024;
+ const SNIFF_BYTES: usize = 64 * 1024;
let prefix = String::from_utf8_lossy(&bytes[..bytes.len().min(SNIFF_BYTES)]);
let parsed = Html::parse_document(prefix.as_ref());
let root = parsed.root_element();
@@ -106,18 +106,31 @@ fn sniff_meta_charset(bytes: &[u8]) -> Option<&'static Encoding> {
}
fn content_type_charset(content: &str) -> Option<&str> {
- for parameter in content.split(';') {
- let Some((name, value)) = parameter.split_once('=') else {
- continue;
- };
- if name.trim().eq_ignore_ascii_case("charset") {
- let label = value.trim().trim_matches(|c| c == '\'' || c == '"').trim();
- if !label.is_empty() {
- return Some(label);
+ let mut start = 0usize;
+ let mut quote = None;
+ for (index, ch) in content.char_indices() {
+ match (quote, ch) {
+ (Some(active), current) if current == active => quote = None,
+ (None, '\'' | '"') => quote = Some(ch),
+ (None, ';') => {
+ if let Some(label) = charset_parameter(&content[start..index]) {
+ return Some(label);
+ }
+ start = index + ch.len_utf8();
}
+ _ => {}
}
}
- None
+ charset_parameter(&content[start..])
+}
+
+fn charset_parameter(parameter: &str) -> Option<&str> {
+ let (name, value) = parameter.split_once('=')?;
+ if !name.trim().eq_ignore_ascii_case("charset") {
+ return None;
+ }
+ let label = value.trim().trim_matches(|c| c == '\'' || c == '"').trim();
+ (!label.is_empty()).then_some(label)
}
#[derive(Default)]
@@ -139,6 +152,7 @@ impl HtmlComplexitySink {
fn push_element(&self, name: &LocalName) {
let mut open = self.open_elements.borrow_mut();
+ close_implied_before_start(&mut open, name.as_ref());
open.push(name.clone());
if open.len() > limits::MAX_XML_DEPTH {
self.depth_limit_exceeded.set(true);
@@ -195,6 +209,24 @@ impl TokenSink for HtmlComplexitySink {
}
}
+fn close_implied_before_start(open: &mut Vec, name: &str) {
+ let implied = match name {
+ "li" => &["li"][..],
+ "p" => &["p"][..],
+ "dt" | "dd" => &["dt", "dd"][..],
+ "rt" | "rp" => &["rt", "rp"][..],
+ "option" => &["option"][..],
+ "optgroup" => &["option", "optgroup"][..],
+ "tr" => &["tr"][..],
+ "td" | "th" => &["td", "th"][..],
+ _ => return,
+ };
+ if let Some(position) = open.iter().rposition(|candidate| implied.contains(&candidate.as_ref()))
+ {
+ open.truncate(position);
+ }
+}
+
fn is_void_html_element(name: &str) -> bool {
matches!(
name,
@@ -345,7 +377,8 @@ impl HtmlCtx for StandaloneCtx {
if src.is_empty() {
return Ok(None);
}
- Ok(is_absolute_uri(src).then(|| ImageSource::External(src.to_owned())))
+ Ok((is_absolute_uri(src) || src.starts_with("//"))
+ .then(|| ImageSource::External(src.to_owned())))
}
fn anchor_id(&self, raw: &str) -> AnchorId {
diff --git a/tests/html.rs b/tests/html.rs
index de27cc91..f4ee48ce 100644
--- a/tests/html.rs
+++ b/tests/html.rs
@@ -1,4 +1,5 @@
-use anydoc::{ConvertError, Format, to_markdown_bytes};
+use anydoc::model::{Block, ImageSource, Inline};
+use anydoc::{ConvertError, Format, to_document, to_markdown_bytes};
#[test]
fn html_extensions_are_named() {
@@ -132,3 +133,58 @@ fn scripts_are_not_document_content() {
let markdown = to_markdown_bytes(html, None).unwrap();
assert_eq!(markdown, "before\n\nafter\n");
}
+
+#[test]
+fn quoted_mime_parameter_semicolon_does_not_fake_charset() {
+ let mut html = br#""#.to_vec();
+ html.push(0x80);
+ html.extend_from_slice(b"
");
+ let markdown = to_markdown_bytes(&html, Some(Format::Html)).unwrap();
+ assert_eq!(markdown, "€\n");
+}
+
+#[test]
+fn optional_li_end_tags_do_not_count_as_nested_depth() {
+ let mut html = String::from("");
+ for i in 0..300 {
+ html.push_str(&format!("item {i}"));
+ }
+ html.push_str(" ");
+ let markdown = to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap();
+ assert!(markdown.contains("item 299"));
+}
+
+#[test]
+fn optional_p_end_tags_do_not_count_as_nested_depth() {
+ let mut html = String::from("");
+ for i in 0..300 {
+ html.push_str(&format!("paragraph {i}"));
+ }
+ let markdown = to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap();
+ assert!(markdown.contains("paragraph 299"));
+}
+
+#[test]
+fn meta_charset_after_first_kib_is_still_honored() {
+ let mut html = b"
");
+ html.push(0xA3);
+ html.extend_from_slice(b"
");
+ let markdown = to_markdown_bytes(&html, Some(Format::Html)).unwrap();
+ assert_eq!(markdown, "Ł\n");
+}
+
+#[test]
+fn protocol_relative_image_is_preserved_as_external() {
+ let html = br#"
"#;
+ let document = to_document(html, Some(Format::Html)).unwrap();
+ match &document.blocks[0] {
+ Block::Paragraph(inlines) => assert!(matches!(
+ &inlines[0],
+ Inline::Image { source: ImageSource::External(url), .. }
+ if url == "//cdn.example.test/image.png"
+ )),
+ other => panic!("expected paragraph, got {other:?}"),
+ }
+}
From df61b31d7dad49715a3745fea5829f0c6a846df3 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Fri, 28 Aug 2026 20:27:30 -0300
Subject: [PATCH 07/29] fix: match HTML5 preflight depth semantics
---
src/formats/html.rs | 85 ++++++++++++++++++++++++++++++++++++++++++++-
tests/html.rs | 61 ++++++++++++++++++++++++++++++++
2 files changed, 145 insertions(+), 1 deletion(-)
diff --git a/src/formats/html.rs b/src/formats/html.rs
index bef02a07..b66c423b 100644
--- a/src/formats/html.rs
+++ b/src/formats/html.rs
@@ -176,7 +176,9 @@ impl TokenSink for HtmlComplexitySink {
StartTag => {
self.bump_node();
let name = tag.name.as_ref();
- if !tag.self_closing && !is_void_html_element(name) {
+ let honor_self_closing = tag.self_closing
+ && html5_self_closing_is_honored(&self.open_elements.borrow(), name);
+ if !is_void_html_element(name) && !honor_self_closing {
self.push_element(&tag.name);
}
match name {
@@ -210,6 +212,12 @@ impl TokenSink for HtmlComplexitySink {
}
fn close_implied_before_start(open: &mut Vec, name: &str) {
+ if is_heading_element(name)
+ && open.last().is_some_and(|candidate| is_heading_element(candidate.as_ref()))
+ {
+ open.pop();
+ }
+
let implied = match name {
"li" => &["li"][..],
"p" => &["p"][..],
@@ -227,6 +235,81 @@ fn close_implied_before_start(open: &mut Vec, name: &str) {
}
}
+fn is_heading_element(name: &str) -> bool {
+ matches!(name, "h1" | "h2" | "h3" | "h4" | "h5" | "h6")
+}
+
+fn html5_self_closing_is_honored(open: &[LocalName], name: &str) -> bool {
+ if matches!(name, "svg" | "math") {
+ return true;
+ }
+
+ let mut in_foreign_content = false;
+ for candidate in open.iter().rev() {
+ match candidate.as_ref() {
+ "foreignobject" | "desc" | "title" | "mi" | "mo" | "mn" | "ms" | "mtext"
+ | "annotation-xml" => return false,
+ "svg" | "math" => {
+ in_foreign_content = true;
+ break;
+ }
+ _ => {}
+ }
+ }
+
+ in_foreign_content && !is_foreign_content_html_breakout(name)
+}
+
+fn is_foreign_content_html_breakout(name: &str) -> bool {
+ matches!(
+ name,
+ "b" | "big"
+ | "blockquote"
+ | "body"
+ | "br"
+ | "center"
+ | "code"
+ | "dd"
+ | "div"
+ | "dl"
+ | "dt"
+ | "em"
+ | "embed"
+ | "font"
+ | "h1"
+ | "h2"
+ | "h3"
+ | "h4"
+ | "h5"
+ | "h6"
+ | "head"
+ | "hr"
+ | "i"
+ | "img"
+ | "li"
+ | "listing"
+ | "menu"
+ | "meta"
+ | "nobr"
