utils.R

# copy objects in one environment to the other
copy_env = function(from, to, keys = ls(envir = from, all.names = TRUE)) {
  if (identical(from, to)) return()
  for (i in keys) assign(i, get(i, envir = from, inherits = FALSE), envir = to)
}


knit_counter = function(init = 0L) {
  n = init
  function(reset = FALSE) {
    if (reset) return(n <<- init)
    n <<- n + 1L
    n - 1L
  }
}

plot_counter = knit_counter(1L)
shot_counter = knit_counter(1L)
chunk_counter = knit_counter(1L)

# a vectorized and better version than evaluate:::line_prompt
line_prompt = function(x, prompt = getOption('prompt'), continue = getOption('continue')) {
  # match a \n, then followed by any character (use zero width assertion)
  paste0(prompt, gsub('(?<=\n)(?=.|\n)', continue, x, perl = TRUE))
}

# add a prefix to output
comment_out = function(x, prefix = '##', which = TRUE, newline = TRUE) {
  x = gsub('[\n]{2,}$', '\n', x)
  if (newline) x = gsub('([^\n]|^)$', '\\1\n', x)  # add \n if not exists
  if (is.null(prefix) || !nzchar(prefix) || is.na(prefix)) return(x)
  prefix = paste(prefix, '')
  x = gsub(' +([\n]*)$', '\\1', x)
  x[which] = line_prompt(x[which], prompt = prefix, continue = prefix)
  x
}

# assign string in comments to a global variable
comment_to_var = function(x, varname, pattern, envir) {
  if (grepl(pattern, x)) {
    assign(varname, sub(pattern, '', x), envir = envir)
    return(TRUE)
  }
  FALSE
}

is_blank = function(x) {
  if (length(x)) all(grepl('^\\s*$', x)) else TRUE
}
valid_path = function(prefix, label) {
  if (length(prefix) == 0L || is.na(prefix) || prefix == 'NA') prefix = ''
  paste0(prefix, label)
}

# define a color variable in TeX
color_def = function(col, variable = 'shadecolor') {
  if (all(is.na(col))) return('')  # no LaTeX code when color is NA
  x = if (length(col) == 1L) sc_split(col) else col
  if ((n <- length(x)) != 3L) {
    if (n == 1L) x = drop(col2rgb(x) / 255) else {
      x = switch(variable, shadecolor = rep(.97, 3), fgcolor = rep(0, 3))
      warning("the color '", col, "' is invalid;",
              'using default color...',
              'see https://yihui.org/knitr/options')
    }
  }
  if (length(x) != 3L) stop('invalid color:', col)
  if (is.numeric(x)) x = round(x, 3L)
  outdec = options(OutDec = '.'); on.exit(options(outdec))
  sprintf('\\definecolor{%s}{rgb}{%s, %s, %s}', variable, x[1], x[2], x[3])
}

# split by semicolon or colon
sc_split = function(string) {
  if (is.call(string)) string = eval(string)
  if (is.numeric(string) || length(string) != 1L) return(string)
  trimws(stringr::str_split(string, ';|,')[[1]])
}

# extract LaTeX packages for tikzDevice
set_preamble = function(input, patterns = knit_patterns$get()) {
  if (!out_format('latex')) return()
  .knitEnv$tikzPackages = .knitEnv$bibliography = NULL
  if (length(db <- patterns$document.begin) != 1L) return()  # no \begin{document} pattern
  if (length(hb <- patterns$header.begin) != 1L) return()  # no \documentclass{} pattern
  idx2 = grep(db, input)[1]
  if (is.na(idx2) || idx2 < 2L) return()
  idx1 = grep(hb, input)[1]
  if (is.na(idx1) || idx1 >= idx2) return()
  txt = one_string(input[idx1:(idx2 - 1L)])  # rough preamble
  idx = stringr::str_locate(txt, hb)  # locate documentclass
  options(tikzDocumentDeclaration = stringr::str_sub(txt, idx[, 1L], idx[, 2L]))
  preamble = pure_preamble(split_lines(stringr::str_sub(txt, idx[, 2L] + 1L)), patterns)
  .knitEnv$tikzPackages = c(.header.sweave.cmd, preamble, '\n')
  .knitEnv$bibliography = grep('^\\\\bibliography.+', input, value = TRUE)
}
# filter out code chunks from preamble if they exist (they do in LyX/Sweave)
pure_preamble = function(preamble, patterns) {
  res = split_file(lines = preamble, set.preamble = FALSE, patterns) # should avoid recursion
  if (!parent_mode()) {
    # when not in parent mode, just return normal texts and skip code
    return(unlist(res))
  }
  owd = setwd(input_dir()); on.exit(setwd(owd))
  progress = opts_knit$get('progress')  # suppress printing of blocks and texts
  opts_knit$set(progress = FALSE); on.exit(opts_knit$set(progress = progress), add = TRUE)
  # run the code in the preamble
  sapply(res, if (opts_knit$get('tangle')) process_tangle else process_group)
}

#' Specify the parent document of child documents
#'
#' This function extracts the LaTeX preamble of the parent document to use for
#' the child document, so that the child document can be compiled as an
#' individual document.
#'
#' When the preamble of the parent document also contains code chunks and inline
#' R code, they will be evaluated as if they were in this child document. For
#' examples, when \pkg{knitr} hooks or other options are set in the preamble of
#' the parent document, it will apply to the child document as well.
#' @param parent Path to the parent document, relative to the current child
#'   document.
#' @return The preamble is extracted and stored to be used later when the
#'   complete output is written.
#' @note Obviously this function is only useful when the output format is LaTeX.
#'   This function only works when the child document is compiled in a
#'   standalone mode using \code{\link{knit}()} (instead of being called in
#'   \code{\link{knit_child}()}); when the parent document is compiled, this
#'   function in the child document will be ignored.
#' @references \url{https://yihui.org/knitr/demo/child/}
#' @export
#' @examples ## can use, e.g. \Sexpr{set_parent('parent_doc.Rnw')} or
#'
#' # <<setup-child, include=FALSE>>=
#'
#' # set_parent('parent_doc.Rnw')
#'
#' # @@
set_parent = function(parent) {
  if (child_mode()) return(invisible(NULL)) # quit if in child mode
  opts_knit$set(parent = TRUE)
  set_preamble(read_utf8(parent))
  invisible(NULL)
}

