Bug Report: R Language Server emits spurious `comment` semantic tokens for first character of every line in `.qmd` files

Summary

When a .qmd file contains one or more R code chunks, the R Language Server (or Quarto VS Code extension) emits a comment semantic token covering the first character of every line in the document — including YAML front matter keys, markdown prose, headings, and LaTeX math delimiters. This colors all those first characters green (#008000), overriding the correct TextMate grammar colors.

Environment

• Editor: Cursor (VS Code-based)
• Quarto extension: quarto.quarto (version: 1.133.0)
• R languageserver: 0.3.18
• OS: Windows 11

Steps to Reproduce

1. Create a .qmd file with a standard YAML front matter and at least one R code chunk, e.g.:

---
title: "Test"
bibliography: refs.bib
format:
  html:
    toc: true
---

# Section heading

This is a paragraph of text.

$$
x = y + z
$$ {#eq-test}

```{r}
#| echo: false
print("hello")
```

2. Open the file in VS Code / Cursor with the Quarto extension installed and an R Language Server active.
3. Observe that the first character of every line (the t in title, the # in # Section heading, the T in This is a paragraph, the $ of $$, etc.) is colored green.
4. Use Developer: Inspect Editor Tokens and Scopes (Ctrl+Shift+P) with the cursor on any affected first character.

Observed Token Info (from Inspect Editor Tokens and Scopes)

Example: first character t of YAML key title:

language          yaml
standard token    String
foreground        #008000
semantic token    comment
  foreground      comment → { "foreground": "#008000" }
textmate scopes   entity.name.tag.yaml
                  string.unquoted.plain.out.yaml
                  meta.embedded.block.frontmatter
                  text.html.quarto
  foreground      meta.embedded.block.frontmatter entity.name.tag.yaml → { "foreground": "#800000" }

Example: first $ of a display math block $$:

language          quarto
standard token    Other
foreground        #008000
semantic token    comment
  foreground      comment → { "foreground": "#008000" }
textmate scopes   keyword.control.import.math.begin.quarto
                  markup.math.block.quarto
                  text.html.quarto
  foreground      keyword.control → { "foreground": "#0000FF" }

The semantic comment token (green, #008000) overrides the correct TextMate colors for every affected character.

Expected Behaviour

The comment semantic token should only be emitted for actual R comments (lines beginning with # inside R code chunks). It should not be applied to:

• YAML front matter keys
• Markdown prose
• Heading markers
• LaTeX math delimiters ($, $$)
• Any other non-R content

Actual Behaviour

The comment semantic token is applied to the first character of every line in the document whenever at least one R code chunk is present.

Key Observations

• The issue is absent when the .qmd file contains no R code chunks.
• Removing all R code chunks from an affected file immediately resolves the green coloring.
• A .qmd file with identical front matter and body but no R chunks shows no issue.
• The issue is therefore triggered by R Language Server activation, not by the YAML or Quarto language servers alone.
• This behaviour was not present in earlier versions; it appears to be a recent regression.

Workaround

Add the following to .vscode/settings.json in the workspace to suppress the spurious color while preserving correct R comment coloring:

{
  "editor.semanticTokenColorCustomizations": {
    "rules": {
      "comment:yaml": "#800000",
      "comment:quarto": "#000000",
      "comment:r": "#008000"
    }
  }
}

Suggested Fix

The R Language Server (or the Quarto extension's LSP bridge) should scope semantic token emission to the language region of each token — i.e., comment tokens should only be emitted within R code block ranges, not propagated to the surrounding markdown/YAML document.

Reproduce the issue in VS Code or Cursor using the provided .qmd example and inspect semantic tokens with Developer: Inspect Editor Tokens and Scopes. Trace comment-token emission in the R Language Server or Quarto extension's LSP bridge, focusing on R code chunk ranges. Done means comment tokens remain on actual R comments without coloring YAML, markdown, headings, or math content.