fix(markdown): leftover extra dd after blank and footnote fence

src/parser/markdown.rs

@@ -181,6 +181,21 @@ fn next_nonblank_indent(lines: &[Line<'_>], start: usize) -> Option<usize> {
         .map(|l| line_indent(l.text))
 }
 
+/// Next non-blank is a definition marker (`: ` / `~ `), including after
+/// quote markers. pulldown keeps the DL open across one blank then
+/// another `<dd>`.
+fn next_nonblank_is_dl_marker(lines: &[Line<'_>], start: usize) -> bool {
+    let Some(line) = lines[start..].iter().find(|l| !l.text.trim().is_empty()) else {
+        return false;
+    };
+    if md_definition_list_marker_len(line.text).is_some() {
+        return true;
+    }
+    let depth = quote_marker_depth(line.text);
+    strip_quote_markers(line.text, depth)
+        .is_some_and(|inner| md_definition_list_marker_len(inner).is_some())
+}
+
 fn starts_html_comment(line: &str) -> bool {
     // CM 4.6 type 2: at most three spaces. Four-space `<!--` is indented
     // code (or lazy title text), not an HTML block.
@@ -2064,7 +2079,8 @@ impl FormatParser for MarkdownParser {
                     && list_hang.is_some_and(|hang| {
                         next_nonblank_indent(&lines, i + 1).is_some_and(|ind| ind >= hang)
                     });
-                if stay_in_item {
+                let stay_in_dl = in_definition_list && next_nonblank_is_dl_marker(&lines, i + 1);
+                if stay_in_item || stay_in_dl {
                     flush_prose_spanned(&mut current_prose, &mut prose_span, &mut regions);
                     if let Some(span) = list_term.take() {
                         if !span.is_empty() {
@@ -2344,8 +2360,43 @@ impl FormatParser for MarkdownParser {
                 );
                 let opener = i;
                 let end = footnote_def_end(&lines, opener);
-                for row in &lines[opener + 1..=end] {
+                let mut j = opener + 1;
+                while j <= end {
+                    let row = &lines[j];
                     let hang = line_indent(row.text);
+                    let inner = row.text.get(hang..).unwrap_or("");
+                    if FENCED_CODE_RE.is_match(inner.trim_start()) {
+                        flush_prose_spanned(&mut current_prose, &mut prose_span, &mut regions);
+                        let fence = FENCED_CODE_RE
+                            .captures(inner.trim_start())
+                            .and_then(|c| c.get(1))
+                            .map(|m| m.as_str())
+                            .unwrap_or("```");
+                        let close = lines[j + 1..=end]
+                            .iter()
+                            .enumerate()
+                            .find_map(|(off, line)| {
+                                let kh = line_indent(line.text);
+                                let ki = line.text.get(kh..).unwrap_or("");
+                                ki.trim_start().starts_with(fence).then_some(j + 1 + off)
+                            })
+                            .unwrap_or(end);
+                        if hang > 0 {
+                            regions.push(SpannedRegion::structure(
+                                input,
+                                ByteSpan::new(row.start, row.start + hang),
+                            ));
+                        }
+                        regions.push(SpannedRegion::code(
+                            input,
+                            None,
+                            ByteSpan::new(row.start + hang, row.end),
+                            ByteSpan::new(row.start + hang, lines[close].end),
+                            ByteSpan::new(lines[close].end, lines[close].end),
+                        ));
+                        j = close + 1;
+                        continue;
+                    }
                     if hang > 0 {
                         let hang_span = ByteSpan::new(row.start, row.start + hang);
                         regions.push(SpannedRegion::structure(input, hang_span));
@@ -2363,6 +2414,7 @@ impl FormatParser for MarkdownParser {
                         input,
                         &mut regions,
                     );
+                    j += 1;
                 }
                 i = end + 1;
                 // pulldown pops the footnote on a column-0 line even
@@ -2652,6 +2704,9 @@ impl FormatParser for MarkdownParser {
                 let marker_len = line_text.len() - text.len();
                 if text.trim().is_empty() {
                     regions.push(SpannedRegion::structure(input, line.span()));
+                    if was_in_definition_list {
+                        in_definition_list = true;
+                    }
                     i += 1;
                     continue;
                 }