# whether to write results as-is?
output_asis = function(x, options) {
  is_blank(x) || options$results == 'asis'
}

# the working directory: use root.dir if specified, otherwise the dir of the
# input file unless knitr.use.cwd = TRUE
input_dir = function() {
  root = opts_knit$get('root.dir')
  root %n% (if (!getOption('knitr.use.cwd', FALSE)) .knitEnv$input.dir) %n% '.'
}

# eval expr under the input dir
in_input_dir = function(expr) in_dir(input_dir(), expr)

# detect if running on CRAN (assuming that CRAN does not set CI or
# NOT_CRAN=true); or set R_CRANDALF=true (is cran) or false (not cran)
is_cran = function() {
  x = Sys.getenv('R_CRANDALF', NA)
  if (!is.na(x)) tolower(x) == 'true' else {
    !any(tolower(Sys.getenv(c('CI', 'NOT_CRAN'))) == 'true')
  }
}

is_cran_check = function() {
  is_cran() && is_R_CMD_check()
}

is_bioc = function() Sys.getenv('BBS_HOME') != ''

# round a number to getOption('digits') decimal places by default, and format()
# it using significant digits if the option knitr.digits.signif = TRUE
round_digits = function(x) {
  if (getOption('knitr.digits.signif', FALSE)) format(x) else {
    as.character(round(x, getOption('digits')))
  }
}

# scientific notation in TeX, HTML and reST
format_sci_one = function(
  x, format = 'latex', times = getOption('knitr.inline.times', '\\times ')
) {

  if (!(class(x)[1] == 'numeric') || is.na(x) || x == 0) return(as.character(x))

  if (is.infinite(x)) {
    return(
      switch(format, latex = {
        sprintf("%s\\infty{}", ifelse(x < 0, "-", ""))
      }, html = {
        sprintf("%s&infin;", ifelse(x < 0, "-", ""))
      }, as.character(x)))
  }

  if (abs(lx <- floor(log10(abs(x)))) < getOption('scipen') + 4L)
    return(round_digits(x)) # no need sci notation

  b = round_digits(x / 10^lx)
  b[b %in% c(1, -1)] = ''

  switch(format, latex = {
    sci_notation('%s%s10^{%s}', b, times, lx)
  },
  html = sci_notation('%s%s10<sup>%s</sup>', b, ' &times; ', lx),
  md   = sci_notation('%s%s10^%s^', b, '&times; ', lx),
  rst  = {
    # if AsIs, use the :math: directive
    if (inherits(x, 'AsIs')) {
      s = sci_notation('%s%s10^{%s}', b, times, lx)
      sprintf(':math:`%s`', s)
    } else {
      # This needs the following line at the top of the file to define |times|
      # .. include <isonum.txt>
      sci_notation('%s%s10 :sup:`%s`', b, ' |times| ', lx)
    }
  }, as.character(x))
}

sci_notation = function(format, base, times, power) {
  sprintf(format, base, ifelse(base == '', '', times), power)
}

# vectorized version of format_sci_one()
format_sci = function(x, ...) {
  if (inherits(x, 'roman')) return(as.character(x))
  vapply(x, format_sci_one, character(1L), ..., USE.NAMES = FALSE)
}

# is tikz device without externalization?
is_tikz_dev = function(options) {
  'tikz' %in% options$dev && !options$external
}

tikz_dict = function(path) {
  paste(sans_ext(basename(path)), 'tikzDictionary', sep = '-')
}

# initially for compatibility with Sweave and old beta versions of knitr
# but now also place to tweak default options
fix_options = function(options) {
  options = as.strict_list(options)

  # if you want to use subfloats, fig.show must be 'hold'
  if (length(options$fig.subcap)) options$fig.show = 'hold'
  # if the animation hook has been set, fig.show must be 'animate'
  if (!is.null(options$animation.hook)) options$fig.show = 'animate'
  # the default device NULL is not valid; use pdf is not set
  if (is.null(options$dev)) options$dev = 'pdf'
  # FALSE means hide for options$results
  if (isFALSE(options$results)) options$results = 'hide'

  # the figure/cache filenames may contain UTF-8 chars, which won't work on
  # Windows, e.g. png() fails if filename contains UTF-8 chars (must use native
  # encoding); in rmarkdown::render(input), basename(input) may implicitly
  # convert the encoding of the filename `input` to UTF-8
  options$fig.path = native_encode(options$fig.path)
  options$cache.path = native_encode(options$cache.path)

  # cache=TRUE -> 3; FALSE -> 0
  if (is.logical(options$cache)) options$cache = options$cache * 3
  if (options$engine != 'R') {
    # non-R code should not use cache = 1, 2
    options$cache = (options$cache > 0) * 3
    # error numbers only make sense to R engine, so 0 -> TRUE; 1, 2 -> FALSE
    if (is.numeric(options$error)) options$error = options$error == 0
  }

  options$eval = unname(options$eval)

