File paths with drake on a shared drive

Question

I am encountering some odd drake behaviour which I just can't figure out. I am trying to add a .rmd to my drake plan. I am working on a remote machine AND on a network drive on that machine. If I try to add an .rmd file to my plan like this:

> library(drake)
> library(rmarkdown)
> 
> list.files()
[1] "drake_testing.Rproj"        "foo.png"             "report.Rmd"                    
> 
> plan <- drake_plan(
+   png("foo.png"),
+   plot(iris$Sepal.Length ~ iris$Sepal.Width),
+   dev.off(),
+   report = render(
+     input = knitr_in("report.Rmd"),
+     output_file = "report.html",
+     quiet = TRUE
+   )
+   
+ )
> 
> plan
# A tibble: 4 x 2
  target         command                                                                               
  <chr>          <expr>                                                                                
1 drake_target_1 png("foo.png")                                                                        
2 drake_target_2 plot(iris$Sepal.Length ~ iris$Sepal.Width)                                            
3 drake_target_3 dev.off()                                                                             
4 report         render(input = knitr_in("report.Rmd"), output_file = "report.html",      quiet = TRUE)
> 
> ## Turn your plan into a set of instructions
> config <- drake_config(plan)
Error: The specified file is not readable: report.Rmd
> 
> traceback()
13: stop(txt, obj, call. = FALSE)
12: .errorhandler("The specified file is not readable: ", object, 
        mode = errormode)
11: digest::digest(object = file, algo = config$hash_algorithm, file = TRUE, 
        serialize = FALSE)
10: rehash_file(file, config)
9: rehash_storage(target = target, file = file, config = config)
8: FUN(X[[i]], ...)
7: lapply(X = X, FUN = FUN, ...)
6: weak_mclapply(X = keys, FUN = FUN, mc.cores = jobs, ...)
5: lightly_parallelize_atomic(X = X, FUN = FUN, jobs = jobs, ...)
4: lightly_parallelize(X = knitr_files, FUN = storage_hash, jobs = config$jobs, 
       config = config)
3: cdl_get_knitr_hash(config)
2: create_drake_layout(plan = plan, envir = envir, verbose = verbose, 
       jobs = jobs_preprocess, console_log_file = console_log_file, 
       trigger = trigger, cache = cache)
1: drake_config(plan)

I have tried the following permutations to make this work:

• Move the .rmd to the local drive and call it with the full path to there
• Add in file.path inside and outside of knitr_in to complete a full path.
• Try using file_in for each of the scenarios above.

I have also tried debugging but I get a little lost when drake turns the file name into a hash then turns it back into the basename of the file (i.e. report.Rmd). The error ultimately happens when digest::digest is called.

Does anyone have experience attempting to figure out something like this?

Answer 1

I think the answer depends on whether you get the same error when you call digest("report.Rmd", file = TRUE) on its own outside drake_config(plan). If it errors (which I am betting it does) there may be something strange about your file system that clashes with R. If that is the case, then there is unfortunately nothing drake can do.

I also suggest some changes to your plan:

plan <- drake_plan(
  plot_step = {
    png(file_out("foo.png")),
    plot(iris$Sepal.Length ~ iris$Sepal.Width),
    dev.off()
  },
  report = render(
    input = knitr_in("report.Rmd"),
    output_file = "report.html",
    quiet = TRUE
  )  
)

Or better yet, compartmentalize your work in reusable functions:

plot_foo = function(filename) {
  png(filename),
  plot(iris$Sepal.Length ~ iris$Sepal.Width),
  dev.off()
}

plan <- drake_plan(
  foo = plot_foo(file_out("foo.png")),
  report = render(
    input = knitr_in("report.Rmd"),
    output_file = "report.html",
    quiet = TRUE
  )  
)

A target is a skippable workflow step with a meaningful return value and/or output file(s). png() and dev.off() are part of the plotting step, and file_out() tells drake to watch foo.png for changes. Also, it is good practice to name your targets. Usually, the return values of targets are meaningful, just like variables in R.