src/sentence/unicode.rs

@@ -1578,6 +1578,8 @@ pub(crate) fn leftover_keyval_cs_name(tail: &str) -> Option<&'static str> {
         "tcbinputrecords",
         "tcbsidebyside",
         "tcbsubskin",
+        "tcbmakedocSubKeys",
+        "tcbmakedocSubKey",
         "tcbmakeprefixed",
         "tcbhyperref",
         "tcbhypernode",
@@ -1791,7 +1793,9 @@ fn leftover_keyval_kind(name: &str) -> VerbKind {
         | "tcbsetmacrotowidthofnode"
         | "tcbsettoheightofnode"
         | "tcbsetmacrotoheightofnode"
-        | "tcbsetfromto" => VerbKind::Listinginput,
+        | "tcbsetfromto"
+        | "tcbmakedocSubKeys"
+        | "tcbmakedocSubKey" => VerbKind::Listinginput,
         "RecustomVerbatimEnvironment"
         | "CustomVerbatimEnvironment"
         | "DefineVerbatimEnvironment"
@@ -3034,7 +3038,8 @@ impl UnicodeSentenceSplitter {
             self.extra_pattern.as_ref(),
         );
         let merged = merge_quoted_punct_splits(merged);
-        merge_splits_inside_delimiters(merged)
+        let merged = merge_splits_inside_delimiters(merged);
+        split_double_bang_then_letter(merged)
     }
 }
 
@@ -3169,7 +3174,10 @@ fn push_segment_preserving_space(dest: &mut String, piece: &str) {
     let need_space = dest.chars().last().is_some_and(|c| !c.is_whitespace())
         && next.is_some_and(|c| {
             !c.is_whitespace() && (c.is_alphanumeric() || matches!(c, '"' | '\'' | '`' | '('))
-        });
+        })
+        // `!!a` is one UAX fragment; inventing `!! a` then splitting
+        // `!! a` on the next pass is a wrap/SemBr cycle.
+        && !dest.ends_with("!!");
     if need_space {
         dest.push(' ');
     }
@@ -3284,6 +3292,84 @@ fn merge_quoted_punct_splits(segments: Vec<String>) -> Vec<String> {
 /// UAX SB11 breaks after ATerm+Sp *before* a closer, so `(. aA. )A` is
 /// `(. aA. ` + `)A`. Glue only the closer; a following capital is a new
 /// sentence (GitHub #266). The tight form `(. aA.)A` already splits.
+///
+/// `!!a` is one UAX fragment. Split so first and second format_text
+/// passes agree (`!!` ends the sentence; the letter starts the next).
+/// Do not split when `!!` sits inside a wrapper this piece closes
+/// (`"!!a"`, `'!!A'A`). DelimState's apostrophe heuristic treats the
+/// closer in `'!!A'A` as `A'A`, so this walk toggles `'` without it.
+/// An unclosed opener (`[=!!a`) still splits.
+fn split_double_bang_then_letter(segments: Vec<String>) -> Vec<String> {
+    let mut out = Vec::new();
+    for seg in segments {
+        let chars: Vec<(usize, char)> = seg.char_indices().collect();
+        let mut start = 0;
+        let mut idx = 0;
+        while idx + 1 < chars.len() {
+            if chars[idx].1 == '!' && chars[idx + 1].1 == '!' {
+                if let Some(&(off, ch)) = chars.get(idx + 2) {
+                    if ch.is_alphabetic() && !wrap_closes_after_bang(&seg[start..], off - start) {
+                        out.push(seg[start..off].to_string());
+                        start = off;
+                        idx += 2;
+                        continue;
+                    }
+                }
+            }
+            idx += 1;
+        }
+        if start < seg.len() {
+            out.push(seg[start..].to_string());
+        }
+    }
+    out
+}
+
+/// True when `!!` at `bang_at` is inside `"…"` / `'…'` / ` ``…'' ` /
+/// `(…)` / `[…]` / `{…}` and that wrapper closes later in `piece`.
+fn wrap_closes_after_bang(piece: &str, bang_at: usize) -> bool {
+    let chars: Vec<(usize, char)> = piece.char_indices().collect();
+    let mut dq = false;
+    let mut sq = false;
+    let mut latex = 0i32;
+    let mut paren = 0i32;
+    let mut bracket = 0i32;
+    let mut brace = 0i32;
+    let mut i = 0;
+    while i < chars.len() {
+        let ch = chars[i].1;
+        let next = chars.get(i + 1).map(|c| c.1);
+        if ch == '`' && next == Some('`') {
+            latex += 1;
+            i += 2;
+        } else if ch == '\'' && next == Some('\'') {
+            latex = (latex - 1).max(0);
+            i += 2;
+        } else {
+            match ch {
+                '"' => dq = !dq,
+                '\'' => sq = !sq,
+                '(' => paren += 1,
+                ')' => paren -= 1,
+                '[' => bracket += 1,
+                ']' => bracket -= 1,
+                '{' => brace += 1,
+                '}' => brace -= 1,
+                _ => {}
+            }
+            i += 1;
+        }
+        let consumed_to = chars.get(i).map_or(piece.len(), |c| c.0);
+        if consumed_to == bang_at {
+            let inside = dq || sq || latex > 0 || paren > 0 || bracket > 0 || brace > 0;
+            if !inside {
+                return false;
+            }
+        }
+    }
+    !dq && !sq && latex <= 0 && paren <= 0 && bracket <= 0 && brace <= 0
+}
+
 fn merge_splits_inside_delimiters(segments: Vec<String>) -> Vec<String> {
     let mut result: Vec<String> = Vec::with_capacity(segments.len());
     let mut state = DelimState::default();
@@ -3572,6 +3658,15 @@ mod tests {
         UnicodeSentenceSplitter::new().split(text)
     }
 