  # handle aspect ratio/figure dimensions; give priority to fig.dim, with
  # warning (seems relatively more likely as user's intention in case when both
  # present) if aspect ratio is specified, calculate figure height
  fix_asp = is.numeric(options$fig.asp)
  if (length(options$fig.dim) == 2L) {
    if (fix_asp) warning('The chunk option fig.asp is ignored since fig.dim is provided.')
    options$fig.width = options$fig.dim[1L]; options$fig.height = options$fig.dim[2L]
  } else {
    if (fix_asp) options$fig.height = options$fig.width * options$fig.asp
  }

  # out.[width|height].px: unit in pixels for sizes
  for (i in c('width', 'height')) {
    options[[sprintf('out.%s.px', i)]] = options[[o <- sprintf('out.%s', i)]] %n%
      (options[[sprintf('fig.%s', i)]] * options$dpi)
    # turn x% to x/100\linewidth or \textheight
    if (is_latex_output()) options[o] = list(latex_percent_size(options[[o]], i))
  }
  # for Retina displays, increase physical size, and decrease output size
  if (is.numeric(r <- options$fig.retina) && r != 1) {
    if (is.null(options[['out.width']])) {
      options$out.width = options$fig.width * options$dpi
    }
    options$dpi = options$dpi * r
  } else {
    options$fig.retina = 1
  }

  # deal with aliases: a1 is real option; a0 is alias
  if (length(a1 <- opts_knit$get('aliases')) && length(a0 <- names(a1))) {
    for (i in seq_along(a1)) {
      # use alias only if the name exists in options
      if (a0[i] %in% names(options)) options[[a1[i]]] = options[[a0[i]]]
    }
  }

  # adjust some options when collapse is TRUE
  if (options$collapse) {
    options[unlist(lapply(
      c('class.', 'attr.'), paste0, c('output', 'message', 'warning', 'error')
    ))] = NULL
  }

  # change default of value conditionally
  if (identical(options$strip.white, I(TRUE)))
    options$strip.white = !options$collapse

  options
}

#' Check the current input and output type
#'
#' The function \code{is_latex_output()} returns \code{TRUE} when the output
#' format is LaTeX; it works for both \file{.Rnw} and R Markdown documents (for
#' the latter, the two Pandoc formats \code{latex} and \code{beamer} are
#' considered LaTeX output). The function \code{is_html_output()} only works for
#' R Markdown documents and will test for several Pandoc HTML based output
#' formats (by default, these formats are considered as HTML formats:
#' \code{c('markdown', 'epub', 'epub2', 'html', 'html4', 'html5', 'revealjs', 's5',
#' 'slideous', 'slidy', 'gfm')}).
#'
#' The function \code{pandoc_to()} returns the Pandoc output format, and
#' \code{pandoc_from()} returns Pandoc input format. \code{pandoc_to(fmt)}
#' allows to check the current output format against a set of format names. Both
#' are to be used with R Markdown documents.
#'
#' These functions may be useful for conditional output that depends on the
#' output format. For example, you may write out a LaTeX table in an R Markdown
#' document when the output format is LaTeX, and an HTML or Markdown table when
#' the output format is HTML. Use \code{pandoc_to(fmt)} to test a more specific
#' Pandoc format.
#'
#' Internally, the Pandoc output format of the current R Markdown document is
#' stored in \code{knitr::\link{opts_knit}$get('rmarkdown.pandoc.to')}, and the
#' Pandoc input format in
#' \code{knitr::\link{opts_knit}$get('rmarkdown.pandoc.from')}
#'
#' @note See available Pandoc formats, in
#'   \href{https://pandoc.org/MANUAL.html}{Pandoc's Manual}
#' @rdname output_type
#' @export
#' @examples
#' # check for output formats type
#' knitr::is_latex_output()
#' knitr::is_html_output()
#' knitr::is_html_output(excludes = c('markdown', 'epub'))
#' # Get current formats
#' knitr::pandoc_from()
#' knitr::pandoc_to()
#' # Test if current output format is 'docx'
#' knitr::pandoc_to('docx')
is_latex_output = function() {
  out_format('latex') || pandoc_to(c('latex', 'beamer'))
}

#' @param fmt A character vector of output formats to be checked against. If not
#'   provided, \code{is_html_output()} uses \code{pandoc_to()}, and
#'   \code{pandoc_to()} returns the output format name.
#' @param excludes A character vector of output formats that should not be
#'   considered as HTML format. Options are: markdown, epub, epub2, html, html4, html5,
#'   revealjs, s5, slideous, slidy, and gfm.
#' @rdname output_type
#' @export
is_html_output = function(fmt = pandoc_to(), excludes = NULL) {
  if (length(fmt) == 0) return(FALSE)
  if (grepl('^markdown', fmt)) fmt = 'markdown'
  if (fmt == 'epub3') fmt = 'epub'
  fmts = c('markdown', 'epub', 'epub2', 'html', 'html4', 'html5', 'revealjs', 's5', 'slideous', 'slidy', 'gfm')
  fmt %in% setdiff(fmts, excludes)
}

#' @rdname output_type
#' @export
pandoc_to = function(fmt) {
  # rmarkdown sets an option for the Pandoc output format from markdown
  to = opts_knit$get('rmarkdown.pandoc.to')
  if (missing(fmt)) to else !is.null(to) && (to %in% fmt)
}

#' @rdname output_type
#' @export
pandoc_from = function() {
  # rmarkdown's input format, obtained from a package option set by rmarkdown
  opts_knit$get('rmarkdown.pandoc.from') %n% 'markdown'
}

# turn percent width/height to LaTeX unit, e.g. out.width = 30% -> .3\linewidth
latex_percent_size = function(x, which = c('width', 'height')) {
  if (!is.character(x)) return(x)
  i = grep('^[0-9.]+%$', x)
  if (length(i) == 0) return(x)
  xi = as.numeric(sub('%$', '', x[i]))
  if (any(is.na(xi))) return(x)
  which = match.arg(which)
  x[i] = paste0(
    formatC(xi / 100, decimal.mark = '.'),
    if (which == 'width') '\\linewidth' else '\\textheight'
  )
  x
}

