Literate R Formats

Literate R Formats & Source Extraction in r-lib/lintr

When modifying how lintr extracts R source code from literate programming and multi-language documents (RMarkdown .Rmd, Quarto .qmd, Sweave .Rnw, or HTML .Rhtml), adhere to the following architectural conventions and extraction principles located primarily in R/extract.R.

1. Line Number Preservation via NA_character_ Masking

• Never drop or collapse lines during extraction: extract_r_source() extracts R code from literate documents by masking non-R lines (markdown text, YAML frontmatter, HTML tags, and skipped code chunks) with NA_character_.
• Preserve 1-to-1 line index mapping: Keeping non-source lines as NA_character_ ensures that line indices in the extracted character vector strictly match the 1-indexed line numbers of the original file (source_expression$lines). This guarantees that diagnostic line numbers (line_number) reported by linters map exactly to the user's source file.

2. Chunk Bounds & eval=FALSE Filtering

• Two-phase chunk boundary detection: get_chunk_positions(pattern, lines) determines chunk bounds by pairing opening pattern matches (filter_chunk_start_positions()) with closing pattern matches (filter_chunk_end_positions()), then retaining only blocks containing at least one line of inner code (ends - starts > 1L).
• Filter non-evaluated chunks: Filter out unevaluated chunks by inspecting both chunk header options and inside-chunk YAML/pipe options (#| eval: false, #| engine: python):

  is_eval_chunk <- function(start, end, lines, pattern) {
    header <- lines[start]
    params_src <- trimws(gsub(pattern$chunk.begin, "\\1", header))
    header_params <- safe_csv_options(params_src)
  
    code <- lines[(start + 1L):(end - 1L)]
    body_params <- tryCatch(
      suppressMessages(suppressWarnings(xfun::divide_chunk("r", code)))$options,
      error = \(e) NULL
    )
  
    engine <- body_params$engine %||% header_params$engine
    if (!is.null(engine) && tolower(as.character(engine[1L])) != "r") {
      return(FALSE)
    }
  
    eval_value <- body_params$eval %||% header_params$eval
    # nolint next: T_and_F_symbol_linter.
    if (identical(eval_value, quote(F))) {
      return(FALSE)
    }
    !isFALSE(eval_value)
  }
  

3. Engine Detection

• Detect R engines on opening lines: is_r_chunk_header() inspects chunk start lines to retain chunks targeting R engines ({r}, {R}, engine = "R") while skipping non-R engines ({python}, {extendr}, {ojs}, {mermaid}, {dot}, engine = "bash").

4. Multi-Format Testing Conventions

• Test across document families: When modifying R/extract.R, verify extraction behavior across all major literate families (.Rmd, .qmd, and .Rnw) inside tests/testthat/test-knitr_formats.R.
• Test zero-chunk documents: Always verify that documents containing zero chunks (such as plain markdown text or YAML header-only files) extract cleanly without throwing index out-of-bounds errors.