+ | "ol"
+ | "p"
+ | "pre"
+ | "ruby"
+ | "s"
+ | "small"
+ | "span"
+ | "strike"
+ | "strong"
+ | "sub"
+ | "sup"
+ | "table"
+ | "tt"
+ | "u"
+ | "ul"
+ | "var"
+ )
+}
+
fn is_void_html_element(name: &str) -> bool {
matches!(
name,
diff --git a/tests/html.rs b/tests/html.rs
index f4ee48ce..b9b82421 100644
--- a/tests/html.rs
+++ b/tests/html.rs
@@ -188,3 +188,64 @@ fn protocol_relative_image_is_preserved_as_external() {
other => panic!("expected paragraph, got {other:?}"),
}
}
+
+fn assert_preflight_depth_limit(error: ConvertError) {
+ match error {
+ ConvertError::ResourceLimit { limit, detail } => {
+ assert_eq!(limit, "max_xml_depth");
+ assert!(
+ detail.contains("before DOM construction"),
+ "expected preflight depth rejection, got: {detail}"
+ );
+ }
+ other => panic!("expected max_xml_depth resource limit, got {other:?}"),
+ }
+}
+
+#[test]
+fn non_void_self_closing_html_tags_still_count_toward_preflight_depth() {
+ let mut html = String::from("");
+ for _ in 0..300 {
+ html.push_str("
");
+ }
+
+ assert_preflight_depth_limit(
+ to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap_err(),
+ );
+}
+
+#[test]
+fn successive_headings_are_implicitly_closed_before_preflight_depth_counting() {
+ let mut html = String::from("");
+ for i in 0..300 {
+ html.push_str(&format!("heading {i}"));
+ }
+
+ let markdown = to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap();
+ assert!(markdown.contains("# heading 299"));
+}
+
+#[test]
+fn foreign_self_closing_svg_elements_do_not_accumulate_html_depth() {
+ let mut html = String::from("");
+ for _ in 0..300 {
+ html.push_str(" ok
");
+
+ let markdown = to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap();
+ assert!(markdown.contains("ok"));
+}
+
+#[test]
+fn html_inside_svg_foreign_object_still_counts_self_closing_non_void_depth() {
+ let mut html = String::from("");
+ for _ in 0..300 {
+ html.push_str("
");
+ }
+ html.push_str("
Date: Fri, 28 Aug 2026 20:58:35 -0300
Subject: [PATCH 08/29] fix: address final HTML review findings
---
Cargo.toml | 2 +-
src/formats/detect.rs | 65 +++++++++++++++++++++++++++++++++++--------
tests/html.rs | 20 +++++++++++++
3 files changed, 75 insertions(+), 12 deletions(-)
diff --git a/Cargo.toml b/Cargo.toml
index e24a9208..0ed2ba91 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -35,4 +35,4 @@ zip = { version = "8.6.0", default-features = false, features = ["deflate"] }
[profile.release]
lto = "thin"
-strip = "symbols"
\ No newline at end of file
+strip = "symbols"
diff --git a/src/formats/detect.rs b/src/formats/detect.rs
index 77072f45..522e513b 100644
--- a/src/formats/detect.rs
+++ b/src/formats/detect.rs
@@ -65,18 +65,61 @@ fn looks_like_html(bytes: &[u8]) -> bool {
fn looks_like_utf16_html(bytes: &[u8], little_endian: bool) -> bool {
let (pairs, _) = bytes.as_chunks::<2>();
- let mut units = pairs.iter().map(|pair| {
- if little_endian { u16::from_le_bytes(*pair) } else { u16::from_be_bytes(*pair) }
- });
- let mut prefix = Vec::with_capacity(64);
- if let Some(first) =
- units.find(|unit| !matches!(*unit, 0x0009 | 0x000A | 0x000C | 0x000D | 0x0020))
- {
- prefix.push(first);
- prefix.extend(units.take(63));
+ let mut index = 0;
+ while pairs.get(index).is_some_and(|pair| {
+ utf16_ascii_unit(*pair, little_endian).is_some_and(|b| b.is_ascii_whitespace())
+ }) {
+ index += 1;
+ }
+
+ if utf16_html_prefix(pairs, index, little_endian, b" Option {
+ let unit = if little_endian { u16::from_le_bytes(pair) } else { u16::from_be_bytes(pair) };
+ (unit <= 0x7F).then_some(unit as u8)
+}
+
+fn utf16_prefix_eq_ignore_ascii_case(
+ pairs: &[[u8; 2]],
+ start: usize,
+ little_endian: bool,
+ prefix: &[u8],
+) -> bool {
+ let Some(slice) = pairs.get(start..start + prefix.len()) else {
+ return false;
+ };
+ slice.iter().zip(prefix).all(|(pair, expected)| {
+ utf16_ascii_unit(*pair, little_endian)
+ .is_some_and(|byte| byte.eq_ignore_ascii_case(expected))
+ })
+}
+
+fn utf16_html_prefix(pairs: &[[u8; 2]], start: usize, little_endian: bool, prefix: &[u8]) -> bool {
+ utf16_prefix_eq_ignore_ascii_case(pairs, start, little_endian, prefix)
+ && pairs.get(start + prefix.len()).is_none_or(|pair| {
+ utf16_ascii_unit(*pair, little_endian)
+ .is_some_and(|b| b.is_ascii_whitespace() || matches!(b, b'>' | b'/'))
+ })
}
fn looks_like_ascii_html(bytes: &[u8]) -> bool {
diff --git a/tests/html.rs b/tests/html.rs
index b9b82421..1718431b 100644
--- a/tests/html.rs
+++ b/tests/html.rs
@@ -59,6 +59,26 @@ fn utf16_html_detection_allows_long_leading_whitespace() {
assert_eq!(Format::from_bytes(&be), Some(Format::Html));
}
+#[test]
+fn utf16le_doctype_allows_long_whitespace_between_keyword_and_name() {
+ let source = format!("", " ".repeat(80));
+ let mut bytes = vec![0xFF, 0xFE];
+ for unit in source.encode_utf16() {
+ bytes.extend_from_slice(&unit.to_le_bytes());
+ }
+ assert_eq!(Format::from_bytes(&bytes), Some(Format::Html));
+}
+
+#[test]
+fn utf16be_doctype_allows_long_whitespace_between_keyword_and_name() {
+ let source = format!("", " ".repeat(80));
+ let mut bytes = vec![0xFE, 0xFF];
+ for unit in source.encode_utf16() {
+ bytes.extend_from_slice(&unit.to_be_bytes());
+ }
+ assert_eq!(Format::from_bytes(&bytes), Some(Format::Html));
+}
+
#[test]
fn unrelated_charset_attribute_does_not_change_decoding() {
let html = "café
".as_bytes();
From fbe5339f2eb23a64f84007bd75c873c70cc9b032 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Fri, 28 Aug 2026 22:26:42 -0300
Subject: [PATCH 09/29] test: consolidate HTML review regressions
---
tests/html.rs | 19 +++++++++++++++++++
1 file changed, 19 insertions(+)
diff --git a/tests/html.rs b/tests/html.rs
index 1718431b..bf489309 100644
--- a/tests/html.rs
+++ b/tests/html.rs
@@ -25,6 +25,12 @@ fn html_prefix_wins_over_embedded_pdf_marker() {
assert_eq!(Format::from_bytes(html), Some(Format::Html));
}
+#[test]
+fn pdf_header_before_html_marker_remains_pdf() {
+ let bytes = b" %PDF-1.7\nnot an HTML root";
+ assert_eq!(Format::from_bytes(bytes), Some(Format::Pdf));
+}
+
#[test]
fn utf16_html_is_detected_from_content() {
let source = "hello";
@@ -245,6 +251,19 @@ fn successive_headings_are_implicitly_closed_before_preflight_depth_counting() {
assert!(markdown.contains("# heading 299"));
}
+#[test]
+fn alternating_headings_are_implicitly_closed_before_preflight_depth_counting() {
+ const HEADINGS: [&str; 6] = ["h1", "h2", "h3", "h4", "h5", "h6"];
+ let mut html = String::from("");
+ for i in 0..300 {
+ let heading = HEADINGS[i % HEADINGS.len()];
+ html.push_str(&format!("<{heading}>heading {i}"));
+ }
+
+ let markdown = to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap();
+ assert!(markdown.contains("heading 299"));
+}
+
#[test]
fn foreign_self_closing_svg_elements_do_not_accumulate_html_depth() {
let mut html = String::from("");
From c21198bf90d9892e2c520bb66abe81775a2dcc9a Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 06:32:16 -0300
Subject: [PATCH 10/29] test: add HTML corpus integration regressions
---