# eval options as symbol/language objects
eval_lang = function(x, envir = knit_global()) {
  if (!is.symbol(x) && !is.language(x)) return(x)
  in_input_dir(eval(x, envir = envir))
}

# check latex packages; if not exist, copy them over to ./
test_latex_pkg = function(name, path) {
  res = try_silent(system(sprintf('%s %s.sty', kpsewhich(), name), intern = TRUE))
  if (inherits(res, 'try-error') || !length(res)) {
    warning("unable to find LaTeX package '", name, "'; will use a copy from knitr")
    file.copy(path, '.')
  }
}

# get child and parent mode
child_mode = function() opts_knit$get('child')
parent_mode = function() opts_knit$get('parent')

# return the output format, or if current format is in specified formats
out_format = function(x) {
  fmt = opts_knit$get('out.format')
  if (missing(x)) fmt else !is.null(fmt) && (fmt %in% x)
}

# tempfile under the current working directory
wd_tempfile = function(...) basename(tempfile(tmpdir = '.', ...))

pandoc_fragment = function(text, to = pandoc_to(), from = pandoc_from()) {
  if (length(text) == 0) return(text)
  f1 = wd_tempfile('pandoc', '.md'); f2 = wd_tempfile('pandoc')
  on.exit(unlink(c(f1, f2)), add = TRUE)
  write_utf8(text, f1)
  rmarkdown::pandoc_convert(f1, to, from, f2, options = if (is_html_output(to)) '--mathjax')
  file_string(f2)
}

#' Path for figure files
#'
#' The filename of figure files is the combination of options \code{fig.path}
#' and \code{label}. This function returns the path of figures for the current
#' chunk by default.
#' @param suffix A filename suffix; if it is non-empty and does not
#'   contain a dot \code{.}, it will be treated as the filename extension (e.g.
#'   \code{png} will be used as \code{.png})
#' @param options A list of options; by default the options of the current chunk.
#' @param number The current figure number. The default is the internal chunk option
#'   \code{fig.cur}, if this is available.
#' @return A character vector of the form \file{fig.path-label-i.suffix}.
#' @note When there are special characters (not alphanumeric or \samp{-} or
#'   \samp{_}) in the path, they will be automatically replaced with \samp{_}.
#'   For example, \file{a b/c.d-} will be sanitized to \file{a_b/c_d-}. This
#'   makes the filenames safe to LaTeX.
#' @export
#' @examples fig_path('.pdf', options = list(fig.path='figure/abc-', label='first-plot'))
#' fig_path('.png', list(fig.path='foo-', label='bar'), 1:10)
fig_path = function(suffix = '', options = opts_current$get(), number) {
  if (suffix != '' && !grepl('[.]', suffix)) suffix = paste0('.', suffix)
  if (missing(number)) number = options$fig.cur %n% 1L
  if (!is.null(number)) suffix = paste0('-', number, suffix)
  path = valid_path(options$fig.path, options$label)
  (if (out_format(c('latex', 'sweave', 'listings'))) sanitize_fn else
    paste0)(path, suffix)
}
# sanitize filename for LaTeX
sanitize_fn = function(path, suffix = '') {
  if (grepl('[^~:_./\\[:alnum:]-]', path)) {
    warning('replaced special characters in figure filename "', path, '" -> "',
            path <- gsub('[^~:_./\\[:alnum:]-]', '_', path), '"')
  }
  # replace . with _ except ../ and ./
  s = strsplit(path, '[/\\\\]')[[1L]]
  i = (s != '.') & (s != '..') & grepl('\\.', s)
  if (any(i)) {
    s[i] = gsub('\\.', '_', s[i])
    path = paste(s, collapse = '/')
    warning('dots in figure paths replaced with _ ("', path, '")')
  }
  paste0(path, suffix)
}

#' Obtain the figure filenames for a chunk
#'
#' Given a chunk label, the figure file extension, the figure number(s), and the
#' chunk option \code{fig.path}, return the filename(s).
#'
#' This function can be used in an inline R expression to write out the figure
#' filenames without hard-coding them. For example, if you created a plot in a
#' code chunk with the label \code{foo} and figure path \file{my-figure/}, you
#' are not recommended to use hard-coded figure paths like
#' \samp{\includegraphics{my-figure/foo-1.pdf}} (in \file{.Rnw} documents) or
#' \samp{![](my-figure/foo-1.png)} (R Markdown) in your document. Instead, you
#' should use \samp{\\Sexpr{fig_chunk('foo', 'pdf')}} or \samp{![](`r
#' fig_chunk('foo', 'png')`)}.
#'
#' You can generate plots in a code chunk but not show them inside the code
#' chunk by using the chunk option \code{fig.show = 'hide'}. Then you can use
#' this function if you want to show them elsewhere.
#' @param label The chunk label.
#' @param ext The figure file extension, e.g. \code{png} or \code{pdf}.
#' @param number The figure number (by default \code{1}).
#' @param fig.path Passed to \code{\link{fig_path}}. By default, the chunk
#'   option \code{fig.path} is used.
#' @return A character vector of filenames.
#' @export
#' @examples library(knitr)
#' fig_chunk('foo', 'png')
#' fig_chunk('foo', 'pdf')
#' fig_chunk('foo', 'svg', 2)  # the second plot of the chunk foo
#' fig_chunk('foo', 'png', 1:5)  # if the chunk foo produced 5 plots
fig_chunk = function(label, ext = '', number, fig.path = opts_chunk$get('fig.path')) {
  fig_path(ext, list(fig.path = fig.path, label = label), number)
}

