r-lib / callr Goto Github PK

View Code? Open in Web Editor NEW

287.0 16.0 34.0 9.83 MB

Call R from R

Home Page: https://callr.r-lib.org/

License: Other

R 100.00%

callr's People

Contributors

Stargazers

Watchers

callr's Issues

r_script

Mimic calling R -e or R -f or Rscript -e or Rscript.

Use Rterm.exe instead of R.exe, on windows

From r-lib/processx#96

And then add a test case for the timeout.

Annoyed warning from `readChar` on Windows

Hi, it's kind of annoying in windows, because the warnings pop up everytime, like below:

In addition: Warning messages:
1: In readChar(path, nchars = file.info(path)$size, ...) :
  can only read in bytes in a non-UTF-8 MBCS locale
2: In readChar(path, nchars = file.info(path)$size, ...) :
  can only read in bytes in a non-UTF-8 MBCS locale

It comes from readChar() in utils.R:

read_char <- function(path, ...) {
  readChar(path, nchars = file.info(path)$size, ...)
}

By checking out the help of readChar():

nchars will be interpreted in bytes not characters in a non-UTF-8 multi-byte locale, with a warning.

Sorry, I'm not familiar with readChar(), but any ideas? Thanks.

EDIT:
I guess by setting useByte = TRUE would be the solution #19

Pass env vars

SUpporting them directly is better, because various configs have various predefined sets that can still be overridden.

R CMD

Call R CMD instead of R. I mean, in addition.

Broken on R-devel?

From CRAN:

Dear maintainer,

Please see the problems shown on
<https://cran.r-project.org/web/checks/check_results_callr.html>.

Please correct before 2018-02-24 to safely retain your package on CRAN.

Best,
-k

Do not omit .Renviron, set R_LIBS in the fake profile

Because .Renviron might actually have some useful variables.

r_session

Create a background R session and run commands in it.

remotify ?

knit_remotely <- remotify(knitr::knit, envir = emptyenv())
knit_remotely(input = "blah.Rmd", output = "blah.md")

The extra argument envir just changes a default value.

Site library paths go missing when changing the library path

When changing the library path with .libPaths(), the site library paths disappear:

callr::r(function() .libPaths())
## [1] "/usr/local/lib/R/site-library" "/usr/lib/R/site-library"      
## [3] "/usr/lib/R/library"           

callr::r(function() {
  .libPaths(tempdir())
  .libPaths()
})
## [1] "/tmp/RtmplLBE1z"    "/usr/lib/R/library"

callr::r(function() Sys.getenv("R_LIBS_SITE"))
## [1] "/tmp/RtmpC00eG2/file41e2d104715"

I'm not sure if this is intended, but I was expecting to see:

.libPaths()
## [1] "/usr/local/lib/R/site-library" "/usr/lib/R/site-library"
## [3] "/usr/lib/R/library"

.libPaths(tempdir())
.libPaths()
## [1] "/tmp/RtmpCo52Xn"               "/usr/local/lib/R/site-library"
## [3] "/usr/lib/R/site-library"       "/usr/lib/R/library"

Sys.getenv("R_LIBS_SITE") 
## [1] "/usr/local/lib/R/site-library:/usr/lib/R/site-library:/usr/lib/R/library"

This was also the behavior in callr::r before ffa393f. The change seems to come from callr now setting R_LIBS_SITE to a temp file. The site library paths are inherited fine, but are lost when .libPaths() constructs new library paths.

Predefined parameter sets

vanilla
copycat

Setting R_LIBS_USER to nonexistant file breaks R CMD build in R <= 3.3

From a Travis CI session with R 3.3:

$ R_LIBS_USER=/dummy R CMD build .                  
* checking for file ‘./DESCRIPTION’ ... OK
* preparing ‘dplyr’:
* checking DESCRIPTION meta-information ... OK
* cleaning src
Error in loadVignetteBuilder(pkgdir, TRUE) : 
  vignette builder 'knitr' not found
Execution halted

The command works if omitting the R_LIBS_USER override. I believe this triggers r-lib/styler#46.