tests/html_corpus.rs | 37 +++++++++++++++++++++++++++++++++++++
1 file changed, 37 insertions(+)
create mode 100644 tests/html_corpus.rs
diff --git a/tests/html_corpus.rs b/tests/html_corpus.rs
new file mode 100644
index 00000000..0225a107
--- /dev/null
+++ b/tests/html_corpus.rs
@@ -0,0 +1,37 @@
+mod common;
+
+use anydoc::{Format, to_markdown, to_markdown_bytes};
+use common::fixture_root;
+use std::fs;
+use std::path::PathBuf;
+
+fn html_fixture(name: &str) -> PathBuf {
+ fixture_root().join("html").join(name)
+}
+
+#[test]
+fn controlled_document_preserves_rich_text_links_and_unicode() {
+ let bytes = fs::read(html_fixture("controlled-document.html")).unwrap();
+ let markdown = to_markdown_bytes(&bytes, Some(Format::Html)).unwrap();
+ assert_eq!(
+ markdown,
+ "# Fixture Document\n\nPlain paragraph with **bold**, *italic*, and [example](https://example.com/page).\n\n## Unicode\n\ncafé Ł music 𝄞 family 👨👩👧.\n\n**Styled bold paragraph.**\n"
+ );
+}
+
+#[test]
+fn controlled_nested_lists_preserve_numbering_and_structure() {
+ let bytes = fs::read(html_fixture("controlled-lists.html")).unwrap();
+ let markdown = to_markdown_bytes(&bytes, Some(Format::Html)).unwrap();
+ assert_eq!(
+ markdown,
+ "## Lists\n\n3. Third\n\n4. Fourth\n\n - Nested A\n\n - Nested B\n"
+ );
+}
+
+#[test]
+fn controlled_spanned_table_matches_existing_xlsx_semantics() {
+ let html = to_markdown(html_fixture("controlled-merged-table.html")).unwrap();
+ let source = to_markdown(fixture_root().join("xlsx").join("handmade-merged.xlsx")).unwrap();
+ assert_eq!(html, source);
+}
From 0c4c81c0dbaf8c6ec4f75ef3c38d35281b062f72 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 06:34:19 -0300
Subject: [PATCH 11/29] test: add controlled HTML corpus fixtures
---
tests/fixtures/html/controlled-document.html | 18 ++++++++++++++++++
tests/fixtures/html/controlled-lists.html | 16 ++++++++++++++++
.../fixtures/html/controlled-merged-table.html | 11 +++++++++++
3 files changed, 45 insertions(+)
create mode 100644 tests/fixtures/html/controlled-document.html
create mode 100644 tests/fixtures/html/controlled-lists.html
create mode 100644 tests/fixtures/html/controlled-merged-table.html
diff --git a/tests/fixtures/html/controlled-document.html b/tests/fixtures/html/controlled-document.html
new file mode 100644
index 00000000..88fa4c65
--- /dev/null
+++ b/tests/fixtures/html/controlled-document.html
@@ -0,0 +1,18 @@
+
+
+
+ Fixture Document
+ Plain paragraph with bold , italic , and example .
+ Unicode
+ café Ł music 𝄞 family 👨👩👧.
+ Styled bold paragraph.
+ drop me
+
+
diff --git a/tests/fixtures/html/controlled-lists.html b/tests/fixtures/html/controlled-lists.html
new file mode 100644
index 00000000..8cd514e0
--- /dev/null
+++ b/tests/fixtures/html/controlled-lists.html
@@ -0,0 +1,16 @@
+
+
+Lists
+
+ Third
+ Fourth
+
+
+
+
+
diff --git a/tests/fixtures/html/controlled-merged-table.html b/tests/fixtures/html/controlled-merged-table.html
new file mode 100644
index 00000000..15ff1adb
--- /dev/null
+++ b/tests/fixtures/html/controlled-merged-table.html
@@ -0,0 +1,11 @@
+
+
+
+ Merged across padded tall b2 3.5 b3
+
+
From adaf98499893d6bdb13446b3262fdeffa9e32e46 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 06:37:22 -0300
Subject: [PATCH 12/29] test: add LibreOffice HTML corpus regressions
---
tests/html_corpus.rs | 64 +++++++++++++++++++++++++++++++++++++++++++-
1 file changed, 63 insertions(+), 1 deletion(-)
diff --git a/tests/html_corpus.rs b/tests/html_corpus.rs
index 0225a107..e148bdc1 100644
--- a/tests/html_corpus.rs
+++ b/tests/html_corpus.rs
@@ -9,6 +9,13 @@ fn html_fixture(name: &str) -> PathBuf {
fixture_root().join("html").join(name)
}
+fn assert_both_contain(source: &str, html: &str, needles: &[&str]) {
+ for needle in needles {
+ assert!(source.contains(needle), "source Markdown missing {needle:?}");
+ assert!(html.contains(needle), "HTML Markdown missing {needle:?}");
+ }
+}
+
#[test]
fn controlled_document_preserves_rich_text_links_and_unicode() {
let bytes = fs::read(html_fixture("controlled-document.html")).unwrap();
@@ -25,7 +32,7 @@ fn controlled_nested_lists_preserve_numbering_and_structure() {
let markdown = to_markdown_bytes(&bytes, Some(Format::Html)).unwrap();
assert_eq!(
markdown,
- "## Lists\n\n3. Third\n\n4. Fourth\n\n - Nested A\n\n - Nested B\n"
+ "## Lists\n\n3. Third\n\n4. Fourth\n\n - Nested A\n - Nested B\n"
);
}
@@ -35,3 +42,58 @@ fn controlled_spanned_table_matches_existing_xlsx_semantics() {
let source = to_markdown(fixture_root().join("xlsx").join("handmade-merged.xlsx")).unwrap();
assert_eq!(html, source);
}
+
+#[test]
+fn libreoffice_docx_text_preserves_core_document_semantics() {
+ let source = to_markdown(fixture_root().join("docx").join("text.docx")).unwrap();
+ let html = to_markdown(html_fixture("libreoffice-docx-text.html")).unwrap();
+
+ assert_both_contain(
+ &source,
+ &html,
+ &[
+ "# Fixture Document",
+ "Plain paragraph with **bold**, *italic*",
+ "## Lists",
+ "First numbered",
+ "## Table",
+ "Wide head",
+ "Music clef 𝄞",
+ "Persian with ZWNJ",
+ "Family emoji",
+ "[example](https://example.com/page)",
+ ],
+ );
+}
+
+#[test]
+fn libreoffice_docx_numbering_preserves_list_content_and_emphasis() {
+ let source =
+ to_markdown(fixture_root().join("docx").join("handmade-numbering.docx")).unwrap();
+ let html = to_markdown(html_fixture("libreoffice-docx-numbering.html")).unwrap();
+
+ assert_both_contain(
+ &source,
+ &html,
+ &[
+ "One-one",
+ "One-two",
+ "One-two-a roman",
+ "Deep bullet",
+ "Interruption paragraph.",
+ "One-four continues the count",
+ "Ten-start via override",
+ "Bold here, and style-false keeps it bold",
+ "direct on makes this italic",
+ ],
+ );
+ assert!(html.contains("1. One-one"));
+ assert!(html.contains("- Deep bullet"));
+}
+
+#[test]
+fn libreoffice_xlsx_merged_table_matches_source_markdown() {
+ let source = to_markdown(fixture_root().join("xlsx").join("handmade-merged.xlsx")).unwrap();
+ let html = to_markdown(html_fixture("libreoffice-xlsx-merged.html")).unwrap();
+ assert_eq!(html, source);
+}
From 529aaa2906c55b26d433257a41c60a4a28fee2c8 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 06:38:53 -0300
Subject: [PATCH 13/29] test: add LibreOffice HTML corpus fixtures
---
.../html/libreoffice-docx-numbering.html | 99 ++++++++++
.../fixtures/html/libreoffice-docx-text.html | 185 ++++++++++++++++++
.../html/libreoffice-xlsx-merged.html | 41 ++++
3 files changed, 325 insertions(+)
create mode 100644 tests/fixtures/html/libreoffice-docx-numbering.html
create mode 100644 tests/fixtures/html/libreoffice-docx-text.html
create mode 100644 tests/fixtures/html/libreoffice-xlsx-merged.html
diff --git a/tests/fixtures/html/libreoffice-docx-numbering.html b/tests/fixtures/html/libreoffice-docx-numbering.html
new file mode 100644
index 00000000..79242746
--- /dev/null
+++ b/tests/fixtures/html/libreoffice-docx-numbering.html
@@ -0,0 +1,99 @@
+
+
+
+ One-one
One-two
+
+ One-two-a
+ roman
One-two-b
+ roman
+
+
+ One-three
Interruption
+paragraph.