#' The global environment in which code chunks are evaluated
#'
#' This function makes the environment of a code chunk accessible inside a
#' chunk.
#'
#' It returns the \code{envir} argument of \code{\link{knit}}, e.g. if we call
#' \code{\link{knit}()} in the global environment, \code{knit_global()} returns
#' R's global environment by default. You can call functions like
#' \code{\link{ls}()} on this environment.
#' @export
knit_global = function() {
  .knitEnv$knit_global %n% globalenv()
}

# Indents a Block
#  Input
#     "library(ggplot2)\nqplot(wt, mpg, data = mtcars)"
#  Output
#          library(ggplot2)
#          qplot(wt, mpg, data  = mtcars)
indent_block = function(block, spaces = '    ') {
  if (is.null(block) || !any(nzchar(block))) return(rep(spaces, length(block)))
  if (spaces == '') return(block)
  line_prompt(block, spaces, spaces)
}

# print knitr logs
print_knitlog = function() {
  if (!opts_knit$get('verbose') || child_mode() || !length(klog <- knit_log$get(drop = FALSE)))
    return()

  for (i in unlist(klog, use.names = FALSE)) {
    cat(sub('\n+$', '', i), '\n\n')
    if (length(code <- knit_code$get(sub('^Chunk ([^:]+):\n.*', '\\1', i))))
      cat(code, sep = '\n')
  }
  cat('\nNumber of messages:\n')
  print(sapply(klog, length))
}

# count the number of lines
line_count = function(x) stringr::str_count(x, '\n') + 1L

has_package = function(pkg) loadable(pkg, FALSE)
has_packages = function(pkgs) {
  for (p in pkgs) if (!has_package(p)) return(FALSE)
  TRUE
}

# if LHS is NULL, return the RHS
`%n%` = function(x, y) if (is.null(x)) y else x

# merge elements of y into x with the same names
merge_list = function(x, y) {
  x[names(y)] = y
  x
}

# find a token in a template, and replace it with a value
insert_template = function(text, token, value, ignore = FALSE) {
  if (is.null(value)) return(text)
  i = grep(token, text); n = length(i)
  if (n > 1) stop("There are multiple tokens in the template: '", token, "'")
  if (n == 0) {
    if (ignore) return(text)
    stop("Couldn't find the token '", token, "' in the template.")
  }
  append(text, value, i)
}

# paths of all figures
all_figs = function(options, ext = options$fig.ext, num = options$fig.num) {
  unlist(lapply(ext, fig_path, options = options, number = seq_len(num)))
}

# evaluate an expression in a diretory and restore wd after that
in_dir = function(dir, expr) {
  if (!is.null(dir)) {
    owd = setwd(dir); on.exit(setwd(owd))
  }
  wd1 = getwd()
  res = expr
  wd2 = getwd()
  if (wd1 != wd2) warning(
    'You changed the working directory to ', wd2, ' (probably via setwd()). ',
    'It will be restored to ', wd1, '. See the Note section in ?knitr::knit'
  )
  res
}

# evaluate under the base.dir
in_base_dir = function(expr) {
  d = opts_knit$get('base.dir')
  if (is.character(d) && !file_test('-d', d)) dir.create(d, recursive = TRUE)
  in_dir(d, expr)
}

# escape special LaTeX characters
escape_latex = function(x, newlines = FALSE, spaces = FALSE) {
  x = gsub('\\\\', '\\\\textbackslash', x)
  x = gsub('([#$%&_{}])', '\\\\\\1', x)
  x = gsub('\\\\textbackslash', '\\\\textbackslash{}', x)
  x = gsub('~', '\\\\textasciitilde{}', x)
  x = gsub('\\^', '\\\\textasciicircum{}', x)
  if (newlines) x = gsub('(?<!\n)\n(?!\n)', '\\\\\\\\', x, perl = TRUE)
  # when there are consecutive spaces, escape each of them except the first one
  if (spaces) x = gsub('(?<= ) ', '\\\\ ', x, perl = TRUE)
  x
}

# escape special HTML chars
escape_html = highr:::escape_html

#' Read source code from R-Forge
#'
#' This function reads source code from the SVN repositories on R-Forge.
#' @param path Relative path to the source script on R-Forge.
#' @param project Name of the R-Forge project.
#' @param extra Extra parameters to be passed to the URL (e.g. \code{extra =
#'   '&revision=48'} to check out the source of revision 48).
#' @return A character vector of the source code.
#' @author Yihui Xie and Peter Ruckdeschel
#' @export
#' @examplesIf interactive()
#' library(knitr)
#' # relies on r-forge.r-project.org being accessible
#' read_rforge('rgl/R/axes.R', project = 'rgl')
#' read_rforge('rgl/R/axes.R', project = 'rgl', extra='&revision=519')
read_rforge = function(path, project, extra = '') {
  base = 'http://r-forge.r-project.org/scm/viewvc.php/*checkout*/pkg'
  read_utf8(sprintf('%s/%s?root=%s%s', base, path, project, extra))
}

split_lines = function(x) xfun::split_lines(x)

