IMO, using pandoc makes the code simple and applicable to more formats (e.g., gfm).
What do you think?

# purloc = purl + pandoc
purloc = function(x, output = file.path(".", xfun::with_ext(x, "R")), ...) {
  input = tempfile(fileext = xfun::file_ext(x))
  file.copy(x, input)

  knitr::pandoc(input, 'commonmark', ext = 'md')
  
  intermediate_md = xfun::with_ext(input, 'md')
  intermediate_md %>%
    readr::read_lines() %>%
    stringr::str_replace_all("^``` r", '```{r}') %>%
    readr::write_lines(intermediate_md)
  knitr::purl(intermediate_md, output = output, ...)
}

@atusy that is a great simplification and improvement on the DIY solution in the original.

The purloc naming is also intuitive and makes sense.

Some questions

Do you think inc_out option is useful? (a quick example of the difference below)

# from inside applied-ml root directory
dir() %>% grep("Part_{1}.*html", ., value = T) %>% sapply(., html_to_r) -> a
dir() %>% grep("Part_{1}.*html", ., value = T) %>% mapply(html_to_r, inc_out=F, .) -> b

a[[2]] %>% cat # with output
b[[2]] %>% cat # without output

For me, it's useful, but maybe not for everyone?

Also, do you agree replacing character entities is useful? I think it is essential (otherwise pipes and some conditionals will appear meaningful in HTML but not in R code)

  replace_character_entities <- function(char_entity){
    xml2::xml_text(xml2::read_html(paste0("<x>", char_entity, "</x>")))
  }

# E.g. 
replace_character_entities("&gt;")
# [1] ">"

Which makes a pipe appear as %>% rather than %>%

I applied this conversion to some test examples but I cannot be certain it will work under all circumstances (one exception that comes to mind is if R code contained some literal >, perhaps in a comment). I think this could be sufficiently rare to not cause too much concern though

About inc_out, I think it is relatively less important.
Because results are expected to be reproducible.
Also, a problem arise when the source Rmd contains code blocks.
They are not output, but are considered as output by html_2_r.
If inc_out is really needed, I think they should be commented out in R.

About special characters, we do not have to care as pandoc takes care of them

echo "<pre>%&gt;</pre>" | pandoc --from html --to gfm
# ```
# %>
# ```