+    #[test]
+    fn single_quote_bang_capital_stays_one_sentence() {
+        assert_eq!(
+            split("'!!A'A"),
+            vec!["'!!A'A".to_string()],
+            "UAX/refine must keep balanced '!!A' as one sentence"
+        );
+    }
+
     #[test]
     fn simple_sentences() {
         assert_eq!(

tests/md_compact_dl_terms.rs

@@ -170,6 +170,36 @@ fn leftover_loose_second_dl_term_after_blank() {
     assert_eq!(format_text(&out, &md_cfg()).unwrap(), out);
 }
 
+#[test]
+fn leftover_loose_extra_definition_after_blank() {
+    let input = concat!(
+        "Term\n",
+        ": First definition sentence. Second sentence.\n",
+        "\n",
+        ": Third definition sentence. Fourth sentence.\n",
+        "\n",
+        "After the list. Next.\n",
+    );
+    let out = format_text(input, &md_cfg()).unwrap();
+    assert!(
+        !out.contains("Third definition sentence. Fourth sentence."),
+        "second dd body must still split, got:\n{out}"
+    );
+    assert!(
+        out.contains("Fourth sentence."),
+        "fourth sentence must stay, got:\n{out}"
+    );
+    assert!(
+        !out.lines().any(|l| l == "Fourth sentence."),
+        "fourth sentence must stay hung, got:\n{out}"
+    );
+    assert!(
+        out.contains("After the list.\nNext."),
+        "following prose must still split, got:\n{out}"
+    );
+    assert_eq!(format_text(&out, &md_cfg()).unwrap(), out);
+}
+
 #[test]
 fn leftover_list_item_second_compact_dl_term_after_definition() {
     let input = concat!(

