Use markdown for italic/bold in titles

[msberends]

Currently, it is quite cumbersome to print parts of labels (like title, subtitle) as italic or bold. For my field of medical microbiology this is quite important; we write names of microorganisms italic, so a title could be "Resistance of Amoxicillin in E. coli"

We now use combinations of expression and bquote to make it possible to parametrise it, but that's really hard and error prone. Could you support anything like we're used to from RMarkdown?

ggplot(data, aes(x, y)) +
  geom_anything +
  labs(title = "Resistance of Amoxicillin in *E. coli*")

I'm not the only one:

• R ggplot How to italicize part one or two words of an axis title http://stackoverflow.com/questions/32555531/ddg#32559133
• Using italic() with a variable in ggplot2 title expression http://stackoverflow.com/questions/31927984/ddg#31928067
• How to italicize part (one or two words) of an axis title https://stackoverflow.com/questions/32555531

Without really convenient solutions.

[clauswilke]

I think everybody agrees something like this is needed. The problem is that we're limited by the capabilities of the underlying graphics library we're using. There is currently no good way to render markdown to an R graphics device.

I've started exploring the techniques that might be used to solve this issue, but we're nowhere near integrating this into ggplot: https://github.com/clauswilke/gridtext

[msberends]

I understand, but what if we do it a little more dirty?

Momentarily, I use this as a hack:

markdown_bolditalic <- function(text) {
  if (text %like% '[*]{2}.+[*]{2}') {
    # bold
    locations <- unlist(gregexpr(pattern = '[**]',
                                 text))
    locations <- locations[1:2]
    text.left <- substr(text,
                        1,
                        min(locations) - 1)
    text.middle <- substr(text,
                          min(locations) + 2,
                          max(locations) - 2)
    text.right <- substr(text,
                         max(locations) + 1,
                         nchar(text))

    text.new <- substitute(
      expression(paste(
        text.left, bold(text.middle), text.right
      )),
      list(
        text.left = text.left,
        text.middle = text.middle,
        text.right = text.right
      )
    )
    text.new %>% eval()

  } else if (text %like% '[*_].+[*_]') {
    # italic
    locations <- unlist(gregexpr(pattern = '[*_]',
                                 text))
    locations <- locations[1:2]
    text.left <- substr(text,
                        1,
                        min(locations) - 1)
    text.middle <- substr(text,
                          min(locations) + 1,
                          max(locations) - 1)
    text.right <- substr(text,
                         max(locations) + 1,
                         nchar(text))

    text.new <- substitute(
      expression(paste(
        text.left, italic(text.middle), text.right
      )),
      list(
        text.left = text.left,
        text.middle = text.middle,
        text.right = text.right
      )
    )
    text.new %>% eval()

  } else {
    text
  }
}

It only works for the first found element, but since I often use only one organism name in a title, it works for me. I know, it’s quite crappy, but maybe this could be used in a form of a loop? I think we only need italic and bold to start with.

[clauswilke]

If the goal is to write a function that can translate markdown into the existing R plotmath interface then there's no need for it to live in the ggplot2 repository. In fact, it's not even ggplot2 specific. It would probably be useful for base R plotting as well. You could consider placing it into a separate package.

[msberends]

I don’t think a plotmath function is the best solution, just proposing one possibility 😊

[baptiste]

FWIW the tikzDevice package can render LaTeX markup, so one could use commonmark to do the translation

library(ggplot2)

p <- ggplot() +
  labs(title = commonmark::markdown_latex("Resistance of Amoxicillin in *E. coli*"))

library(tikzDevice)
ggsave('test.tex', p, device = tikz, standAlone=TRUE, width=3, height=2)

[msberends]

Awesome suggestion and a lot closer indeed, but requires quite some dependencies.

[jemus42]

Falling back on TeX for mere inline text formatting seems to be, uh, well, the German idiom is "mit Kanonen auf Spatzen schießen" – i.e. a huge solution for a tiny problem.

[paleolimbot]

This is probably closer to possible now with markdown plus gridtext:

html <- markdown::markdownToHTML(
  text = "Resistance of Amoxicillin in **E. coli**",
  fragment.only = TRUE
)
html
#> [1] "<p>Resistance of Amoxicillin in <strong>E. coli</strong></p>\n"

# gridtext doesn't understand "<strong>" yet
html <- gsub("<(/?)strong>", "<\\1b>", html)

grid::grid.draw(gridtext::rich_text_grob(html))

^{Created on 2019-06-17 by the reprex package (v0.2.1)}

[clauswilke]

gridtext is not ready to be used yet. I'll completely rewrite this summer. Adding support for <strong> would be trivial, but making the grob play nicely with ggplot2 is more complicated.

[thomasp85]

Ultimately this is capabilities I’d like to add at the graphic engine layer

[clauswilke]

Thomas, I agree. Just to make sure whatever I do now will be reusable later: Do you expect that there will have to be some sort of text layouting code, or do you plan to hand all of that off to some outside rendering library? My sense is, but I could be wrong, that if we want to continue supporting all the existing R graphics devices, we need to provide our own layouting code, just like is currently the case with plotmath. To this end, I'm currently planning to write some basic layouting code in C++, and it could be integrated into the graphics engine at a later point. Do you think this is reasonable, or do you have other ideas?

Writing a layouting engine is a fair amount of work, but trying to repurpose an existing one (as described e.g. here) comes with all sorts of limitations and performance issues.

[thomasp85]

I’m honestly not sure how this would come out. The way the graphic engine passes off text to the graphic devices makes me think we can solve most of it cleanly in the graphic engine while still supporting old device but I’m not sure.

tldr; I don’t know what the best course of action is for you