+
+ One-four
+ continues the count
+
+ Two-one
+ independent counter
+
+ Ten-start
+ via override
+
+ Letter-A via
+ numStyleLink
+
+ Level
+ replaced and start overridden together
+Suppressed numbering
+paragraph
+
+ Style-numbered
+ paragraph
+
+
+
+
+
+
+ pStyle-bound
+ level one
+
+
+
+
+
+pStyle-bound level
+two
+
+
+
+
+
+
+ pStyle-bound
+ level one again
+
+
+
+
+
+Inherited
+subheading
+Bold here, and
+style-false keeps it bold
+toggled
+off by a true in the character style
+double toggle
+cancels to plain
+direct
+off wins absolutely
+direct on makes
+this italic
+
+
\ No newline at end of file
diff --git a/tests/fixtures/html/libreoffice-docx-text.html b/tests/fixtures/html/libreoffice-docx-text.html
new file mode 100644
index 00000000..d86cd5c8
--- /dev/null
+++ b/tests/fixtures/html/libreoffice-docx-text.html
@@ -0,0 +1,185 @@
+
+
+
+ Fixture Document
+Plain paragraph with bold , italic , and struck
+runs.
+Style-bold
+paragraph with a NotBold-styled
+span inside.
+
+
+ Lists
+
+
+
+ First numbered
+ Second numbered
+
+
+ Alpha sub one
+ Alpha sub two
+
+
+ Roman sub sub
+
+
+ Third numbered
+
+Interrupting paragraph between lists.
+
+
+ Fourth, continuing the count
+
+
+ Roman starting at four
+
+
+ Roman five
+
+
+ Bullet one
+ Bullet two
+
+
+
+
+ Table
+
+
+ Wide head
+
+ End
+
+
+
+ Tall
+
+ B2
+
+ C2
+
+
+
+ B3
+
+ C3
+
+
+ Notes and special text
+
+
+Music clef 𝄞 appears before this footnote
+reference.
+
+An endnote follows herei
+
+Persian with ZWNJ: میخواهم
+
+Markdown specials: *stars* _under_ [bracket] `tick` #hash 1. dotted |
+pipe.
+
+
+ Links and anchors
+
+
+External link to example
+
+Relative link to a
+sibling file
+
+
+Jump to the bookmarked
+paragraph
+
+
+ Objects
+
+
+Inline image:
+
+
+Text box: after the box.
+ Inside the text box.
+
+
+
+ Quote and code
+
+
+Value below one millionth: 0.0000004 should survive.
+
+
+
+
\ No newline at end of file
diff --git a/tests/fixtures/html/libreoffice-xlsx-merged.html b/tests/fixtures/html/libreoffice-xlsx-merged.html
new file mode 100644
index 00000000..fcb0f498
--- /dev/null
+++ b/tests/fixtures/html/libreoffice-xlsx-merged.html
@@ -0,0 +1,41 @@
+
+
+
+
+
+
+
+ Merged across
+ padded
+
+
+ tall
+ b2
+ 3.5
+
+
+ b3
+
+
+
+
+
+
From 2bca8eeeed75730e40ed52bdb9e4aabb51c3ab01 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 06:45:08 -0300
Subject: [PATCH 14/29] test: cover structural children in nested list wrappers
---
tests/html.rs | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/tests/html.rs b/tests/html.rs
index bf489309..1099d129 100644
--- a/tests/html.rs
+++ b/tests/html.rs
@@ -288,3 +288,15 @@ fn html_inside_svg_foreign_object_still_counts_self_closing_non_void_depth() {
to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap_err(),
);
}
+
+#[test]
+fn nested_list_wrappers_preserve_structural_children() {
+ let html = br#"
+
+ Nested heading
Date: Sat, 29 Aug 2026 06:47:47 -0300
Subject: [PATCH 15/29] fix: preserve structural children in HTML lists
---
src/shared/html.rs | 106 ++++++++++++++++++++++++++++++++++-----------
1 file changed, 81 insertions(+), 25 deletions(-)
diff --git a/src/shared/html.rs b/src/shared/html.rs
index 3a7590b6..ccb4b18c 100644
--- a/src/shared/html.rs
+++ b/src/shared/html.rs
@@ -297,6 +297,25 @@ impl Builder<'_> {
Ok(b.finish())
}
+ /// Render one element in a fresh block context, preserving the element's
+ /// own block semantics. This is used for malformed-but-repairable list
+ /// markup where structural elements appear directly under `ul`/`ol`.
+ fn sub_element_blocks(
+ &mut self,
+ elem: &Element,
+ delta: StyleDelta,
+ ) -> Result, ConvertError> {
+ let mut b = Builder {
+ blocks: Vec::new(),
+ inlines: Vec::new(),
+ css: self.css,
+ ctx: self.ctx,
+ start_boundary: true,
+ };
+ b.walk_elem(elem, delta)?;
+ Ok(b.finish())
+ }
+
/// Element-level props: matching stylesheet rules and the inline `style`
/// attribute merged in one cascade order — normal rules, inline style,
/// `!important` rules, `!important` inline style.
@@ -539,19 +558,40 @@ impl Builder<'_> {
delta: StyleDelta,
) -> Result, ConvertError> {
let ordered = elem.local == "ol";
- let items: Vec<&Element> = elem.child_elems().filter(|e| e.local == "li").collect();
- if items.is_empty() {
- return Ok(Vec::new());
- }
+ let children: Vec<&Element> = elem.child_elems().collect();
+ let items: Vec<&Element> =
+ children.iter().copied().filter(|e| e.local == "li").collect();
+
if !ordered {
- let mut list_items = Vec::with_capacity(items.len());
- for li in &items {
- list_items
- .push(ListItem { blocks: self.sub_blocks(li, delta)?, marker_label: None });
+ let mut out = Vec::new();
+ let mut list_items = Vec::new();
+ for child in children {
+ if child.local == "li" {
+ list_items.push(ListItem {
+ blocks: self.sub_blocks(child, delta)?,
+ marker_label: None,
+ });
+ continue;
+ }
+ if !list_items.is_empty() {
+ out.push(Block::List(List {
+ marker: MarkerKind::Bullet,
+ start: 1,
+ items: std::mem::take(&mut list_items),
+ }));
+ }
+ out.extend(self.sub_element_blocks(child, delta)?);
+ }
+ if !list_items.is_empty() {
+ out.push(Block::List(List {
+ marker: MarkerKind::Bullet,
+ start: 1,
+ items: list_items,
+ }));
}
- let list = List { marker: MarkerKind::Bullet, start: 1, items: list_items };
- return Ok(vec![Block::List(list)]);
+ return Ok(out);
}
+
let marker = match elem.attr_any("type") {
Some("a") => MarkerKind::LowerAlpha,
Some("A") => MarkerKind::UpperAlpha,
@@ -575,24 +615,40 @@ impl Builder<'_> {
// step must not overflow.
next = if reversed { next.saturating_sub(1) } else { next.saturating_add(1) };
}
- // Zero/negative numbers are valid ordered-list values but cannot be
- // a `start` for the renderer's start+index numbering; such lists
- // carry every number as an explicit literal marker instead.
