this.path aims to provide a mechanism for retrieving the path of an R script within itself without needing to explicitly write its path elsewhere. Additionally, it provides a mechanism for specifying file paths against the executing script’s directory (without changing the working directory).
Install it from CRAN:
install.packages("this.path")Install the development version from GitHub:
install.packages("this.path",
repos = "https://raw.githubusercontent.com/ArcadeAntics/PACKAGES")The most important functions from this.path are
this.path(), this.dir(), here(),
and this.proj():
this.path() returns the normalized path of the
executing script.
this.dir() return the normalized path of the
directory in which the executing script is located.
here() constructs file paths against the executing
script’s directory.
this.proj() constructs file paths against the
executing project’s directory instead of the executing script’s
directory.
this.path also provides functions for constructing and manipulating file paths:
path.join(), basename2(), and
dirname2() are drop in replacements for
file.path(), basename(), and
dirname() which better handle drives and network
shares.
splitext(), removeext(),
ext(), and ext<-() split a path into root
and extension, remove a file extension, get an extension, or set an
extension for a file path.
path.split(), path.split.1(), and
path.unsplit() split the path to a file into
components.
relpath() and rel2here() turn absolute
paths into relative paths.
New additions to this.path include:
LINENO() returns the line number of the executing
expression in the executing script.
wrap.source(), inside.source(),
set.this.path(), and unset.this.path()
implement this.path() for any source()-like
functions outside of the builtins.
shFILE() looks through the command line arguments,
extracting FILE from either of the following:
-f FILE or --file=FILE
The only equivalent to this.path() (that I have seen) is
whereami::thisfile(). R package whereami
has many issues that this.path resolves:
this.path() works with sys.source(),
debugSource() in ‘RStudio’,
testthat::source_file(), knitr::knit(),
box::use(), and compiler::loadcmp(). It also
works in interactive mode inside ‘Rgui’, ‘RStudio’, ‘VSCode’, and
‘Jupyter’.
Related to source(), this.path():
2.1. takes argument chdir into account.
2.2. recognizes that file = "",
file = "clipboard", and file = "stdin" are not
referring files and skips them.
2.3. accounts for file being a URL pathname.
2.4. accounts for file being a connection instead of a
character string.
2.5. skips calls in which file is missing and
exprs is used instead.
if an R script is being run from a shell,
this.path() does a better job of extracting the command
line argument -f FILE or --file=FILE.
this.path() saves all normalized paths within their
appropriate environments, making it much faster subsequent times within
the same script, and independent of working directory.
if this.path() does not find an executing script, it
throws an error. This is better than whereami::thisfile()
which returns NULL when it cannot find the executing
script. If the executing script cannot be found, obviously there is a
problem that needs to be addressed, so the script should not
continue.
whereami also has some objectively incorrect coding issues:
It treats R and Rscript as two separate applications to look for command line arguments. At least since R 2.5.0 (>= 15 years ago), Rscript directly calls R, so there should not be separate cases. Additionally, it does NOT take into account the differences between the command line applications on Windows and under Unix-alikes.
It checks for uses of knitr::knit() after checking
the entire call stack for source() and after checking the
command line arguments. This is incorrect, it should be checking for
knitr::knit() at the same time as it checks for
source().
When whereami is loaded or attached, it changes
option keep.source to TRUE. A package should
never be changing global options without asking / /
informing the user beforehand. This may not harm your code, but it is
still bad practice that should be avoided, especially for a CRAN
package.
When you call whereami::thisfile(), it forces
knitr to be loaded as well. Again, it is changing your
environment without permission. It should use something like
isNamespaceLoaded("knitr") instead of
requireNamespace("knitr") because, contrary to what the
package claims, whereami enhances
knitr rather than requires it.
Code aside, whereami suggests that an R script needing to know its own path should only be done if absolutely necessary, and that it should be set outside the context of the R script if possible. I find this vague and unconvincing. Other scripting languages have methods of requesting a script’s path without issue, so R should too. We should not be scaring programmers into thinking that this is rule-breaking or bad practice.
If you are not using this.path for the function
this.path(), you are probably using it for
this.path::here(). The only equivalent (that I know of) is
R package here with its function
here::here(). this.path provides a
mechanism for specifying a path relative to the executing script’s
directory, while here provides a mechanism for specifying a path
relative to the project’s directory.
As a long time here user, I appreciate all the functionality and convenience this package offered, but it has some faults that I could not ignore (which this.path fixes). I do not dislike here, it just no longer fits my use-cases.
here::here() returns the initial working directory
when it cannot find the project’s directory. If the project’s directory
cannot be found, there is something wrong and an error should be thrown,
but it doesn’t, and this leads to incorrect / / unexpected
behaviour.
here does not work when the initial working directory is set outside the context of the project. Occasionally, I would set the working directory to a temporary directory where my R script would create a log file of the details of the script. This leads to here not being able to find the project’s directory and incorrectly returning the initial working directory.
here does not work for projects containing sub-projects. In my scenario, I had a project “A”, and later I had a project “B” of which “A” is a sub-project. I would run a script in “B” which runs another script in “A”, but the project root is already set to the root of “B”, so the script in “A” fails.
here does not work when a project is stored on the internet. When I say on the internet, I am not referring to a project stored on a network share, that works correctly. I am talking about projects uploaded to a website. The work I do requires that all source code and input be publicly available through our website, and that the code should run as expected (slow as it may be, and only code that exclusively reads files and / / or produces graphics). Since the project root of here cannot be a URL pathname, I cannot use it.
It should be noted that if you prefer specifying files relative to
the project’s directory instead of the executing script’s directory, you
could use this.path::this.proj() instead. It behaves very
similarly to here::here(), but can handle multiple projects
in use at once.
There are a few alternatives to this.path(), though they
are very limiting.
You could always change the working directory to the directory of the executing script before running it. This would be
cd /path/to
Rscript ./file.Ror
source("/path/to/file.R", chdir = TRUE)This works for a lot of use-cases, except for interactive use where you probably are not changing the working directory as you move amongst files, and it will not if your R script is stored on a website. Sometimes it is convenient to have the working directory set elsewhere. This also means that you cannot make R scripts that act like executables. If you need to call other scripts in the same directory, you could not call those other scripts without the original script’s path.
utils::getSrcFilename()You could use utils::getSrcFilename(). Everywhere you
would use this.path(), use
utils::getSrcFilename(function() NULL, full.names = TRUE)
instead (yes, it is quite lengthy), and everywhere you would use
this.dir(), use
utils::getSrcDirectory(function() NULL) (again, quite
lengthy).
While this will work for R scripts uploaded to a website, this will
not work in interactive use since you have to source() your
scripts. Also, it means you must set option keep.source to
TRUE; once again this may not be a big deal, but something
that may cause issues. This means you could not run your R scripts from
a shell ever again, making it an incredibly inconvenient substitute.
If you think I have overlooked something in whereami or here, or think there are any improvements I could make to this.path, please let me know, I am open to all suggestions! And I hope this package serves you well!