Restore backslash escapes before parsing TeX math content

[rhennigan]

Summary

A markdown-escaped character inside a math block was mangled instead of rendered. $$\$123.45 / \$6.78$$ produced garbage where the dollar signs should be.

Before, on main:

In[1]:= Needs[ "Wolfram`Chatbook`" ]
In[2]:= FormatChatOutput[ "Here is some TeX: $$\$123.45 / \$6.78$$" ] // InputForm

<|"boxes" -> FormBox[RowBox[{"ï¸36ï¹123.45", "/", "ï¸36ï¹6.78"}], TraditionalForm],
  "errors" -> {}, "input" -> "$123.45 / $6.78", "state" -> "Boxes"|>

After:

<|"boxes" -> FormBox[RowBox[{"$123.45", "/", "$6.78"}], TraditionalForm],
  "errors" -> {}, "input" -> "\\$123.45 / \\$6.78", "state" -> "Boxes"|>

Root cause

reformatTextData0 rewrites markdown escapes like \$ into a private sentinel (esc, a \[EntityStart]…\[EntityEnd] pair) before the text is parsed into cells, then restores them afterward with $mdUnescapeRules. Those rules drop the backslash, which is right for body text — there the backslash only marks the escape, so \$ should render as a bare $.

Math content never takes that path. It is handed to makeTeXBoxes as a raw string, so the sentinel reached the TeX parser verbatim and came back as the mojibake above, once those private use characters were reinterpreted.

The two halves of the "before" output fail differently, which is worth noting since only one of them is visible to users:

• "boxes" was mangled by the TeX parser, and no longer contained a sentinel for anything downstream to recognize.
• "input" still held real sentinels, so the outer unescape sweep in reformatTextData0 restored them — to a bare $, silently stripping the TeX escape rather than showing mojibake.

Fix

Unescape math with the new $texUnescapeRules, which restores the sentinel to \<char> and keeps the backslash. Inside $$...$$ the escape belongs to TeX rather than markdown, so the backslash has to reach the parser: \$ is how TeX renders a literal dollar sign, and likewise \_ and \#. The remaining escaped characters (`, *, |) keep whatever meaning TeX assigns the sequence, which still beats passing the sentinel through.

The change sits on the general mathCell definition only. The digits-only and system-name definitions above it can't hold a sentinel, and the fallback to inlineCodeCell returns strings, which $mdUnescapeRules still catches on the way out.

Test plan

• New Tests/Formatting.wlt — 5/5 pass. Formatting.wl had no test coverage before this.
• Confirmed the tests actually catch the bug: both TeX tests were run against the pre-fix code and fail there, reproducing the exact mojibake above. Without that check a passing test proves nothing here.
• TeX-Escaped-Dollar pins the reported case; TeX-Escaped-Underscore-And-Hash covers \_ and \# to show the rule generalizes.
• Markdown-Escaped-Dollar is a guard, not a regression test: it passes both before and after, and pins the neighboring markdown path (\$ → $ in body text) that this change could otherwise have quietly broken. The two unescape rule sets are easy to confuse.
• CodeInspector clean on both changed files.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR fixes TeX math rendering in FormatChatOutput when markdown-escaped characters (notably \$) appear inside $$...$$ blocks, by restoring backslash escapes before passing math content to the TeX parser. This aligns math handling with how markdown escapes are already round-tripped elsewhere in reformatTextData0, while preserving TeX’s need to see escapes like \$, \_, and \# intact.

Changes:

• Add a dedicated $texUnescapeRules rule set to restore the escape sentinel back to \-prefixed sequences for TeX parsing.
• Apply $texUnescapeRules to mathCell content before calling makeTeXBoxes.
• Add a new Tests/Formatting.wlt file covering TeX-escaped $, _, #, plus a guard test for markdown-body \$ behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
Source/Chatbook/Formatting.wl	Restores backslash escapes from the markdown sentinel before TeX parsing for mathCell strings.
Tests/Formatting.wlt	Adds regression and guard tests validating correct TeX-escape handling vs. markdown-body unescaping.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.