- if numbers.iter().any(|&n| n < 1) {
- let mut list_items = Vec::with_capacity(items.len());
- for (li, &n) in items.iter().zip(&numbers) {
- list_items.push(ListItem {
- blocks: self.sub_blocks(li, delta)?,
- marker_label: Some(format!("{n}.")),
- });
- }
- return Ok(vec![Block::List(List { marker, start: 1, items: list_items })]);
- }
+
+ // Zero/negative numbers are valid but cannot be represented by the
+ // renderer's positive `start` field, so those items carry explicit
+ // literal marker labels. Structural siblings still stay in order.
+ let explicit_markers = numbers.iter().any(|&n| n < 1);
let mut out: Vec = Vec::new();
let mut current: Option = None;
let mut last_number = 0i64;
- for (li, &number) in items.iter().zip(&numbers) {
- let item = ListItem { blocks: self.sub_blocks(li, delta)?, marker_label: None };
+ let mut item_index = 0usize;
+
+ for child in children {
+ if child.local != "li" {
+ if let Some(list) = current.take() {
+ out.push(Block::List(list));
+ }
+ out.extend(self.sub_element_blocks(child, delta)?);
+ continue;
+ }
+
+ let number = numbers[item_index];
+ item_index += 1;
+ let item = ListItem {
+ blocks: self.sub_blocks(child, delta)?,
+ marker_label: explicit_markers.then(|| format!("{number}.")),
+ };
+
+ if explicit_markers {
+ if current.is_none() {
+ current = Some(List { marker, start: 1, items: Vec::new() });
+ }
+ current.as_mut().unwrap().items.push(item);
+ continue;
+ }
+
let contiguous = current.is_some() && last_number.checked_add(1) == Some(number);
if !contiguous {
if let Some(list) = current.take() {
From 6372cf0846bed82c4457553135c0da5cb4d030c3 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 06:49:35 -0300
Subject: [PATCH 16/29] style: format HTML corpus changes
---
src/shared/html.rs | 3 +--
tests/html_corpus.rs | 8 ++------
2 files changed, 3 insertions(+), 8 deletions(-)
diff --git a/src/shared/html.rs b/src/shared/html.rs
index ccb4b18c..a5192d7a 100644
--- a/src/shared/html.rs
+++ b/src/shared/html.rs
@@ -559,8 +559,7 @@ impl Builder<'_> {
) -> Result, ConvertError> {
let ordered = elem.local == "ol";
let children: Vec<&Element> = elem.child_elems().collect();
- let items: Vec<&Element> =
- children.iter().copied().filter(|e| e.local == "li").collect();
+ let items: Vec<&Element> = children.iter().copied().filter(|e| e.local == "li").collect();
if !ordered {
let mut out = Vec::new();
diff --git a/tests/html_corpus.rs b/tests/html_corpus.rs
index e148bdc1..a62b2b12 100644
--- a/tests/html_corpus.rs
+++ b/tests/html_corpus.rs
@@ -30,10 +30,7 @@ fn controlled_document_preserves_rich_text_links_and_unicode() {
fn controlled_nested_lists_preserve_numbering_and_structure() {
let bytes = fs::read(html_fixture("controlled-lists.html")).unwrap();
let markdown = to_markdown_bytes(&bytes, Some(Format::Html)).unwrap();
- assert_eq!(
- markdown,
- "## Lists\n\n3. Third\n\n4. Fourth\n\n - Nested A\n - Nested B\n"
- );
+ assert_eq!(markdown, "## Lists\n\n3. Third\n\n4. Fourth\n\n - Nested A\n - Nested B\n");
}
#[test]
@@ -68,8 +65,7 @@ fn libreoffice_docx_text_preserves_core_document_semantics() {
#[test]
fn libreoffice_docx_numbering_preserves_list_content_and_emphasis() {
- let source =
- to_markdown(fixture_root().join("docx").join("handmade-numbering.docx")).unwrap();
+ let source = to_markdown(fixture_root().join("docx").join("handmade-numbering.docx")).unwrap();
let html = to_markdown(html_fixture("libreoffice-docx-numbering.html")).unwrap();
assert_both_contain(
From bbff29ab56a736266d957b6e9b26b9b492dcb523 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 06:53:23 -0300
Subject: [PATCH 17/29] test: refine LibreOffice numbering invariants
---
tests/html_corpus.rs | 6 +++++-
1 file changed, 5 insertions(+), 1 deletion(-)
diff --git a/tests/html_corpus.rs b/tests/html_corpus.rs
index a62b2b12..b345bb8c 100644
--- a/tests/html_corpus.rs
+++ b/tests/html_corpus.rs
@@ -79,7 +79,11 @@ fn libreoffice_docx_numbering_preserves_list_content_and_emphasis() {
"Interruption paragraph.",
"One-four continues the count",
"Ten-start via override",
- "Bold here, and style-false keeps it bold",
+ "pStyle-bound level one",
+ "pStyle-bound level two",
+ "pStyle-bound level one again",
+ "Bold here,",
+ "style-false keeps it bold",
"direct on makes this italic",
],
);
From fe53a0a4adca72e83ca37bafe5c41d5c19fe2e2a Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 06:55:08 -0300
Subject: [PATCH 18/29] test: include HTML fixtures in corpus detection
---
tests/snapshots.rs | 1 +
1 file changed, 1 insertion(+)
diff --git a/tests/snapshots.rs b/tests/snapshots.rs
index 25f997c2..9b77eee6 100644
--- a/tests/snapshots.rs
+++ b/tests/snapshots.rs
@@ -85,6 +85,7 @@ fn fixtures_detect_from_content() {
("doc", Some(Format::Doc)),
("docx", Some(Format::Docx)),
("epub", Some(Format::Epub)),
+ ("html", Some(Format::Html)),
("odp", Some(Format::Odp)),
("ods", Some(Format::Ods)),
("odt", Some(Format::Odt)),
From b6c08895f8b16351629ed6614786589b03ac4756 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 07:02:21 -0300
Subject: [PATCH 19/29] test: add HTML corpus snapshots
---
...shots__html__controlled-document.html.snap | 13 +++
...napshots__html__controlled-lists.html.snap | 12 +++
...s__html__controlled-merged-table.html.snap | 9 ++
...html__libreoffice-docx-numbering.html.snap | 49 +++++++++++
...ots__html__libreoffice-docx-text.html.snap | 83 +++++++++++++++++++
...s__html__libreoffice-xlsx-merged.html.snap | 9 ++
6 files changed, 175 insertions(+)
create mode 100644 tests/snapshots/snapshots__html__controlled-document.html.snap
create mode 100644 tests/snapshots/snapshots__html__controlled-lists.html.snap
create mode 100644 tests/snapshots/snapshots__html__controlled-merged-table.html.snap
create mode 100644 tests/snapshots/snapshots__html__libreoffice-docx-numbering.html.snap
create mode 100644 tests/snapshots/snapshots__html__libreoffice-docx-text.html.snap
create mode 100644 tests/snapshots/snapshots__html__libreoffice-xlsx-merged.html.snap
diff --git a/tests/snapshots/snapshots__html__controlled-document.html.snap b/tests/snapshots/snapshots__html__controlled-document.html.snap
new file mode 100644
index 00000000..0e3964eb
--- /dev/null
+++ b/tests/snapshots/snapshots__html__controlled-document.html.snap
@@ -0,0 +1,13 @@
+---
+source: tests/snapshots.rs
+expression: output
+---
+# Fixture Document
+
+Plain paragraph with **bold**, *italic*, and [example](https://example.com/page).
+
+## Unicode
+
+café Ł music 𝄞 family 👨👩👧.
