Reproduction steps

1. Use any LSP server that emits Cyrillic text in Diagnostic.message.
2. Open a file that triggers a diagnostic whose message contains both:
   • a Cyrillic letter from the ranges U+043A–U+0440 (к л м н о п р) or U+0421–U+042F (С Т У Ф Х Ц Ч Ш Щ Ъ Ы Ь Э Ю Я), and
   • any markdown-special character (', ", (, ), #, /, …).
3. Inspect the diagnostic message in the editor.

Current behavior

Spurious \ characters appear before specific Cyrillic letters:

Отсутствует \п\р\обе\л с\п\рава \от ','

В\мест\о уста\ревшег\о св\ойства "\Эта\Ф\о\р\ма" с\ледует ис\п\о\льз\овать "\Эт\отОбъе\кт"

П\р\оцеду\ра \нах\одится в\не \об\ласти (#Об\ласть/#Region). Весь \к\од \м\оду\ля д\о\лже\н быть \о\рга\низ\ова\н в \об\ласти д\ля \лучшей ст\ру\кту\ры.

Copying the message yields the clean text — the backslashes are inserted by the renderer.

Expected behavior

Display the message exactly as the LSP server sent it, e.g.:

Отсутствует пробел справа от ','

Root cause

crates/markdown/src/markdown.rs, pub fn escape (~line 457):

for c in s.chars() {
    escaper.next(c as u8).write_to(c, &mut output);
}

MarkdownEscaper::next decides to prefix a backslash via byte.is_ascii_punctuation(). For non-ASCII codepoints, c as u8 truncates to the low 8 bits, which can land inside ASCII-punctuation ranges — e.g. р (U+0440) → 0x40 → @, о (U+043E) → 0x3E → >, Э (U+042D) → 0x2D → -. Per CommonMark §6.1, \X is consumed only when X is ASCII punctuation, so for non-ASCII characters the backslash is rendered literally.

The earlier fix in #51766 addressed visible escapes inside indented code blocks but did not touch this lossy-cast path. The bug affects any script outside ASCII whose codepoints land in those low-bit ranges — Cyrillic, Greek, Hebrew, Arabic, CJK, etc.

Zed version and system specs

Zed 1.0.1, macOS 15.7.4

Attach Zed log file

n/a

Relevant Zed settings / Keymap

n/a

(for AI issues) Model provider details

n/a

WSL

No