Added an option to tidy_source() to enforce a strict maximum line length.

[krivit]

This patch adds another option, width.strict= to tidy_source(), defaulting to FALSE. If TRUE, instead of calling deparse() with width.cutoff= directly, it calls a new function, strict_deparse(..., width.max), which performs a binary search for the highest width.cutoff= value to pass to deparse() such that the longest line in its output is, after stripping trailing whitespace, at most width.max= in length.

It seems to work quite well in most situations, though it still doesn't handle inline comments very well: the magic constant %InLiNe_IdEnTiFiEr% counts towards the line length, though it shouldn't. A possible solution might be to replace this magic constant with something much shorter, perhaps involving non-ASCII characters (which are legal in variable names in R, as far as I know) to reduce chances of a collision.

[krivit]

The behaviour with inline comments is now a little better. Unfortunately, if deparse() wraps a line just before a comment, it will get "merged" back in the unmasking code, potentially overflowing.

I've tried changing that code, but it doesn't work very well in knitr when output is interspersed with lines of code: the inline comment ends up going after the output.

[yihui]

Much appreciated! I'll review this tomorrow.

[pablo14]

I did a simpler (and less efficient) hack to tidy_block function. It tries different width until the desired sentence width is reached. Otherwise, it returns the original value.

It works for the pdf book I'm writing in most of the cases. I don't know if this approach may conflict with other use-cases.

PS: bookdown + formatR are awesome!

# wrapper around parse() and deparse(), iterativelly it tries to re-write every line with the correct width
tidy_block = function(text, width = getOption('width'), arrow = FALSE) {
  exprs = parse_only(text)
  if (length(exprs) == 0) return(character(0))
  exprs = if (arrow) replace_assignment(exprs) else as.list(exprs)
  expr_2=sapply(exprs, function(e) paste(base::deparse(e, width), collapse = '\n'))

  expr_2ver=NULL

  # for each overflow expr, iterate until it reaches the condition
  for(i in 1:length(expr_2))
  {
    expr = parse_only(expr_2[i])

    # flagging if exceeds max
    if(nchar(expr)>width)
    {
      new_width=width-1

      flag_end_while=T
      while(flag_end_while)
      {
        #new_expr=paste(base::deparse(expr, new_width), collapse = '\n')
        new_expr=sapply(expr, function(e) paste(base::deparse(e, new_width), collapse = '\n'))

        # split in several sentences
        new_expr_split=strsplit(new_expr, "\n")[[1]]

        # trim left spaces
        new_expr_split_t=trimws(new_expr_split, which = "both")

        # calculate sentence length
        expr_len=nchar(new_expr_split_t)

        # if all expr are under original width, then add expr to ok
        if(all(expr_len <= width))
        {
          expr_2ver=rbind(expr_2ver, new_expr)
          flag_end_while=F
        }

        new_width=new_width-1

        # seting min condition
        if(new_width<10)
        {
          new_expr=sapply(expr, function(e) paste(base::deparse(e, width), collapse = '\n'))
          expr_2ver=rbind(expr_2ver, new_expr)
          flag_end_while=F
        }

      } # end while
    } else {
      new_expr=sapply(expr, function(e) paste(base::deparse(e, width), collapse = '\n'))
      expr_2ver=rbind(expr_2ver, new_expr)
    }# end if width ok
  } # end for

  rownames(expr_2ver)=NULL
  return(expr_2ver[,1])
}

[krivit]

Just FYI, I submitted an Enhancement request to implement this at the deparse() level. (https://bugs.r-project.org/bugzilla/show_bug.cgi?id=17390)

[krivit]

The R Bugzilla (https://bugs.r-project.org/bugzilla/show_bug.cgi?id=17390) request was closed, though they are open to the possibility of a patch.

[krivit]

@yihui , what about using r-lib/styler? It looks like its API might allow line width enforcement in some form. Also, I think it might handle comments more elegantly than formatR.

[lorenzwalthert]

For the record: styler (as of version 1.0.2) does not currently support line width enforcement, but there is WIP to achieve this (contributed by @krivit, will be finalized by @lorenzwalthert), see r-lib/styler#414.