knitr::purl fail with parametrized Rmd when params is used in chunk option

[cderv]

From ropensci/targets#256

When tangle = TRUE and Rmd uses parameters, it works ok if params are used in code chunks but it fails if params is used in one of the chunk option.

lines <- c(
  "---",
  "title: parameters",
  "output: html_document",
  "params:",
  "  paramcd: TRUE",
  "---",
  "",
  "```{r, eval = params$paramcd}",
  "print(params$paramcd)",
  "print(object)",
  "```"
)
writeLines(lines, "report.Rmd")
knitr::knit("report.Rmd", tangle = TRUE, quiet = TRUE)
#> Error in eval(x, envir = envir) : objet 'params' introuvable
#> [1] "report.R"
xfun::file_string("report.R")
#> params <-
#> list(paramcd = TRUE)

lines[8] <- "```{r, eval = TRUE}"
writeLines(lines, "report.Rmd")
knitr::knit("report.Rmd", tangle = TRUE, quiet = TRUE)
#> [1] "report.R"
xfun::file_string("report.R")
#> params <-
#> list(paramcd = TRUE)
#> 
#> ## ---- eval = TRUE-------------------------------------------------------------
#> print(params$paramcd)
#> print(object)

It seems like a bug.

Also, it seems knitr is not aware about parameters defined in a Rmd file when knitting. But this may be unsupported as parametrized report is a rmarkdown feature mainly

lines <- c(
  "---",
  "title: knitr with params",
  "output: html_document",
  "params:",
  "  paramcd: TRUE",
  "---",
  "",
  "```{r, eval = TRUE}",
  "str(params)",
  "```"
)
writeLines(lines, "report.Rmd")
res <- knitr::knit("report.Rmd", quiet = TRUE)
xfun::file_string(res)
#> ---
#> title: knitr with params
#> output: html_document
#> params:
#>   paramcd: TRUE
#> ---
#> 
#> ```r
#> str(params)
#> #> Error in str(params): object 'params' not found
#> ```

[yihui]

But this may be unsupported as parameterized report is a rmarkdown feature mainly

That's it.

[cderv]

Then we can say it is a feature request to be able to "purl" Rmd file, even when using parameters ?

It could also be another tool than knitr, as parsing code chunks and inline code from a Rmd chunk is something that seem to be often needed. For now, I think every usage like that uses purl().

[yihui]

In principle, purl() doesn't evaluate any code in the document (only knit() does), so it won't recognize params$paramcd. To support this feature, we may have to break the design.

[cderv]

I see. Thanks!

I think there is a need for Rmd code parsing with evaluation (as the one in targets - cc @wlandau @yonicd) but maybe it has to be different from purl. I'll keep think about it for the long term, looking at other existing use case.

[yihui]

I'm okay with supporting this case. I have already had some code supporting the special case of R Markdown in knitr. Parsing YAML doesn't involve "evaluating" code, so that's probably fine.

In fact, params is supported when you do not use it in chunk options:

knitr/R/output.R

Lines 235 to 238 in 0272a83

	params = knit_params(text)
	params = if (length(params))
	c('params <-', capture.output(dput(flatten_params(params), '')), '')
	.knitEnv$tangle.params = params # for hook_purl()

The deeper problem is that I have never been satisfied with purl(). It is hard and sometimes impossible to do it right without evaluating code. The true way to purl code from a document is actually to use the hook knitr::hook_purl().

[wlandau]

The true way to purl code from a document is actually to use the hook knitr::hook_purl().

I think there is still a need to statically extract code from a knitr document from outside the document itself (examples from targets and drake).

[yihui]

Yes, there is a need. The problem is that it is not possible to both statically and faithfully extract code in certain cases. The hook_purl is must more reliable than purl(), but the obvious disadvantage is that it requires knitting the document.