Fails on install for windows

Greetings,

I am installing a package that uses @master as a dependency.
After attempting to install both via the package, as well as directly the installation fails on windows 7

Specifically "make -f "Makevars.win"

Why are we setting `NOT_CRAN`?

I think this is not needed any more.

pkgdown site.

Fails: :(

> pkgdown::build_site()
══ Building pkgdown site ═══════════════════════════════════════════════════════
Reading from: '/Users/gaborcsardi/works/callr'
Writing to:   '/Users/gaborcsardi/works/callr/docs'
── Initialising site ───────────────────────────────────────────────────────────
Writing 'sitemap.xml'
── Building home ───────────────────────────────────────────────────────────────
Writing 'index.html'
── Building function reference ─────────────────────────────────────────────────
Loading callr
Reading 'man/callr.Rd'
Reading 'man/convert_and_check_my_args.Rd'
Reading 'man/get_result.Rd'
Reading 'man/make_error.Rd'
Reading 'man/r.Rd'
Reading 'man/r_bg.Rd'
Error in asNamespace(ns) : not a namespace
> traceback()
28: stop("not a namespace")
27: asNamespace(ns)
26: getNamespaceName(pkg)
25: unname(getNamespaceName(pkg))
24: ns_env_name(get_env(obj))
23: href_topic_local(in_braces)
22: as_html.tag_link(.x[[i]], ...)
21: .f(.x[[i]], ...)
20: .Call(map_impl, environment(), ".x", ".f", "character")
19: purrr::map_chr(x, as_html, ...)
18: flatten_para(text, ...)
17: as_data.tag_value(.x[[i]], ...)
16: .f(.x[[i]], ...)
15: .Call(map_impl, environment(), ".x", ".f", "list")
14: purrr::map(., as_data)
13: function_list[[i]](value)
12: freduce(value, `_function_list`)
11: `_fseq`(`_lhs`)
10: eval(quote(`_fseq`(`_lhs`)), env, env)
9: eval(quote(`_fseq`(`_lhs`)), env, env)
8: withVisible(eval(quote(`_fseq`(`_lhs`)), env, env))
7: sections %>% purrr::map(as_data) %>% purrr::map(add_slug)
6: data_reference_topic(topic, pkg, examples = examples, run_dont_run = run_dont_run,
       mathjax = mathjax)
5: .f(.x[[i]], ...)
4: .Call(map_impl, environment(), ".x", ".f", "list")
3: purrr::map(topics, build_reference_topic, pkg = pkg, lazy = lazy,
       examples = examples, run_dont_run = run_dont_run, mathjax = mathjax)
2: build_reference(pkg, lazy = lazy, examples = examples, run_dont_run = run_dont_run,
       mathjax = mathjax, seed = seed, override = override, preview = FALSE)
1: pkgdown::build_site()

Spinner is messing up `devtools::check()` output

E.g.

 OKhecking for file ‘devtools/DESCRIPTION’ ...

I can't figure out a good way to get back to the end of the previous line using ANSII escape codes to fix this, maybe you have some ideas.

Get the standard output and error from the child

I am not sure how, maybe in a file?

Useful R CMD envvars?

Devtools has this:

r_env_vars <- function() {
  vars <- c(
    "R_LIBS" = paste(.libPaths(), collapse = .Platform$path.sep),
    "CYGWIN" = "nodosfilewarning",
    # When R CMD check runs tests, it sets R_TESTS. When the tests
    # themselves run R CMD xxxx, as is the case with the tests in
    # devtools, having R_TESTS set causes errors because it confuses
    # the R subprocesses. Un-setting it here avoids those problems.
    "R_TESTS" = "",
    "R_BROWSER" = "false",
    "R_PDFVIEWER" = "false",
    "TAR" = auto_tar())

  if (is.na(Sys.getenv("NOT_CRAN", unset = NA))) {
    vars[["NOT_CRAN"]] <- "true"
  }

  vars
}

# Determine the best setting for the TAR environmental variable
# This is needed for R <= 2.15.2 to use internal tar. Later versions don't need
# this workaround, and they use R_BUILD_TAR instead of TAR, so this has no
# effect on them.
auto_tar <- function() {
  tar <- Sys.getenv("TAR", unset = NA)
  if (!is.na(tar)) return(tar)

  windows <- .Platform$OS.type == "windows"
  no_rtools <- is.null(get_rtools_path())
  if (windows && no_rtools) "internal" else ""
}

Would you consider adding to callr?