+
+**Styled bold paragraph.**
diff --git a/tests/snapshots/snapshots__html__controlled-lists.html.snap b/tests/snapshots/snapshots__html__controlled-lists.html.snap
new file mode 100644
index 00000000..f52b3598
--- /dev/null
+++ b/tests/snapshots/snapshots__html__controlled-lists.html.snap
@@ -0,0 +1,12 @@
+---
+source: tests/snapshots.rs
+expression: output
+---
+## Lists
+
+3. Third
+
+4. Fourth
+
+ - Nested A
+ - Nested B
diff --git a/tests/snapshots/snapshots__html__controlled-merged-table.html.snap b/tests/snapshots/snapshots__html__controlled-merged-table.html.snap
new file mode 100644
index 00000000..d32b3294
--- /dev/null
+++ b/tests/snapshots/snapshots__html__controlled-merged-table.html.snap
@@ -0,0 +1,9 @@
+---
+source: tests/snapshots.rs
+expression: output
+---
+| | | |
+| --- | --- | --- |
+| Merged across | | padded |
+| tall | b2 | 3.5 |
+| | b3 | |
diff --git a/tests/snapshots/snapshots__html__libreoffice-docx-numbering.html.snap b/tests/snapshots/snapshots__html__libreoffice-docx-numbering.html.snap
new file mode 100644
index 00000000..ebd4ae55
--- /dev/null
+++ b/tests/snapshots/snapshots__html__libreoffice-docx-numbering.html.snap
@@ -0,0 +1,49 @@
+---
+source: tests/snapshots.rs
+expression: output
+---
+1. One-one
+
+2. One-two
+
+ - i. One-two-a roman
+
+ - ii. One-two-b roman
+
+ - Deep bullet
+
+3. One-three
+
+Interruption paragraph.
+
+4. One-four continues the count
+
+1. Two-one independent counter
+
+10. Ten-start via override
+
+- A. Letter-A via numStyleLink
+
+- G. Level replaced and start overridden together
+
+Suppressed numbering paragraph
+
+8. Style-numbered paragraph
+
+1. pStyle-bound level one
+
+pStyle-bound level two
+
+2. pStyle-bound level one again
+
+**Inherited subheading**
+
+**Bold here,** and style-false keeps it bold
+
+toggled off by a true in the character style
+
+**double toggle cancels to plain**
+
+direct off wins absolutely
+
+*direct on makes this italic*
diff --git a/tests/snapshots/snapshots__html__libreoffice-docx-text.html.snap b/tests/snapshots/snapshots__html__libreoffice-docx-text.html.snap
new file mode 100644
index 00000000..36dbbb5f
--- /dev/null
+++ b/tests/snapshots/snapshots__html__libreoffice-docx-text.html.snap
@@ -0,0 +1,83 @@
+---
+source: tests/snapshots.rs
+expression: output
+---
+1. # Fixture Document
+
+Plain paragraph with **bold**, *italic*, and ~~struck~~ runs.
+
+**Style-bold paragraph with a** NotBold-styled span **inside.**
+
+1. ## Lists
+
+1. First numbered
+
+2. Second numbered
+
+ - a. Alpha sub one
+
+ - b. Alpha sub two
+
+ - i. Roman sub sub
+
+3. Third numbered
+
+Interrupting paragraph between lists.
+
+4. Fourth, continuing the count
+
+- IV. Roman starting at four
+
+- I. Roman five
+
+- Bullet one
+
+- Bullet two
+
+ - Nested bullet
+
+2. ## Table
+
+| | | | |
+| --- | --- | --- | --- |
+| Wide head | | End | |
+| Tall | B2 | | C2 |
+| | B3 | | C3 |
+
+3. ## Notes and special text
+
+Music clef 𝄞 appears before this footnote[1](#sdfootnote1sym) reference.
+
+An endnote follows here
Date: Sat, 29 Aug 2026 07:08:09 -0300
Subject: [PATCH 20/29] test: keep corpus walk helper local
---
tests/common/mod.rs | 13 -------------
tests/snapshots.rs | 18 ++++++++++++++++--
2 files changed, 16 insertions(+), 15 deletions(-)
diff --git a/tests/common/mod.rs b/tests/common/mod.rs
index cb693101..3ba22488 100644
--- a/tests/common/mod.rs
+++ b/tests/common/mod.rs
@@ -5,16 +5,3 @@ use std::path::{Path, PathBuf};
pub fn fixture_root() -> PathBuf {
Path::new(env!("CARGO_MANIFEST_DIR")).join("tests").join("fixtures")
}
-
-/// Recursively collect every file under `dir`, sorted for determinism.
-pub fn walk(dir: &Path, out: &mut Vec) {
- let mut entries: Vec<_> = std::fs::read_dir(dir).unwrap().map(|e| e.unwrap().path()).collect();
- entries.sort();
- for path in entries {
- if path.is_dir() {
- walk(&path, out);
- } else {
- out.push(path);
- }
- }
-}
diff --git a/tests/snapshots.rs b/tests/snapshots.rs
index 9b77eee6..687d378f 100644
--- a/tests/snapshots.rs
+++ b/tests/snapshots.rs
@@ -9,9 +9,23 @@
mod common;
-use common::{fixture_root, walk};
+use common::fixture_root;
use std::fmt::Write as _;
-use std::path::Path;
+use std::path::{Path, PathBuf};
+
+
+/// Recursively collect every file under `dir`, sorted for determinism.
+fn walk(dir: &Path, out: &mut Vec) {
+ let mut entries: Vec<_> = std::fs::read_dir(dir).unwrap().map(|e| e.unwrap().path()).collect();
+ entries.sort();
+ for path in entries {
+ if path.is_dir() {
+ walk(&path, out);
+ } else {
+ out.push(path);
+ }
+ }
+}
/// Convert one file, capturing panics so a bad parser records a baseline
/// instead of aborting the whole harness.
From 6c78e2ce49067e4ed730fa748d3d6d70e6e91c82 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 07:09:22 -0300
Subject: [PATCH 21/29] style: format corpus walk helper
---
tests/snapshots.rs | 1 -
1 file changed, 1 deletion(-)
diff --git a/tests/snapshots.rs b/tests/snapshots.rs
index 687d378f..1129a2f8 100644
--- a/tests/snapshots.rs
+++ b/tests/snapshots.rs
@@ -13,7 +13,6 @@ use common::fixture_root;
use std::fmt::Write as _;
use std::path::{Path, PathBuf};
-
/// Recursively collect every file under `dir`, sorted for determinism.
fn walk(dir: &Path, out: &mut Vec) {
let mut entries: Vec<_> = std::fs::read_dir(dir).unwrap().map(|e| e.unwrap().path()).collect();
From ab653fb9cfb0a36c24a93f261a7a766bb119d86b Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 07:11:52 -0300
Subject: [PATCH 22/29] test: isolate HTML corpus fixture helper
---
tests/common/mod.rs | 13 +++++++++++++
tests/html_corpus.rs | 7 ++++---
tests/snapshots.rs | 17 ++---------------
3 files changed, 19 insertions(+), 18 deletions(-)
diff --git a/tests/common/mod.rs b/tests/common/mod.rs
index 3ba22488..cb693101 100644
--- a/tests/common/mod.rs
+++ b/tests/common/mod.rs
@@ -5,3 +5,16 @@ use std::path::{Path, PathBuf};
pub fn fixture_root() -> PathBuf {
Path::new(env!("CARGO_MANIFEST_DIR")).join("tests").join("fixtures")
}
+
+/// Recursively collect every file under `dir`, sorted for determinism.
+pub fn walk(dir: &Path, out: &mut Vec) {
+ let mut entries: Vec<_> = std::fs::read_dir(dir).unwrap().map(|e| e.unwrap().path()).collect();
+ entries.sort();
+ for path in entries {
+ if path.is_dir() {
+ walk(&path, out);
+ } else {
+ out.push(path);
+ }
+ }
+}
diff --git a/tests/html_corpus.rs b/tests/html_corpus.rs
index b345bb8c..8d1a552a 100644
--- a/tests/html_corpus.rs
+++ b/tests/html_corpus.rs
@@ -1,10 +1,11 @@
-mod common;
-
use anydoc::{Format, to_markdown, to_markdown_bytes};
-use common::fixture_root;
use std::fs;
use std::path::PathBuf;
+fn fixture_root() -> PathBuf {
+ PathBuf::from(env!("CARGO_MANIFEST_DIR")).join("tests").join("fixtures")
+}
+
fn html_fixture(name: &str) -> PathBuf {
fixture_root().join("html").join(name)
}
diff --git a/tests/snapshots.rs b/tests/snapshots.rs
index 1129a2f8..9b77eee6 100644
--- a/tests/snapshots.rs
+++ b/tests/snapshots.rs
@@ -9,22 +9,9 @@
mod common;
-use common::fixture_root;
+use common::{fixture_root, walk};
use std::fmt::Write as _;
-use std::path::{Path, PathBuf};
-
-/// Recursively collect every file under `dir`, sorted for determinism.