# if a string is encoded in UTF-8, convert it to native encoding
native_encode = function(x, to = '') {
  from = if (any(Encoding(x) == 'UTF-8')) 'UTF-8' else ''
  x2 = iconv(x, from, to)
  if (!any(is.na(x2))) return(x2)  # use conversion only if it succeeds
  warning('some characters may not work under the current locale')
  x
}

is_native_enc = function(x) {
  x == 'native.enc' || x == ''
}

is_utf8 = function(x) {
  !any(is.na(iconv(x, 'UTF-8', 'UTF-8')))
}

#' Wrap long lines in Rmd files
#'
#' This function wraps long paragraphs in an R Markdown file. Other elements are
#' not wrapped: the YAML preamble, fenced code blocks, section headers and
#' indented elements. The main reason for wrapping long lines is to make it
#' easier to review differences in version control.
#' @param file The input Rmd file.
#' @param width The expected line width.
#' @param text A character vector of text lines, as an alternative to \code{file}. If
#'   \code{text} is not \code{NULL}, \code{file} is ignored.
#' @param backup Path to back up the original file in case anything goes
#'   wrong. If set to \code{NULL}, no backup is made. The default value is constructed
#'   from \code{file} by adding \code{__} before the base filename.
#' @return If \code{file} is provided, it is overwritten; if \code{text} is
#'   provided, a character vector is returned.
#' @note Currently it does not wrap blockquotes or lists (ordered or unordered).
#'   This feature may or may not be added in the future.
#' @export
#' @examples wrap_rmd(text = c('```', '1+1', '```', '- a list item', '> a quote', '',
#' paste(rep('this is a normal paragraph', 5), collapse = ' ')))
wrap_rmd = function(file, width = 80, text = NULL, backup) {
  x = if (is.null(text)) read_utf8(file) else split_lines(text)
  x = strip_white(x)  # strip blank lines in the beginning and end
  if ((n <- length(x)) <= 1L) return(x)  # are you kidding?
  idx = NULL  # collect the lines to exclude from wrapping
  i = grep('^---$', x)  # yaml preamble
  if (length(i) > 1 && i[1L] == 1L) idx = c(idx, i[1L]:i[2L])
  i = grep('^(```|\\{% (end|)highlight [a-z ]*%\\}|</?script.*>)', x)  # code blocks
  if (length(i)) {
    if (length(i) %% 2L != 0L) stop('markers for code blocks must be paired up')
    idx = c(idx, unlist(apply(matrix(i, ncol = 2L, byrow = TRUE), 1L,
                              function(z) z[1L]:z[2L])))
  }
  # section headers, indented code blocks and latex math
  idx = c(idx, grep('^(#|===|---|    |\t)', x))
  # blank lines
  idx = c(idx, grep('^\\s*$', x))
  # TODO: this is naive -- I treat a line as (a part of) a normal paragraph if
  # it does not start with a space, or > (blockquotes) or -, *, 1. (lists), and
  # only wrap paragraphs
  idx = c(idx, grep('^\\s*( |> |- |\\* |\\d+ )', x))
  idx = unique(idx)
  if (length(idx) == n) return(x)  # no need to wrap anything

  i = logical(n); i[idx] = TRUE; r = rle(i)
  n = length(r$lengths); txt = vector('list', n); j = c(0L, cumsum(r$lengths))
  for (i in seq_len(n)) {
    block = x[seq(j[i] + 1L, j[i + 1])]
    txt[[i]] = if (r$value[i]) {
      # those lines not to be wrapped
      gsub('\\s+$', '', block)  # strip pending spaces
    } else {
      strwrap(one_string(block), width)
    }
  }
  txt = unlist(txt)
  if (is.null(text)) {
    if (missing(backup)) backup = file.path(dirname(file), paste0('__', basename(file)))
    if (!is.null(backup)) file.copy(file, backup, overwrite = TRUE)
    write_utf8(txt, file)
  } else txt
}

# change the default device to an appropriate device when the output is html
# (e.g. markdown, reST, AsciiDoc)
set_html_dev = function() {
  # only change if device has not been set
  if (!is.null(opts_chunk$get('dev'))) return()
  # in some cases, png() does not work (e.g. options('bitmapType') == 'Xlib' on
  # headless servers); use svg then
  opts_chunk$set(dev = if (dev_available('png')) 'png' else 'svg')
}

# locate kpsewhich especially for Mac OS because /usr/texbin may not be in PATH
kpsewhich = function() {
  if (Sys.info()['sysname'] != 'Darwin' || !file.exists(x <- '/usr/texbin/kpsewhich')
      || nzchar(Sys.which('kpsewhich')))
    'kpsewhich' else x
}

# check if a utility exists; if it does, save its availability in opts_knit
has_utility = function(name, package = name) {
  name2 = paste('util', name, sep = '_')  # e.g. util_pdfcrop
  if (is.logical(yes <- opts_knit$get(name2))) return(yes)
  # a special case: use tools::find_gs_cmd() to find ghostscript
  if (name == 'ghostscript') name = tools::find_gs_cmd()
  yes = nzchar(Sys.which(name))
  if (!yes) warning(package, ' not installed or not in PATH')
  opts_knit$set(setNames(list(yes), name2))
  yes
}

#' Query the current input filename
#'
#' Returns the name of the input file passed to \code{\link{knit}()}.
#' @param dir Boolean; whether to prepend the current working directory to the file path,
#'   i.e. whether to return an absolute path or a relative path.
#' @return A character string, if this function is called inside an input
#'   document. Otherwise \code{NULL}.
#' @export
current_input = function(dir = FALSE) {
  input = knit_concord$get('infile')
  outwd = opts_knit$get('output.dir')
  if (is.null(input)) return()
  if (dir) {
    if (is.null(outwd)) {
      warning('Cannot determine the directory of the input document')
      dir = FALSE
    }
  }
  if (!dir) return(basename(input))
  if (is_abs_path(input)) input else file.path(outwd, input)
}

# import output handlers from evaluate
default_handlers = evaluate:::default_output_handler
# change the value handler in evaluate default handlers
knit_handlers = function(fun, options) {
  if (!is.function(fun)) fun = function(x, ...) {
    res = withVisible(knit_print(x, ...))
    # indicate the htmlwidget result with a special class so we can attach
    # the figure caption to it later in sew.knit_asis
    if (inherits(x, 'htmlwidget'))
      class(res$value) = c(class(res$value), 'knit_asis_htmlwidget')
    if (res$visible) res$value else invisible(res$value)
  }
  if (length(formals(fun)) < 2)
    stop("the chunk option 'render' must be a function of the form ",
         "function(x, options) or function(x, ...)")
  merge_list(default_handlers, list(
    value = function(x, visible) {
      if (visible) fun(x, options = options)
    },
    calling_handlers = options$calling.handlers
  ))
}

