Wrapping an R script: how do I compose the R_LIBS_SITE environment variable?

Wrapping an R script: how do I compose the R_LIBS_SITE environment variable?

[divoplade]

Dear guix,

I am using guix to write R packages. For convenience, I created a small
R script that does the job.

To make this script work as-is, I wrapped it using wrap-program. wrap-
program creates a shell script that sets up environment variables and
then call the real script, because R needs to be able to find all the
packages and dependencies. More specifically, I need to extend the
R_LIBS_SITE environment variable to point to all the R dependencies of
my package, as well as their recursive dependencies.

How do I compute that? I only care about R dependencies, so using the
whole recursive dependency tree of my package seems too much.

My current solution is to save the R_LIBS_SITE environment from the
build procedure and just use that. The obvious problem is that it also
contains build dependencies, and I don't think it will work correctly
for cross-compiled packages.

Do you have a better idea?

Best regards,

divoplade

Re: Wrapping an R script: how do I compose the R_LIBS_SITE environment variable?

[zimoun]

Hi,

On Fri, 05 Feb 2021 at 17:14, divoplade <d@divoplade.fr> wrote:

> I am using guix to write R packages. For convenience, I created a small
> R script that does the job.

Is “guix import cran” not enough?  What is the source of these R packages?

Well, could you be more specific about “using guix to write R packages”?


> To make this script work as-is, I wrapped it using wrap-program. wrap-
> program creates a shell script that sets up environment variables and
> then call the real script, because R needs to be able to find all the
> packages and dependencies. More specifically, I need to extend the
> R_LIBS_SITE environment variable to point to all the R dependencies of
> my package, as well as their recursive dependencies.

This seems the job of Guix.

> How do I compute that? I only care about R dependencies, so using the
> whole recursive dependency tree of my package seems too much.

Ricardo wrote a R helper:

   <https://elephly.net/paste/1612619217.R.html>

Basically, it does more or less what “guix install” does but directly in
the R repl.  Maybe it is what you want.


Hope that helps,
simon

Re: Wrapping an R script: how do I compose the R_LIBS_SITE environment variable?

[Ricardo Wurmus]

zimoun <zimon.toutoune@gmail.com> writes:

> Ricardo wrote a R helper:
>
>    <https://elephly.net/paste/1612619217.R.html>
>
> Basically, it does more or less what “guix install” does but directly in
> the R repl.  Maybe it is what you want.

It also imports packages on the fly in case they don’t exist in Guix
yet.

-- 
Ricardo

Re: Wrapping an R script: how do I compose the R_LIBS_SITE environment variable?

[divoplade]

[-- Attachment #1: Type: text/plain, Size: 891 bytes --]

Hello,

Sorry, I realize I was not clear enough.

Attached is an example of a script written in R that has non-trivial
recursive dependencies.

You can run the example as:

guix environment --ad-hoc --container -l package.scm -- hello
You will normally see something like:

$ guix environment --ad-hoc --container -l /tmp/package.scm -- hello
# A tibble: 10 x 2
       x     y
   <int> <dbl>
 1     1     1
 2     2     4
 3     3     9
 4     4    16
 5     5    25
 6     6    36
 7     7    49
 8     8    64
 9     9    81
10    10   100
sh: rm: command not found

If you look at the package definition, you see that I call wrap-program 
in order to set the R_LIBS_SITE environment variable. The components of
R_LIBS_SITE is the R dependencies of my package, and their recursive R
dependencies. But I don't know how to get them.

I hope this example is better at explaining the problem.