-fn walk(dir: &Path, out: &mut Vec) {
- let mut entries: Vec<_> = std::fs::read_dir(dir).unwrap().map(|e| e.unwrap().path()).collect();
- entries.sort();
- for path in entries {
- if path.is_dir() {
- walk(&path, out);
- } else {
- out.push(path);
- }
- }
-}
+use std::path::Path;
/// Convert one file, capturing panics so a bad parser records a baseline
/// instead of aborting the whole harness.
From 6c706695aa055584e370f4887a87ece1b2ba3ee6 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 07:37:25 -0300
Subject: [PATCH 23/29] test: cover non-rendering children inside lists
---
tests/html_list.rs | 27 +++++++++++++++++++++++++++
1 file changed, 27 insertions(+)
create mode 100644 tests/html_list.rs
diff --git a/tests/html_list.rs b/tests/html_list.rs
new file mode 100644
index 00000000..e6ed851c
--- /dev/null
+++ b/tests/html_list.rs
@@ -0,0 +1,27 @@
+use anydoc::{Format, to_markdown_bytes};
+
+#[test]
+fn unordered_list_ignores_non_rendering_children_without_splitting() {
+ let html = br#"
+ "#;
+
+ let markdown = to_markdown_bytes(html, Some(Format::Html)).unwrap();
+ assert_eq!(markdown, "- one\n- two\n");
+}
+
+#[test]
+fn ordered_list_ignores_non_rendering_children_without_splitting() {
+ let html = br#"
+
+ one
+
+ two
+ "#;
+
+ let markdown = to_markdown_bytes(html, Some(Format::Html)).unwrap();
+ assert_eq!(markdown, "1. one\n2. two\n");
+}
From 75cc2d272e1db7ec73a7e0ecfbc90bc244b69fbb Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 07:40:59 -0300
Subject: [PATCH 24/29] fix: ignore non-rendering children when grouping lists
---
src/shared/html.rs | 12 ++++++++++--
1 file changed, 10 insertions(+), 2 deletions(-)
diff --git a/src/shared/html.rs b/src/shared/html.rs
index a5192d7a..e35473b4 100644
--- a/src/shared/html.rs
+++ b/src/shared/html.rs
@@ -572,6 +572,10 @@ impl Builder<'_> {
});
continue;
}
+ let child_blocks = self.sub_element_blocks(child, delta)?;
+ if child_blocks.is_empty() {
+ continue;
+ }
if !list_items.is_empty() {
out.push(Block::List(List {
marker: MarkerKind::Bullet,
@@ -579,7 +583,7 @@ impl Builder<'_> {
items: std::mem::take(&mut list_items),
}));
}
- out.extend(self.sub_element_blocks(child, delta)?);
+ out.extend(child_blocks);
}
if !list_items.is_empty() {
out.push(Block::List(List {
@@ -626,10 +630,14 @@ impl Builder<'_> {
for child in children {
if child.local != "li" {
+ let child_blocks = self.sub_element_blocks(child, delta)?;
+ if child_blocks.is_empty() {
+ continue;
+ }
if let Some(list) = current.take() {
out.push(Block::List(list));
}
- out.extend(self.sub_element_blocks(child, delta)?);
+ out.extend(child_blocks);
continue;
}
From c6b7bb18608d4bde426a71d9a5300070d8f653fc Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sat, 29 Aug 2026 21:22:23 -0300
Subject: [PATCH 25/29] fix: preserve structural depth across anchor recovery
---
src/formats/html.rs | 6 ++++++
tests/html.rs | 24 ++++++++++++++++++++++++
2 files changed, 30 insertions(+)
diff --git a/src/formats/html.rs b/src/formats/html.rs
index b66c423b..e55eb345 100644
--- a/src/formats/html.rs
+++ b/src/formats/html.rs
@@ -212,6 +212,12 @@ impl TokenSink for HtmlComplexitySink {
}
fn close_implied_before_start(open: &mut Vec, name: &str) {
+ if name == "a"
+ && let Some(position) = open.iter().rposition(|candidate| candidate.as_ref() == "a")
+ {
+ open.remove(position);
+ }
+
if is_heading_element(name)
&& open.last().is_some_and(|candidate| is_heading_element(candidate.as_ref()))
{
diff --git a/tests/html.rs b/tests/html.rs
index 1099d129..57ab7eaa 100644
--- a/tests/html.rs
+++ b/tests/html.rs
@@ -300,3 +300,27 @@ fn nested_list_wrappers_preserve_structural_children() {
assert!(markdown.contains("Nested heading"), "{markdown:?}");
assert!(markdown.contains("| A |"), "{markdown:?}");
}
+
+#[test]
+fn repeated_unclosed_anchors_do_not_trigger_depth_limit() {
+ let mut html = String::from("");
+ for index in 0..300 {
+ html.push_str(&format!("link"));
+ }
+ let result = to_markdown_bytes(html.as_bytes(), Some(Format::Html));
+ assert!(
+ !matches!(result, Err(ConvertError::ResourceLimit { limit: "max_xml_depth", .. })),
+ "HTML5 repairs repeated anchors; preflight must not reject them as excessive depth"
+ );
+}
+
+#[test]
+fn intervening_blocks_between_unclosed_anchors_still_hit_preflight_depth_limit() {
+ let mut html = String::from("");
+ for index in 0..300 {
+ html.push_str(&format!(" "));
+ }
+
+ let error = to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap_err();
+ assert_preflight_depth_limit(error);
+}
From ee58d6a1aaafc64155077858d9a431de4839cb08 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sun, 30 Aug 2026 08:02:05 +0000
Subject: [PATCH 26/29] fix: handle framesets and relative HTML images
---
src/formats/html.rs | 16 ++++++++++------
tests/html.rs | 20 ++++++++++++++++++++
2 files changed, 30 insertions(+), 6 deletions(-)
diff --git a/src/formats/html.rs b/src/formats/html.rs
index e55eb345..effeacb1 100644
--- a/src/formats/html.rs
+++ b/src/formats/html.rs
@@ -46,10 +46,15 @@ pub fn parse(bytes: &[u8]) -> Result {
css.add(&style.text().collect::());
}
- let body = root
- .descendent_elements()
- .find(|e| e.value().name() == "body")
- .ok_or_else(|| ConvertError::malformed("HTML parser produced no body element"))?;
+ let body = match root.descendent_elements().find(|e| e.value().name() == "body") {
+ Some(body) => body,
+ None if root.descendent_elements().any(|e| e.value().name() == "frameset") => {
+ return Ok(Document::default());
+ }
+ None => {
+ return Err(ConvertError::malformed("HTML parser produced no body element"));
+ }
+ };
let mut node_count = 0usize;
let body = adapt_element(body, 1, &mut node_count)?;
@@ -466,8 +471,7 @@ impl HtmlCtx for StandaloneCtx {
if src.is_empty() {
return Ok(None);
}
- Ok((is_absolute_uri(src) || src.starts_with("//"))
- .then(|| ImageSource::External(src.to_owned())))
+ Ok(Some(ImageSource::External(src.to_owned())))
}
fn anchor_id(&self, raw: &str) -> AnchorId {
diff --git a/tests/html.rs b/tests/html.rs
index 57ab7eaa..cf65d76a 100644
--- a/tests/html.rs
+++ b/tests/html.rs
@@ -324,3 +324,23 @@ fn intervening_blocks_between_unclosed_anchors_still_hit_preflight_depth_limit()
let error = to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap_err();
assert_preflight_depth_limit(error);
}
+
+#[test]
+fn frameset_html_converts_without_body_error() {
+ let html = br#"frames
"#;
+ let document = to_document(html, Some(Format::Html)).unwrap();
+ match &document.blocks[0] {
+ Block::Paragraph(inlines) => assert!(matches!(
+ &inlines[0],
+ Inline::Image { source: ImageSource::External(url), .. }
+ if url == "images/pixel.png"
+ )),
+ other => panic!("expected paragraph, got {other:?}"),
+ }
+}
From ce7948287c8b4a62666f851a5845beabb091ef38 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin Barbacena
Date: Sun, 30 Aug 2026 08:06:01 +0000
Subject: [PATCH 27/29] test: update HTML image snapshots
---
.../snapshots__html__libreoffice-docx-text.html.snap | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/tests/snapshots/snapshots__html__libreoffice-docx-text.html.snap b/tests/snapshots/snapshots__html__libreoffice-docx-text.html.snap
index 36dbbb5f..566b5585 100644
--- a/tests/snapshots/snapshots__html__libreoffice-docx-text.html.snap
+++ b/tests/snapshots/snapshots__html__libreoffice-docx-text.html.snap
@@ -66,13 +66,13 @@ Jump to [the bookmarked paragraph](#plainmark).