tests/md_link_ref_footnotes.rs

@@ -63,6 +63,40 @@ fn ticket_fixture() -> &'static str {
     )
 }
 
+#[test]
+fn leftover_footnote_fence_continuation_is_code() {
+    let input = concat!(
+        "[^1]: First sentence. Second sentence.\n",
+        "\n",
+        "    ```\n",
+        "    code. yes\n",
+        "    ```\n",
+        "\n",
+        "After the note. Next.\n",
+    );
+    let regions = MarkdownParser.parse(input);
+    assert!(
+        regions.iter().any(|r| matches!(
+            r,
+            Region::Code { body, .. } if body.contains("code. yes")
+        )),
+        "indented fence in a footnote must be Code, got {regions:?}"
+    );
+    let out = format_text(input, &md_cfg()).unwrap();
+    assert!(
+        out.contains("code. yes"),
+        "fence body must stay one line, got:\n{out}"
+    );
+    assert!(
+        !out.contains("code.\nyes"),
+        "must not SemBr-split inside the fence, got:\n{out}"
+    );
+    assert!(
+        out.contains("After the note.\nNext."),
+        "following prose must still split, got:\n{out}"
+    );
+}
+
 #[test]
 fn leftover_footnote_col0_after_continuation_is_new_paragraph() {
     let input = concat!(

tests/sentence_delim_props.rs

@@ -22,6 +22,78 @@ fn plaintext_cfg() -> FormatConfig {
 
 /// PR #52 CI seed: period before backticks, then quote-backtick.
 /// Markup closer-split must not invent a newline inside a DelimState span.
+#[test]
+fn plaintext_setext_looking_bang_letter_is_idempotent() {
+    let input = ".=A[=!!a";
+    let out = format_plain(input);
+    assert_eq!(
+        format_plain(&out),
+        out,
+        "idempotence\n in={input:?}\n out={out:?}"
+    );
+}
+
+/// Quoted `!!a` is one DelimState span. Splitting `!!` then the letter
+/// parks a newline inside the quotes and fails proptest_balanced_wrappers.
+#[test]
+fn plaintext_single_quoted_bang_capital_is_span_safe() {
+    let input = "'!!A'A";
+    let out = format_plain(input);
+    assert_eq!(
+        format_plain(&out),
+        out,
+        "idempotence\n in={input:?}\n out={out:?}"
+    );
+    assert!(
+        newlines_respect_delimiter_spans(&out),
+        "span newline\n in={input:?}\n out={out:?}"
+    );
+    assert!(
+        !out.contains("!!\n"),
+        "must not split !!A inside quotes\n in={input:?}\n out={out:?}"
+    );
+}
+
+/// Quoted `!!a` is one DelimState span. Splitting `!!` then the letter
+/// parks a newline inside the quotes and fails proptest_balanced_wrappers.
+#[test]
+fn plaintext_latex_quoted_bang_letter_is_span_safe() {
+    let input = "``!!a''";
+    let out = format_plain(input);
+    assert_eq!(
+        format_plain(&out),
+        out,
+        "idempotence\n in={input:?}\n out={out:?}"
+    );
+    assert!(
+        newlines_respect_delimiter_spans(&out),
+        "span newline\n in={input:?}\n out={out:?}"
+    );
+    assert!(
+        !out.contains("!!\n"),
+        "must not split !!a inside latex quotes\n in={input:?}\n out={out:?}"
+    );
+}
+
+#[test]
+fn plaintext_quoted_bang_letter_is_span_safe() {
+    let input = "\"!!a\"";
+    let out = format_plain(input);
+    assert_eq!(
+        format_plain(&out),
+        out,
+        "idempotence\n in={input:?}\n out={out:?}"
+    );
+    assert!(
+        newlines_respect_delimiter_spans(&out),
+        "span newline\n in={input:?}\n out={out:?}"
+    );
+    assert!(
+        !out.contains("!!\n"),
+        "must not split !!a inside quotes\n in={input:?}\n out={out:?}"
+    );
+}
+
 #[test]
 fn plaintext_period_before_backticks_quote_is_span_safe() {
     let input = ".`` \"`";