[-- Attachment #2: package.scm --]
[-- Type: text/x-scheme, Size: 2051 bytes --]

(use-modules (guix packages)
	     (guix build-system trivial)
	     (guix gexp)
	     (gnu packages base)
	     (gnu packages bash)
	     (gnu packages statistics)
	     (gnu packages cran))

(package
 (name "r-hello")
 (version "0.0.0")
 (source (plain-file "empty" ""))
 (build-system trivial-build-system)
 (arguments
  '(#:modules ((guix build utils))
    #:builder
    (begin
      (use-modules (guix build utils))
      (let ((out (assoc-ref %outputs "out"))
	    (r-in (assoc-ref %build-inputs "r-minimal"))
	    (bash-in (assoc-ref %build-inputs "bash")))
	(mkdir out)
	(mkdir (string-append out "/bin"))
	;; Bug? PATH is not set
	(setenv "PATH" (string-append bash-in "/bin"))
	(call-with-output-file (string-append out "/bin/hello")
	  (lambda (port)
	    (format port "#!~a/bin/Rscript

library (\"magrittr\")

data <- (tibble::tibble (x = 1:10)
    %>% dplyr::mutate (y = x ^ 2))

print (data)
"
		    r-in)))
	(chmod (string-append out "/bin/hello") #o755)
	(wrap-program
	 (string-append out "/bin/hello")
	 `("R_LIBS_SITE" ":" =
	   ,(map (lambda (r-package)
		   ;; R_LIBS_SITE should point to the site-library
		   ;; folder of the packages
		   (string-append (assoc-ref %build-inputs r-package)
				  "/site-library"))
		 ;; Here are my R dependencies:
		 '("r-magrittr" "r-tibble" "r-dplyr"
		   ;; However, running hello requires these recursive
		   ;; dependencies. How am I supposed to know? What if
		   ;; they change with future versions of my
		   ;; dependencies?
		   "r-rlang" "r-vctrs" "r-r6" "r-generics" "r-glue"
		   "r-lifecycle" "r-ellipsis" "r-pillar" "r-crayon"
		   "r-pkgconfig" "r-tidyselect" "r-purrr" "r-cli"
		   "r-assertthat" "r-fansi" "r-utf8"))))))))
 (native-inputs
  `(("bash" ,bash)))
 (inputs
  `(("r-minimal" ,r-minimal)
    ("bash" ,bash)))
 (propagated-inputs
  `(("r-magrittr" ,r-magrittr)
    ("r-tibble" ,r-tibble)
    ("r-dplyr" ,r-dplyr)))
 (synopsis "A script written in R")
 (description "How can I list the recursive R dependencies?")
 (home-page "http://example.com")
 (license '...))

Re: Wrapping an R script: how do I compose the R_LIBS_SITE environment variable?

[Ricardo Wurmus]

divoplade <d@divoplade.fr> writes:

> Attached is an example of a script written in R that has non-trivial
> recursive dependencies.

Why don’t you use the r-build-system here?

> If you look at the package definition, you see that I call wrap-program 
> in order to set the R_LIBS_SITE environment variable. The components of
> R_LIBS_SITE is the R dependencies of my package, and their recursive R
> dependencies. But I don't know how to get them.

The r-build-system has a procedure “generate-site-path” to generate
R_LIBS_SITE.

-- 
Ricardo

Re: Wrapping an R script: how do I compose the R_LIBS_SITE environment variable?

[divoplade]

Hello,

Le dimanche 07 février 2021 à 08:11 +0100, Ricardo Wurmus a écrit :
> The r-build-system has a procedure “generate-site-path” to generate
> R_LIBS_SITE.

Thank you, I did not know about this function. It is not exported, but
I can copy it, or re-use the R_LIBS_SITE environment variable after the
install phase.

Re: Wrapping an R script: how do I compose the R_LIBS_SITE environment variable?

[zimoun]

Hi,

On Sun, 07 Feb 2021 at 10:30, divoplade <d@divoplade.fr> wrote:
> Le dimanche 07 février 2021 à 08:11 +0100, Ricardo Wurmus a écrit :

>> The r-build-system has a procedure “generate-site-path” to generate
>> R_LIBS_SITE.
>
> Thank you, I did not know about this function. It is not exported, but
> I can copy it, or re-use the R_LIBS_SITE environment variable after the
> install phase.

The Guile modules are a bit mysterious for me but you can try:

  (define generate-site-path
     (@@ (guix build r-build-system) generate-site-path))

instead of copying.


Cheers,
simon