5. ## Objects
-Inline image: done.
+Inline image:  done.
Text box: after the box.
Inside the text box.
-Shape1
+
6. ## Quote and code
From b7a119b7a964a9827bb470be2fe7a4e813b97c2d Mon Sep 17 00:00:00 2001
From: Marcell Manfrin
Date: Mon, 31 Aug 2026 22:29:24 -0300
Subject: [PATCH 28/29] fix: close implied paragraphs and preserve bare-hash
links in HTML
Cubic findings (PR 149 review run ed78d83a, shared HTML frontend):
1. The complexity preflight did not model HTML5's implicit closure
on block-level start tags, so repeated
pairs accumulated
phantom nesting depth: documents html5ever repairs into shallow trees
were rejected and the depth accounting diverged from the DOM the
limits are meant to model. Close an innermost open
when a
block-level start tag arrives; deeper arrangements remain
over-counted, keeping the preflight fail-closed.
2. href="#" produced LinkTarget::Anchor(""), which the Markdown
renderer cannot resolve, so the link was dropped to plain text.
Preserve the empty fragment as a relative "#" URL instead.
Regressions: 200
pairs now convert (depth 200 < limit), 300
pairs are still rejected before DOM construction, and a bare-hash link
renders as [top](#).
---
src/formats/html.rs | 56 +++++++++++++++++++++++++++++++++++++++++++++
tests/html.rs | 30 ++++++++++++++++++++++++
2 files changed, 86 insertions(+)
diff --git a/src/formats/html.rs b/src/formats/html.rs
index effeacb1..490ce8a4 100644
--- a/src/formats/html.rs
+++ b/src/formats/html.rs
@@ -223,6 +223,17 @@ fn close_implied_before_start(open: &mut Vec
, name: &str) {
open.remove(position);
}
+ // HTML5 implicitly closes an open when a block-level start tag
+ // arrives. Model the case where the
is the innermost open element;
+ // deeper arrangements stay over-counted, keeping the preflight
+ // fail-closed. (hr also closes
in HTML5 but is void, so it never
+ // reaches this hook; the leftover
only over-counts.)
+ if is_paragraph_closing_element(name)
+ && open.last().is_some_and(|candidate| candidate.as_ref() == "p")
+ {
+ open.pop();
+ }
+
if is_heading_element(name)
&& open.last().is_some_and(|candidate| is_heading_element(candidate.as_ref()))
{
@@ -250,6 +261,45 @@ fn is_heading_element(name: &str) -> bool {
matches!(name, "h1" | "h2" | "h3" | "h4" | "h5" | "h6")
}
+/// Block-level start tags that make HTML5 implicitly close an open
.
+fn is_paragraph_closing_element(name: &str) -> bool {
+ matches!(
+ name,
+ "address"
+ | "article"
+ | "aside"
+ | "blockquote"
+ | "details"
+ | "dialog"
+ | "div"
+ | "dl"
+ | "fieldset"
+ | "figcaption"
+ | "figure"
+ | "footer"
+ | "form"
+ | "h1"
+ | "h2"
+ | "h3"
+ | "h4"
+ | "h5"
+ | "h6"
+ | "header"
+ | "hgroup"
+ | "hr"
+ | "main"
+ | "menu"
+ | "nav"
+ | "ol"
+ | "p"
+ | "pre"
+ | "section"
+ | "summary"
+ | "table"
+ | "ul"
+ )
+}
+
fn html5_self_closing_is_honored(open: &[LocalName], name: &str) -> bool {
if matches!(name, "svg" | "math") {
return true;
@@ -456,6 +506,12 @@ impl HtmlCtx for StandaloneCtx {
return None;
}
if let Some(fragment) = href.strip_prefix('#') {
+ if fragment.is_empty() {
+ // A bare "#" points at the document itself. Keep it as a
+ // relative URL so the link survives rendering instead of
+ // becoming an unresolvable empty anchor.
+ return Some(LinkTarget::Relative(href.to_owned()));
+ }
let fragment = crate::package::path::decode_fragment(fragment);
return Some(LinkTarget::Anchor(fragment));
}
diff --git a/tests/html.rs b/tests/html.rs
index cf65d76a..caf7744d 100644
--- a/tests/html.rs
+++ b/tests/html.rs
@@ -344,3 +344,33 @@ fn relative_image_is_preserved_without_fetching() {
other => panic!("expected paragraph, got {other:?}"),
}
}
+
+#[test]
+fn paragraph_div_pairs_do_not_accumulate_paragraph_depth() {
+ // html5ever implicitly closes each
when the
arrives, so the
+ // real DOM nests 200 divs - well under max_xml_depth.
+ let mut html = String::from("");
+ for _ in 0..200 {
+ html.push_str("
text
");
+ }
+ html.push_str("");
+ assert!(to_markdown_bytes(html.as_bytes(), Some(Format::Html)).is_ok());
+}
+
+#[test]
+fn deeply_nested_paragraph_div_pairs_are_rejected_before_dom() {
+ let mut html = String::from("");
+ for _ in 0..300 {
+ html.push_str("
text
");
+ }
+ html.push_str("");
+ let error = to_markdown_bytes(html.as_bytes(), Some(Format::Html)).unwrap_err();
+ assert!(matches!(error, ConvertError::ResourceLimit { limit: "max_xml_depth", .. }));
+}
+
+#[test]
+fn bare_hash_link_is_preserved_as_relative_url() {
+ let html = br##"
top
"##;
+ let markdown = to_markdown_bytes(html, Some(Format::Html)).unwrap();
+ assert_eq!(markdown, "[top](#)\n");
+}
From 41a30a51544def72ddca777c201c45c16cc930f6 Mon Sep 17 00:00:00 2001
From: Marcell Manfrin
Date: Mon, 31 Aug 2026 23:24:58 -0300
Subject: [PATCH 29/29] fix: complete the paragraph-closing start tag set in
the HTML preflight
Cubic (PR 149, src/formats/html.rs:308) claimed dialog and summary do
not implicitly close an open ; html5ever's in-body mode
(tree_builder/rules.rs) proves they do. Keep both and complete the
preflight list to the parser's full close_p_element_in_button_scope
set: center, dir, listing, plaintext, and search were missing. table
is retained: html5ever closes
for
outside quirks mode,
and the preflight assumes standards mode.
---
src/formats/html.rs | 9 ++++++++-
1 file changed, 8 insertions(+), 1 deletion(-)
diff --git a/src/formats/html.rs b/src/formats/html.rs
index 490ce8a4..f9680ca6 100644
--- a/src/formats/html.rs
+++ b/src/formats/html.rs
@@ -261,7 +261,9 @@ fn is_heading_element(name: &str) -> bool {
matches!(name, "h1" | "h2" | "h3" | "h4" | "h5" | "h6")
}
-/// Block-level start tags that make HTML5 implicitly close an open .
+/// Block-level start tags that make HTML5 implicitly close an open
+/// (html5ever's in-body close_p_element_in_button_scope arms; table closes
+///
only outside quirks mode, and the preflight assumes standards mode).
fn is_paragraph_closing_element(name: &str) -> bool {
matches!(
name,
@@ -269,8 +271,10 @@ fn is_paragraph_closing_element(name: &str) -> bool {
| "article"
| "aside"
| "blockquote"
+ | "center"
| "details"
| "dialog"
+ | "dir"
| "div"
| "dl"
| "fieldset"
@@ -287,12 +291,15 @@ fn is_paragraph_closing_element(name: &str) -> bool {
| "header"
| "hgroup"
| "hr"
+ | "listing"
| "main"
| "menu"
| "nav"
| "ol"
| "p"
+ | "plaintext"
| "pre"
+ | "search"
| "section"
| "summary"
| "table"