# is the inst dir under . or ..? differs in R CMD build/INSTALL and devtools/roxygen2
inst_dir = function(...) {
  p = file.path(c('..', '.'), 'inst', ...)
  p[file.exists(p)]
}

# a restricted version of is.numeric (e.g. do not treat chron::chron() as
# numeric since their behavior may be somewhat unpredictable, e.g. through
# round(), #1118); see #1396 for difftime
is_numeric = function(x) {
  class(x)[1] %in% c('numeric', 'integer', 'difftime', 'complex')
}

# create \label{x} or (\#x); the latter is current an internal hack for bookdown
create_label = function(..., latex = FALSE) {
  if (isTRUE(opts_knit$get('bookdown.internal.label'))) {
    lab1 = '(\\#'; lab2 = ')'
  } else if (latex) {
    lab1 = '\\label{'; lab2 = '}'
  } else {
    return('')  # we don't want the label at all
  }
  paste(c(lab1, ..., lab2), collapse = '')
}

#' Combine multiple words into a single string
#'
#' When a value from an inline R expression is a character vector of multiple
#' elements, we may want to combine them into a phrase like \samp{a and b}, or
#' \code{a, b, and c}. That is what this a helper function does.
#'
#' If the length of the input \code{words} is smaller than or equal to 1,
#' \code{words} is returned. When \code{words} is of length 2, the first word
#' and second word are combined using the \code{and} string, or if blank,
#' \code{sep} if is used. When the length is greater than 2, \code{sep} is used
#' to separate all words, and the \code{and} string is prepended to the last
#' word.
#' @param words A character vector.
#' @param sep Separator to be inserted between words.
#' @param and Character string to be prepended to the last word.
#' @param before,after A character string to be added before/after each word.
#' @param oxford_comma Whether to insert the separator between the last two
#'   elements in the list.
#' @return A character string marked by \code{xfun::\link{raw_string}()}.
#' @export
#' @examples combine_words('a'); combine_words(c('a', 'b'))
#' combine_words(c('a', 'b', 'c'))
#' combine_words(c('a', 'b', 'c'), sep = ' / ', and = '')
#' combine_words(c('a', 'b', 'c'), and = '')
#' combine_words(c('a', 'b', 'c'), before = '"', after = '"')
#' combine_words(c('a', 'b', 'c'), before = '"', after = '"', oxford_comma=FALSE)
combine_words = function(
  words, sep = ', ', and = ' and ', before = '', after = before, oxford_comma = TRUE
) {
  n = length(words); rs = xfun::raw_string
  if (n == 0) return(words)
  words = paste0(before, words, after)
  if (n == 1) return(rs(words))
  if (n == 2) return(rs(paste(words, collapse = if (is_blank(and)) sep else and)))
  if (oxford_comma && grepl('^ ', and) && grepl(' $', sep)) and = gsub('^ ', '', and)
  words[n] = paste0(and, words[n])
  # combine the last two words directly without the comma
  if (!oxford_comma) {
    words[n - 1] = paste0(words[n - 1:0], collapse = '')
    words = words[-n]
  }
  rs(paste(words, collapse = sep))
}

warning2 = function(...) warning(..., call. = FALSE)
stop2 = function(...) stop(..., call. = FALSE)

warn_options_unsupported = function(option, to) {
  warning2('Chunk option ', option, ' is not supported for ', to, ' output')
}

raw_markers = c('!!!!!RAW-KNITR-CONTENT', 'RAW-KNITR-CONTENT!!!!!')

#' @export
#' @rdname raw_output
extract_raw_output = function(text, markers = raw_markers) {
  r = sprintf('%s(.*?)%s', markers[1], markers[2])
  x = one_string(text)
  m = gregexpr(r, x)
  s = regmatches(x, m)
  n = length(s[[1]])
  if (n == 0) return(list(value = text, chunks = character()))

  chunks = tokens = character(n)
  for (i in seq_len(n)) {
    chunks[i] = sub(r, '\\1', s[[1]][i])
    tokens[i] = digest(chunks[i])
    s[[1]][i] = gsub(r, paste0(markers[1], tokens[i], markers[2]), s[[1]][i])
  }
  regmatches(x, m) = s

  list(value = x, chunks = setNames(chunks, tokens))
}

#' @export
#' @rdname raw_output
restore_raw_output = function(text, chunks, markers = raw_markers) {
  if ((n <- length(chunks)) == 0) return(text)
  text = enc2utf8(text); chunks = enc2utf8(chunks); tokens = names(chunks)
  for (i in seq_len(n)) {
    r = paste0(markers[1], tokens[i], markers[2])
    text = gsub(r, chunks[i], text, fixed = TRUE, useBytes = TRUE)
  }
  Encoding(text) = 'UTF-8'
  text
}

