setting output_dir in render() will create absolute path for knitr fig.path

[cderv]

See this reproducible example

xfun::write_utf8(c(
  '```{r plot, fig.cap = "A plot", out.width = "100%"}',
  'plot(mtcars)',
  '```'
), "test.Rmd")

callr::r(function() {  
  res <- rmarkdown::render("test.Rmd", rmarkdown::md_document())
  grep("img", xfun::read_utf8(res), value = TRUE)
})
#> [1] "<img src=\"test_files/figure-markdown_strict/plot-1.png\" alt=\"A plot\" width=\"100%\" />"

callr::r(function() {
  res <- rmarkdown::render("test.Rmd", rmarkdown::md_document(), output_dir = "dummy")
  grep("img", xfun::read_utf8(res), value = TRUE)
})
#> [1] "<img src=\"C:/Users/chris/AppData/Local/Temp/Rtmpy8Seib/reprexe05041605211/dummy/test_files/figure-markdown_strict/plot-1.png\" alt=\"A plot\" width=\"100%\" />"

Only difference between the two calls is setting the output_dir. I don't think this is expected behavior but this is what happens in render()

• If output_dir is set, it will be normalized https://github.com/rstudio/rmarkdown/blob/4b119a9782f6153ed0c28514e1360b9a7229e010/R/render.R#L336-L342 https://github.com/rstudio/rmarkdown/blob/4b119a9782f6153ed0c28514e1360b9a7229e010/R/render.R#L488-L492
• Then it will be used to determine the path for figures https://github.com/rstudio/rmarkdown/blob/4b119a9782f6153ed0c28514e1360b9a7229e010/R/render.R#L500-L502
• And used to set fig.path https://github.com/rstudio/rmarkdown/blob/4b119a9782f6153ed0c28514e1360b9a7229e010/R/render.R#L632-L635

Maybe in this last step we should explicitly set fig.path to a relative one ? Using xfun::relative_path() ?

The handling of paths in render is always source of unexpected behavior when one the argument is set.

Related in a way to #1508

[yihui]

I was aware of this problem long time ago, and tried to make the figure path relative since I didn't like the absolute paths, but I remember the interaction among output_file, output_dir, and intermediate_dir was so complicated that I had to give up. ..

[cderv]

Maybe we could try for a next version to improve / fix / rethink all the path interaction. I feel this is a larger work as it seems we keep fixing things here and there.

I may try my idea of just setting relative path here for figure in this case, but if we agree on larger work maybe I'll wait for when I have time to tackle all this.

[yihui]

I may try my idea of just setting relative path here for figure in this case

Feel free to try. I don't remember what I tried exactly last time or what problem I ran into.

[darckula]

Hi guys, I just found myself trapped on the same issue. Funny to see that this topic was open "recently".

In my case, I am generating reports with rmarkdown with output format github_document. My directory structure is as follows:

Rstudio project directory:

• place for markdown documents *.md
• folder "charts" where I want to locate the graphs generated by markdown when generating the ggplot graphs
• folder "Rmds" where I locate the rmd source files.

Example:

./charts/
      fig1.png
./Rmds/
      my_file.Rmd
./my_file.md

In order to change the output directory for the md document, I use: rmarkdown::render("Rmds/my_file.Rmd", output_dir=".")

The problem is that this command changes the fig paths to be absolute (as opposite to relative) to my computer. Reading this thread, I used the chunk option fig.path="../charts" which worked well and put the charts in the intended folder... BUT, the md file generated has also that path for grab the charts, this is, the md document tries to find the graphs under "../charts" and it should be "charts" (not the parent directory but the current).

I cannot find a solution for this... not sure if this example can assist you to figure out the problem. If I can be of any help to test something, I'd be pleased.

thanks!

[cderv]

@yihui as reported in https://github.com/rstudio/rticles/issues/528 this in fact has a big impact on vignettes.

I believe when vignettes are build, output_dir() is set https://github.com/yihui/knitr/blob/master/R/utils-vignettes.R#L64-L68 This follows a fix you made to prevent a CRAN workaround https://github.com/yihui/knitr/commit/b80ece1c6a8eeb2c3ee6cd177d6b34891f162d4d

RStudio IDE will also set output_dir in some cases to a temp directory whe clicking on Knit button https://github.com/rstudio/rstudio/blob/60f5a749a3f469e680655c6c7906610086950604/src/cpp/session/modules/rmarkdown/SessionRMarkdown.cpp#L555-L568 And vignette is one of the case where this happens

• tempfile is used for package documentation
• and anything in vignettes/ will trigger it

I don't know if this is recent or not but definitely happening.

We should definitely try to find a way to improve that for next release.

[yihui]

It sounds like the problem should affect all PDF vignettes that contain plots? If it is such a serious problem, I wonder why we didn't receive more reports.

Anyway, it's definitely worth fixing, but it may be quite tricky...

[cderv]

It sounds like the problem should affect all PDF vignettes that contain plots? If it is such a serious problem, I wonder why we didn't receive more reports.

Yes I wondered the same. 🤔 I think it is worth looking into to see if we can something scoped to vignette at first maybe. I understand this could be quite tricky.

Maybe this absolute path is not really the issue for vignette (as what happens in https://github.com/rstudio/rticles/issues/528) and the issue in only on Windows environment with absolute path in .tex files and some missing escaping for LaTeX . I'll look into this.