I think you're probably already handling R_LIBS
auto_tar can probably be dropped now

Compilation error on Intel C compiler

I am trying to build R and the Tidyverse with the Intel compiler suite, and callr is failing to build. I have included the error message below; it looks to me like perhaps there is a missing #include that gcc is more forgiving about?

** package ‘callr’ successfully unpacked and MD5 sums checked
** libs
icc -std=c99 -qopenmp -O3 -ipo -xHost -multiple-processes  -Wall tools/px.c -o tools/px
tools/px.c(76): warning #266: function "usleep" declared implicitly
        if (fnum > 0) usleep(fnum * 1000.0 * 1000.0);
                      ^

icc -std=c99 -qopenmp -O3 -ipo -xHost -multiple-processes  supervisor/supervisor.c supervisor/utils.c \
      -o supervisor/supervisor
supervisor/supervisor.c(87): warning #266: function "usleep" declared implicitly
      usleep(milliseconds * 1000);
      ^

supervisor/supervisor.c(130): warning #266: function "kill" declared implicitly
      int res = kill(pid, 0);
                ^

supervisor/supervisor.c(152): warning #266: function "kill" declared implicitly
          kill(children[i], SIGTERM);
          ^

supervisor/supervisor.c(199): warning #266: function "kill" declared implicitly
              kill(children[i], SIGKILL);
              ^

supervisor/supervisor.c(328): warning #266: function "fileno" declared implicitly
      if (fcntl(fileno(fp_input), F_SETFL, O_NONBLOCK) == -1) {
                ^

supervisor/supervisor.c(347): error: incomplete type is not allowed
      struct sigaction sa;
                       ^

supervisor/supervisor.c(349): warning #266: function "sigemptyset" declared implicitly
      sigemptyset(&sa.sa_mask);
      ^

supervisor/supervisor.c(350): warning #266: function "sigaction" declared implicitly
      if (sigaction(SIGINT, &sa, NULL)  == -1 ||
          ^

icc: error #10298: problem during post processing of parallel object compilation
compilation aborted for supervisor/supervisor.c (code 2)
make: *** [supervisor/supervisor] Error 2
ERROR: compilation failed for package ‘callr’

callr::r function cannot call quit()

Because that makes the R process, well, quit.

A workaround would be to have an argument for optional quit status, and then quit() would be explicitly called with this status.

New CRAN release

The last release was published about 2 years ago. Is there any plan to publish a new release? It is such a great tool to enhance R performance and flexibility.

Disable handle inheritance in child process

CF r-lib/processx#132

Get both stdout and stderr

I'm wondering if it is possible to get both output streams together, with lines in the right (chronological) order. One way to achieve this with system2 is to redirect stdout and stderr to the same connection, but this does not seem to work with callr, c.f. the following small snippet:

library(callr)

f = function() {
  print("print")
  message("message")
  return("return")
}
out = tempfile()

# with callr
r(f, stdout = out, stderr = out)
readLines(out)
# -> "print" missing

# with system2
rscript = tempfile()
fp = file(rscript, "w")
dump("f", file = fp)
writeLines("f()", fp)
close(fp)

system2("Rscript", rscript, stdout = out, stderr = out)
readLines(out)

Shall we have built-in progress bar support for r_session?

Or just let the users do it any way they want?

Set repository

Devtools has this:

# Create a temporary .Rprofile based on the current "repos" option
# and return a named vector that corresponds to environment variables
# that need to be set to use this .Rprofile
r_profile <- function() {
  tmp_user_profile <- file.path(tempdir(), "Rprofile-devtools")
  tmp_user_profile_con <- file(tmp_user_profile, "w")
  on.exit(close(tmp_user_profile_con), add = TRUE)
  writeLines("options(repos =", tmp_user_profile_con)
  dput(getOption("repos"), tmp_user_profile_con)
  writeLines(")", tmp_user_profile_con)

  c(R_PROFILE_USER = tmp_user_profile)
}

To ensure that the repository from the current session gets carried over to the child. Would you consider this for callr?

