fix(lexer): don't split table cells on | inside links/code

The table-row lexer collected every `|` as a cell separator, so a pipe
inside a `[[url|title]]` wikilink (or `{{…}}` transclusion / `` `…` ``
inline code) was mistaken for a column boundary — mangling the link and
adding a phantom column in the AST and HTML export.

Replace the naive scan in `try_lex_table_row` with `cell_bar_positions`,
which skips pipes inside those constructs (matching the regions
`lex_inline` tokenises as a unit). Unclosed constructs fall back to
literal openers so a malformed cell can't swallow the rest of the row,
and the scan only slices at ASCII openers to stay UTF-8 safe.

Backslash-escaped `\|` is intentionally left unhandled: the inline lexer
has no escape handling, so honouring it here would leave a stray
backslash in the output. Documented as a matched follow-up.

Adds lexer + HTML-export regression tests.

Closes #1

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Lzhq43v1qpkZczN65dqXa3

crates/nuwiki-core/src/syntax/vimwiki/lexer.rs

@@ -617,15 +617,10 @@ impl<'src> LexState<'src> {
             end -= 1;
         }
 
-        // Collect cell boundaries.
+        // Collect cell boundaries. A `|` inside a `[[…]]` wiklink, a `{{…}}`
-        let mut sep_positions = Vec::new();
+        // transclusion, or `` `…` `` inline code is literal cell content, not
-        let mut k = i;
+        // a separator, so those regions are skipped (see `cell_bar_positions`).
-        while k < end {
+        let sep_positions = cell_bar_positions(line, i, end);
-            if bytes[k] == b'|' {
-                sep_positions.push(k);
-            }
-            k += 1;
-        }
 
         if sep_positions.len() < 2 {
             return false;
@@ -1068,6 +1063,51 @@ impl<'src> LexState<'src> {
 /// bracketed description like `[[page|Task [B-1]]]` keeps the `[B-1]` and closes
 /// at the final `]]`. A stray single `]` (no matching `[`) is treated as literal
 /// body text. Returns `None` if no closing `]]` is found.
+/// Byte offsets of the `|` characters in `line[i..end]` that act as
+/// table-cell separators. Pipes inside a `[[…]]` wikilink, a `{{…}}`
+/// transclusion, or `` `…` `` inline code are literal cell content — they
+/// mirror the regions [`VimwikiLexer::lex_inline`] tokenises as a single unit
+/// — so they are not reported as separators. An *unclosed* construct is
+/// treated as a literal opener (it does not start a skip region), so a
+/// malformed cell can never swallow the rest of the row.
+///
+/// Note: a backslash-escaped `\|` is intentionally *not* honoured here,
+/// because the inline lexer has no escape handling and would render the
+/// backslash literally; supporting it would need a matching change there.
+fn cell_bar_positions(line: &str, i: usize, end: usize) -> Vec<usize> {
+    let bytes = line.as_bytes();
+    let mut positions = Vec::new();
+    let mut k = i;
+    while k < end {
+        // These openers are ASCII, so `bytes[k]` never matches a UTF-8
+        // continuation byte and every slice below starts on a char boundary.
+        match bytes[k] {
+            b'[' if bytes.get(k + 1) == Some(&b'[') => {
+                if let Some(close) = find_wikilink_close(&line[k + 2..end]) {
+                    k += close + 4; // `[[` + inner + `]]`
+                    continue;
+                }
+            }
+            b'{' if bytes.get(k + 1) == Some(&b'{') => {
+                if let Some(close) = line[k + 2..end].find("}}") {
+                    k += close + 4; // `{{` + inner + `}}`
+                    continue;
+                }
+            }
+            b'`' => {
+                if let Some(close) = line[k + 1..end].find('`') {
+                    k += close + 2; // opening + inner + closing backtick
+                    continue;
+                }
+            }
+            b'|' => positions.push(k),
+            _ => {}
+        }
+        k += 1;
+    }
+    positions
+}
+
 fn find_wikilink_close(s: &str) -> Option<usize> {
     let b = s.as_bytes();
     let mut depth: u32 = 0;

crates/nuwiki-core/tests/html_renderer.rs

@@ -245,6 +245,19 @@ fn table_with_header_separator() {
     assert!(out.contains("<td> 1 </td>"));
 }
 
+#[test]
+fn table_cell_with_wikilink_pipe_exports_one_cell() {
+    // Regression: the `|` in `[[url|title]]` must stay inside the link, so the
+    // row keeps two columns and the link renders as a single anchor. Before the
+    // fix the cell split into `[[https://example.com` + `My Site]]`, adding a
+    // phantom <td> and breaking the link.
+    let out = render("| Name | Link |\n|---|---|\n| foo | [[https://example.com|My Site]] |\n");
+    assert!(
+        out.contains("<td> foo </td><td> <a href=\"https://example.com\">My Site</a> </td>"),
+        "got: {out}"
+    );
+}
+
 #[test]
 fn single_comment_becomes_html_comment() {
     let out = render("%% hidden\n");

crates/nuwiki-core/tests/lexer.rs

@@ -454,6 +454,48 @@ fn table_col_and_row_span_cells() {
     );
 }
 
+#[test]
+fn table_cell_with_wikilink_pipe_is_not_split() {
+    // The `|` inside `[[url|title]]` is part of the link, not a cell
+    // boundary: the row must lex to exactly two cells (three TableSeps).
+    assert_eq!(
+        lex("| foo | [[https://example.com|My Site]] |\n"),
+        vec![
+            TableSep,
+            text(" foo "),
+            TableSep,
+            text(" "),
+            WikiLinkOpen,
+            text("https://example.com"),
+            WikiLinkSep,
+            text("My Site"),
+            WikiLinkClose,
+            text(" "),
+            TableSep,
+            Newline,
+        ]
+    );
+}
+
+#[test]
+fn table_cell_with_transclusion_and_code_pipes_are_not_split() {
+    // `|` inside `{{…}}` and inside `` `…` `` are both literal.
+    let lexed = lex("| {{a|b}} | `x|y` |\n");
+    let seps = lexed.iter().filter(|t| matches!(t, TableSep)).count();
+    assert_eq!(seps, 3, "expected 2 cells: {lexed:?}");
+}
+
+#[test]
+fn table_cell_with_unclosed_wikilink_falls_back_to_pipe_split() {
+    // A malformed, unclosed `[[` must not swallow the rest of the row —
+    // the pipes still delimit cells.
+    let seps = lex("| [[oops | tail |\n")
+        .iter()
+        .filter(|t| matches!(t, TableSep))
+        .count();
+    assert_eq!(seps, 3);
+}
+
 // ===== Definition list =====
 
 #[test]