[v3.0.2] List rendering robustness fixes (itemize / enumerate)

[OlgaRedozubova]

Summary

Two groups of renderer fixes, shipped together as one 3.0.2 release:

• Lists (1–4): four unrelated inputs made LaTeX itemize / enumerate lists render incorrectly; the fixes live in the list-env and footnote block rules.

• Code blocks (5): code-text styles were absolute (px / rem), so a code block did not scale when a consumer sizes a rendered block by a single font-size; the styles are now relative.

• Version bump: 3.0.1 → 3.0.2

• Specs: pr-specs/2026-07-list-rendering-robustness.md, pr-specs/2026-07-code-block-font-scaling.md

• Changelog: doc/changelog.md

• Full test suite green.

---

1. Marker padding for block-content items

A top-level list gets its padding-inline-start from the widest custom \item[...] marker, but the width was only measured on the inline item path. Items whose content is a block environment (\begin{figure}, \begin{tabular}, a code fence) were skipped, so a list whose long-marker items all hold block content lost its padding.

MMD example

\begin{itemize}
\item[11.33] In Fig. 11-20, when pump YA delivers 5.00 cfs.
\begin{figure}
\includegraphics[alt={},max width=\textwidth]{https://cdn.mathpix.com/snip/images/QxUdugRlfaW7h_IxvYNbgRDXjd0egDGrEXiSjKH7FGU.original.fullsize.png}
\caption{Fig. 11-20}
\end{figure}
\item[11.34] In Fig. 11-21, which system has the greater capacity.
\begin{figure}
\includegraphics[alt={},max width=\textwidth]{https://cdn.mathpix.com/snip/images/QxUdugRlfaW7h_IxvYNbgRDXjd0egDGrEXiSjKH7FGU.original.fullsize.png}
\caption{Fig. 11-21}
\end{figure}
\end{itemize}

Before	After

---

2. Marker width: fullwidth/CJK, math, and wrapped content

Marker width summed String.length over top-level text tokens only, so several markers were undercounted and the list got too small an indent (marker overlaps the content):

• Fullwidth / CJK (\item[11．], U+FF0E) counted as narrow ASCII — now East-Asian Wide/Fullwidth chars count as 2.
• Math (\item[$x^4+x^4$]) contributed 0 — now uses the token's rendered widthEx.
• Wrapped content (\item[\textbf{…}]) contributed 0 — now measured through the wrapper's children.

ASCII text markers are unchanged.

MMD examples

\begin{itemize}
\item[I] Item 1．
\item[11．] Item 11．．
\item[I] Item 1．
\end{itemize}

\begin{itemize}
\item[$x^4 + x^4$] a footnote note or item body
\end{itemize}

\begin{itemize}
\item[\textbf{x^4 + x^4}] a footnote note or item body
\end{itemize}

Case	Before	After
fullwidth 11．
math $x^4+x^4$
bold \textbf{…}

---

3. No env-state leak from an aborted list parse

The list block rule parses speculatively into a buffered state that shares env by prototype. On abort (unclosed list) or a silent probe it returned without restoring env.isBlock (and env.inheritedListType); the leaked isBlock = true then let the inline list fallback fire on the following content, so an unclosed itemize before a tabular rendered a broken partial list with empty <> item bodies. Those transient fields are now restored on both paths, so the unclosed list degrades to plain text — exactly as it does without a tabular.

MMD example

\begin{itemize}
\item[(d1)] Asad Ali Khan
\item[(d2)] Asad Ali Khan
\begin{tabular}{|l|l|}
cell
\end{tabular}
\begin{itemize}

Before	After

---

4. Footnote block rules stop at a list start

The \footnote / \footnotetext block rules scan forward for their open tag, terminating at block boundaries so they don't swallow following blocks. The LaTeX list rule was not a terminator, so a \begin{itemize} between a paragraph and a later footnote (no blank line) did not stop the scan: the list was swallowed and rendered as literal text (a blank line masked the bug). The LaTeX list rule is now a terminator for both — \footnotetext via its full set, \footnote via a minimal fence + Lists (keeping its cheap scan).

This is also a performance fix: on repeated paragraph + list-with-footnotetext units without blank separators the missing terminator made the scan run across every list into the rest of the document — O(N²), seconds on large inputs; terminating at the list makes it linear (guarded by a scaling test).

MMD example

Wiener's life work, its enormous range notwithstanding, exhibits a coherence of thought from start to finish, reminiscent of a great work of art. To write a biography of a person of so towering a stature is an
\begin{itemize}
\item[] \footnotetext{
1 Fortunately, this correspondence is now being studied by Dr. Albert C. Lewis of the Bertrand Russell Editorial Project, McMaster University, Ontario, Canada, and may see the light of day within a few years.
}
\end{itemize}

Before	After

---

5. Code-block styles scale with the em context

Code-text styles were pinned to absolute values, so when a consumer scales a rendered block by setting a single font-size on the container (e.g. image export), everything scaled except the code — px is fixed and rem resolves against the root, not the block's em. The four properties are now relative, calibrated so a 16px base is pixel-identical to before (only code padding moves 16px → 15px):

• #setText pre { font-size: 0.9375em; } (was 85%)
• #setText pre code { font-size: inherit; } (was 15px)
• #setText pre code { line-height: 1.6; } (was 24px)
• #setText pre code { padding: 1em; } (was 1rem)

Styles only — no change to lstlisting / fenced-code markup.

MMD example

\begin{lstlisting}
def area(r):
    return 3.14159 * r * r
\end{lstlisting}

Scale the rendered block by setting a large font-size on the container (as image export does).

Before	After

---

Testing

List cases in tests/_data/_lists/_data.js and tests/_list-marker-padding.js: block-content markers (figure/fence), fullwidth 11．, math and \textbf markers, an unclosed list + tabular, a list after a paragraph with a multiline \footnote{} / \footnotetext{}, and a markdown list not swallowed before a \footnote.

Guards: a scaling test (tests/_footnotes_latex.js) rejecting the O(N²) footnote scan; a silent-Lists env invariant (tests/_parse-isolation.js); selector-scoped code-style assertions (tests/_styles.js).

Full suite green.

Non-goals

• The list padding heuristic itself (px-per-char, the > 3 threshold) is unchanged.
• Malformed list input still degrades gracefully; the goal is plain text, not a partial list.
• Nested lists keep the default indent even with long markers (pre-existing; separate ticket).
• Code blocks: rendering at the default 16px base is unchanged apart from the 1px code padding; inline code, pre scroll/overflow, highlight colors, and table-cell padding are untouched.