I think recent changes have broken devel devtools on travis

e.g. travis-ci/travis-ci#7576

@gaborcsardi Looks to be an issue with how arguments are being escaped in callr or proccessx. This is the only recent change in devtools + dependencies that I think would cause it.

Output is messed up sometimes

Happens e.g. when calling devtools::build on devtools itself.

Maybe this is a processx issue, actually.

Syntactic sugar for functions in packages

callr::r(stats::acf, args = list(runif(100)))

should be the same as

callr::r(function(...) stats::acf(...), args = list(runif(100)))

Return the full error stack from the other process

somehow....

Compilation fails if RTools is not installed in default folder

Hi,

Compilation fails when RTools is installed in a location other than C:\RTools.
Since RTools/bin is added to the PATH as a common practice, I suggest to call gcc without specifying exact location, or use the BINPREF variable, as discussed here

Thanks, Ido

Timeout in r_session$run()

Bump version (e.g. to 2.0.3-9000) to distinguish from CRAN 2.0.3

Please bump developer version to distinguish from version on CRAN.

Pipe the output of the child process

Optionally.

Tests

Make `r_safe` the default, i.e. r can be just an alias to it.

Error in process_kill(self, private, grace) : process_kill: Operation not permitted

When I run the following repeatedly:

p <- process$new('ls')
p$kill()

This error occasionally comes up:

> p <- process$new('ls')
> p$kill()
[1] TRUE
> p <- process$new('ls')
> p$kill()
[1] FALSE
> p <- process$new('ls')
> p$kill()
Error in process_kill(self, private, grace) : 
  process_kill: Operation not permitted
> p <- process$new('ls')
> p$kill()
[1] FALSE

Could this be due to a race where the process stops after the R function process_kill() is called, but before the C function callr_kill gets to the part where it does stuff with the process?

In Travis logs, I also see callr_kill: No child processes. I haven't figured out how to reproduce it locally, but I suspect it has a similar cause.

Error in process_signal(self, private, signal) : 
  callr_get_exit_status: No child processes
Calls: <Anonymous> ... <Anonymous> -> sd_stop -> <Anonymous> -> process_signal
Error in process_kill(self, private, grace) : 
  callr_kill: No child processes

https://travis-ci.org/rstudio/shinytest/jobs/359462139#L3125-L3130

Nice README

Installation failed: 'rcmd_safe_env' is not an exported object from 'namespace:callr'

When installing packages via devtools::install_github("repository") I am receiving the error that `rcmd_safe_env' is not in the namespace of callr. I checked, and it looks like it is in fact in the namespace.

I am using Ubuntu, if that is helpful.

callr not working from testthat

We have a testthat test case that uses callr that does not work on travis or when called from R CMD check. Interestingly, it works fine if testthat is called directly. I made a demo repo at https://github.com/hannesmuehleisen/callrtestthattravistest. Error thrown:

Running ‘testthat.R’
ERROR
Running the tests in ‘tests/testthat.R’ failed.
Last 13 lines of output:
  cannot open the connection
  1: callr::r(do_something_dumb, show = TRUE) at testthat/test_01_callr.R:9
  2: r_internal(func, args, libpath = libpath, repos = repos, stdout = stdout, stderr = stderr, 
         error = error, cmdargs = cmdargs, show = show, callback = callback, system_profile = system_profile, 
         user_profile = user_profile, env = env)
  3: get_result(res)
  4: readRDS(res)
  5: gzfile(file, "rb")

Travis error is here: https://travis-ci.org/hannesmuehleisen/callrtestthattravistest/builds/189459759

Names of the `repos` option are being stripped

Backstory here: Bioconductor/BiocManager#4 (comment)

Unless told otherwise, I will make a PR.

Uninformative error if session terminates

# devtools::install_github("romainfrancois/trump")

callr::r(function() library(trump))
#> Error in readRDS(res) : error reading from connection 
#> 3. readRDS(res) 
#> 2. get_result(output = out, options) 
#> 1. callr::r(function() library(trump))

rcmd without CMD

the rcmd() function, as the stated by the name, runs a 'R CMD xxx' command.

