[R-pkg-devel] Writing to files without altering working directory in R package

[Jeff Newmiller]

I find the idea of using setwd in examples disturbing, because I try to avoid using setwd at all as a way to delegate my choice of working directory to the operating system actions taken when starting R. In other words, I recommend to  beginners that they start R in their working directory and leave it there all the time, using relative paths as much as possible. Since examples are likely to be run in the middle of an analysis in order to develop the next step in the analysis, leaving my wd elsewhere would be counterproductive. Having a leftover variable that tells me where the data were written is simpler than using setwd to and from the temp directory. More hand holding than that belongs in a tutorial or vignette.

Note that if you wrap the example up in a vignette then changing to the temp directory seems fair to me (if CRAN allows it) because the context is more like  a complete, self-contained analysis.

On May 7, 2019 11:32:49 AM PDT, Henrik Bengtsson <henrik.bengtsson using gmail.com> wrote:
>I agree that setwd(), tempdir(), etc. clutter up examples.  At the
>same time, I respect the CRAN policy - it's conservative approach has
>helped us avoid a wild-west working environment. One approach that
>might help bring some standardization, instead of each package
>developer rolls their own, and avoid having to have that tempdir code
>in there is to have example() handle this.  For example, add a
>workdir=tempdir() argument to example() and have example() change to
>that directory temporarily, and report on the working directory at the
>end.
>
>> example(foo)
>...
>Any output files produced by this example may be found in the
>temporary working directory '/tmp/alice/Rtmp4faDCv'.
>
>This could be a message or a warning.  One could also imagine using
>workdir=getOption("example.workdir", tempdir()) as the default.  If
>properly done, this would allow the user to set
>options(example.workdir = quote(getwd())) if they'd like to write to
>the current directory.
>
>What it does not solve is when a user cut'n'paste code, but maybe that
>could be considered acceptable because in that case the user is not
>running/source code, but calling it themself.
>
>My $.02
>
>/Henrik
>
>
>
>
>These messages can be skipped if called with example(...,
>workdir=getwd()).
>
>On Tue, May 7, 2019 at 10:57 AM Duncan Murdoch
><murdoch.duncan using gmail.com> wrote:
>>
>> On 06/05/2019 12:16 p.m., Jim Hester wrote:
>> > For what it's worth, the recommendation to use `tempfile()` is very
>> > confusing to R users.
>> >
>> > Often users (particularly new users) jump directly to examples when
>> > reading documentation and when you have these more complicated
>> > examples they do not realize they can just use a simple string
>> > literal.
>> >
>> > See https://github.com/tidyverse/readr/issues/635 for an issue
>where
>> > multiple users explicitly request examples which do _not_ use
>> > `tempfile()`.
>>
>> I think beginners rarely like the help pages, and that's really to be
>> expected:  help pages need to be complete and correct, and beginners
>> really need a small subset of possibilities.  Beginners need
>tutorials.
>>
>> In the Github issue, you mentioned pollution of the working
>directory,
>> but I think a bigger problem is destruction of user data, e.g. if a
>user
>> has a file "wine.csv", and the example writes to that file.
>>
>> So you need to do something to protect users.  I think making
>examples a
>> little complicated by using setwd(tempdir()) with a literal filename,
>or
>> writing to tempfile() is worth it.
>>
>> Duncan Murdoch
>>
>> >
>> > On Fri, May 3, 2019 at 7:59 PM Duncan Murdoch
><murdoch.duncan using gmail.com> wrote:
>> >>
>> >> On 03/05/2019 6:33 p.m., Jarrett Phillips wrote:
>> >>> Hello,
>> >>>
>> >>> My R package has a function with an argument to specify whether
>numerical
>> >>> results should be outputted to a CSV file.
>> >>>
>> >>> CRAN policy stipulates verbatim that
>> >>>
>> >>> Packages should not write in the user’s home filespace (including
>> >>> clipboards), nor anywhere else on the file system apart from the
>R
>> >>> session’s temporary directory (or during installation in the
>location
>> >>> pointed to by TMPDIR: and such usage should be cleaned up).
>Installing into
>> >>> the system’s R installation (e.g., scripts to its bin directory)
>is not
>> >>> allowed.
>> >>>
>> >>> I know I should use tempdir() within my package function, but
>I've not seen
>> >>> any examples on how this is best done within existing R packages.
>> >>>
>> >>> Within my package documentation examples for my function, I have
>the lines:
>> >>>
>> >>>    \dontshow{.oldwd <- setwd(tempdir())}
>> >>>
>> >>> ... some R code ...
>> >>>
>> >>> \dontshow{setwd(.oldwd)}
>> >>>
>> >>> but I have been informed that this is not the accepted way.
>> >>>
>> >>> Any ideas from the community on how do do this properly?
>> >>
>> >> Use the tempfile() function to generate a filename in the
>temporary
>> >> directory.  You might want to use the "pattern" or "fileext"
>arguments,
>> >> but don't change the "tmpdir" argument.
>> >>
>> >> Then write to that file.
>> >>
>> >> For example,
>> >>
>> >> filename <- tempfile(fileext = ".csv")
>> >> write.csv(df, filename)
>> >>
>> >> It's a good idea to clean up afterwards using
>> >>
>> >> unlink(filename)
>> >>
>> >> ______________________________________________
>> >> R-package-devel using r-project.org mailing list
>> >> https://stat.ethz.ch/mailman/listinfo/r-package-devel
>>
>> ______________________________________________
>> R-package-devel using r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-package-devel
>
>______________________________________________
>R-package-devel using r-project.org mailing list
>https://stat.ethz.ch/mailman/listinfo/r-package-devel

-- 
Sent from my phone. Please excuse my brevity.