#' Mark character strings as raw output that should not be converted
#'
#' These functions provide a mechanism to protect the character output of R code
#' chunks. The output is annotated with special markers in \code{raw_output};
#' \code{extract_raw_output()} will extract raw output wrapped in the markers,
#' and replace the raw output with its MD5 digest; \code{restore_raw_output()}
#' will restore the MD5 digest with the original raw output.
#'
#' This mechanism is designed primarily for R Markdown pre/post-processors. In
#' an R code chunk, you generate \code{raw_output()} to the Markdown output. In
#' the pre-processor, you can \code{extract_raw_output()} from the Markdown
#' file, store the raw output and MD5 digests, and remove the actual raw output
#' from Markdown so Pandoc will never see it. In the post-processor, you can
#' read the Pandoc output (e.g., an HTML or RTF file), and restore the raw
#' output.
#' @param x The character vector to be protected.
#' @param markers A length-2 character vector to be used to wrap \code{x};
#'   see \code{knitr:::raw_markers} for the default value.
#' @param ... Arguments to be passed to \code{\link{asis_output}()}.
#' @param text For \code{extract_raw_output()}, the content of the input file
#'   (e.g. Markdown); for \code{restore_raw_output()}, the content of the output
#'   file (e.g. HTML generated by Pandoc from Markdown).
#' @param chunks A named character vector returned from
#'   \code{extract_raw_output()}.
#' @return For \code{extract_raw_output()}, a list of two components:
#'   \code{value} (the \code{text} with raw output replaced by MD5 digests) and
#'   \code{chunks} (a named character vector, of which the names are MD5 digests
#'   and values are the raw output). For \code{restore_raw_output()}, the
#'   restored \code{text}.
#' @export
#' @examples library(knitr)
#' out = c('*hello*', raw_output('<special>content</special> *protect* me!'), '*world*')
#' pre = extract_raw_output(out)
#' str(pre)
#' pre$value = gsub('[*]([^*]+)[*]', '<em>\\1</em>', pre$value)  # think this as Pandoc conversion
#' pre$value
#' # raw output was protected from the conversion (e.g. *protect* was not converted)
#' restore_raw_output(pre$value, pre$chunks)
raw_output = function(x, markers = raw_markers, ...) {
  asis_output(paste(c(markers[1], x, markers[2]), collapse = ''), ...)
}


#' Mark character strings as raw blocks in R Markdown
#'
#' Wraps content in a raw attribute block, which protects it from being escaped
#' by Pandoc. See \url{https://pandoc.org/MANUAL.html#generic-raw-attribute}.
#' Functions \code{raw_latex()} and \code{raw_html()} are shorthands of
#' \code{raw_block(x, 'latex')} and \code{raw_block(x, 'html')}, respectively.
#' @param x The character vector to be protected.
#' @param type The type of raw blocks (i.e., the Pandoc output format). If you
#'   are not sure about the Pandoc output format of your document, insert a code
#'   chunk \code{knitr:::pandoc_to()} and see what it returns after the document
#'   is compiled.
#' @param ... Arguments to be passed to \code{\link{asis_output}()}.
#' @export
#' @examples
#' knitr::raw_latex('\\emph{some text}')
raw_block = function(x, type = 'latex', ...) {
  if (!rmarkdown::pandoc_available('2.0.0')) warning('raw_block() requires Pandoc >= 2.0.0')
  x = c(sprintf('\n```{=%s}', type), x, '```\n')
  asis_output(one_string(x), ...)
}

#' @rdname raw_block
#' @export
raw_latex = function(x, ...) raw_block(x, 'latex', ...)

#' @rdname raw_block
#' @export
raw_html = function(x, ...) raw_block(x, 'html', ...)

optipng = function(...) {
  warning2('knitr:::optipng() has been deprecated; please use xfun::optipng()')
  xfun::optipng(...)
}

digest = function(x) {
  if (loadable('digest')) digest::digest(x) else digest2(x)
}

# compatible with digest::digest() but definitely slower because of file I/O
digest2 = function(x) {
  s = serialize(x, NULL, ascii = FALSE)
  if (length(s) > 14) s = s[-(1:14)]  # https://d.cosx.org/d/419804
  writeBin(s, f <- tempfile())
  on.exit(unlink(f), add = TRUE)
  unname(tools::md5sum(f))
}

# not removing the serialize() header (first few bytes)
digest3 = function(x) {
  f = tempfile(); on.exit(unlink(f), add = TRUE)
  s = file(f, open = 'wb')
  serialize(x, s, ascii = FALSE)
  close(s)
  unname(tools::md5sum(f))
}

# collapse by \n
one_string = function(x, ...) paste(x, ..., collapse = '\n')

# double quote a vector and combine by "; "
quote_vec = function(x, sep = '; ') paste0(sprintf('"%s"', x), collapse = sep)

# c(1, 1, 1, 2, 3, 3) -> c(1a, 1b, 1c, 2a, 3a, 3b)
make_unique = function(x) {
  if (length(x) == 0) return(x)
  x2 = make.unique(x)
  if (all(i <- x2 == x)) return(x)
  x2[i] = paste0(x2[i], '.0')
  i = as.numeric(sub('.*[.]([0-9]+)$', '\\1', x2)) + 1
  s = letters[i]
  s = ifelse(is.na(s), i, s)
  paste0(x, s)
}

#' Encode an image file to a data URI
#'
#' This function is the same as \code{xfun::\link{base64_uri}()} (only with a
#' different function name). It can encode an image file as a base64 string,
#' which can be used in the \code{img} tag in HTML.
#' @param f Path to the image file.
#' @return The data URI as a character string.
#' @author Wush Wu and Yihui Xie
#' @export
#' @references \url{https://en.wikipedia.org/wiki/Data_URI_scheme}
#' @examples uri = image_uri(file.path(R.home('doc'), 'html', 'logo.jpg'))
#' if (interactive()) {cat(sprintf('<img src="%s" />', uri), file = 'logo.html')
#' browseURL('logo.html') # you can check its HTML source
#' }
image_uri = function(f) xfun::base64_uri(f)

# Change all "[label](url)" to "label" except leave "`[label](url)`" as is
remove_urls = function(x) {
  # regex adapted from https://dev.to/mattkenefick/regex-convert-markdown-links-to-html-anchors-f7j
  gsub("(?<!`)\\[([^]]+)\\]\\(([^)]+)\\)(?!`)", "\\1", x, perl = TRUE)
}