What if I want to run 'R --version' ? Or R RHOME ?
Should I'd be better off using r(function(x) R.version), but it's a bit slower.

Thanks.

passing expression to r()

naive question: why not taking an expression argument to r() ?

Is there any issue ? Because it also covers the function call case.
I mean

r(quote(summary(mycars)), vars = list(mycars)))

instead of

r(function(x) summary(x), args = list(mycars))

I cannot redirect the I/O streams from r_bg().

# The error message is cryptic.
callr::r_bg(func = function() 1 + "A", stdout = "~/out.txt", stderr = "~/err.txt")
#> Error in process_initialize(self, private, command, args, stdout, stderr, : callr error
# No traceback or warnings are available to the master process.
traceback()
#> No traceback available
warnings()
#> NULL
# Neither I/O stream made it to a text file.
any(c("out.txt", "err.txt") %in% list.files("~"))
#> [1] FALSE
# Using development callr: 3d398568995e5769eb9cab844ca735662e009485
sessionInfo()
#> R version 3.4.3 (2017-11-30)
#> Platform: x86_64-pc-linux-gnu (64-bit)
#> Running under: KDE neon User Edition 5.12
#> 
#> Matrix products: default
#> BLAS: /usr/lib/libblas/libblas.so.3.6.0
#> LAPACK: /usr/lib/lapack/liblapack.so.3.6.0
#> 
#> locale:
#>  [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C              
#>  [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8    
#>  [5] LC_MONETARY=en_US.UTF-8    LC_MESSAGES=en_US.UTF-8   
#>  [7] LC_PAPER=en_US.UTF-8       LC_NAME=C                 
#>  [9] LC_ADDRESS=C               LC_TELEPHONE=C            
#> [11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C       
#> 
#> attached base packages:
#> [1] stats     graphics  grDevices utils     datasets  methods   base     
#> 
#> loaded via a namespace (and not attached):
#>  [1] Rcpp_0.12.15     crayon_1.3.4     digest_0.6.15    rprojroot_1.3-2 
#>  [5] assertthat_0.2.0 R6_2.2.2         backports_1.1.2  formatR_1.5     
#>  [9] magrittr_1.5     evaluate_0.10.1  debugme_1.1.0    stringi_1.1.6   
#> [13] callr_2.0.2.9001 rmarkdown_1.8    tools_3.4.3      stringr_1.3.0   
#> [17] yaml_2.1.16      compiler_3.4.3   htmltools_0.3.6  knitr_1.20

Option to print cmd used in rcmd()

As well as output = TRUE it would be useful to have (say) input = TRUE which would print the complete command run by rcmd().

lapply()-like functionality

I picture a version of lapply() that distributes the list over multiple processes spawned by callr::r() or callr::rcmd(). Is this part of the plan for future development?

Better name for `r_eval`

Maybe I'll just call it r?

Confused about s3 dispatch in r_safe

I'm not sure if this qualifies as a bug, but I don't understand what's happening here. Consider the following code snippet:

callr::r_safe(
  function(n) {
    point <- function(x, y) {
      stopifnot(is.numeric(x), is.numeric(y))
      structure(list(x = x, y = y), class = "point")
    }

    abs.point <- function(x) {
      sqrt(x$x ^ 2 + x$y ^ 2)
    }

    points <- mapply(point, runif(n), runif(n), SIMPLIFY = FALSE)

    sapply(points, abs)
  },
  args = list(n = 5L)
)

If abs.point is called as abs(points[[1]]), it works as I would expect. But using sapply to iterate, I get a non-numeric argument to mathematical function error.

The actual problem I'm facing is that something along the lines of the above example is being used in an rmarkdown document and pkgdown uses r_safe to knit the .Rmd file.

Print stderr on fail_on_status?

Might be useful for debugging

Handle crashes in r_session

Might be different for the --interactive case.

r-lib / callr Goto Github PK

callr's People

Contributors

Stargazers

Watchers

Forkers

callr's Issues

Recommend Projects

Recommend Topics

Recommend Org

Jobs