A cross-platform interface to file system operations, built on top of the 'libuv' C library.
fs provides a cross-platform, uniform interface to file system operations. It shares the same back-end component as nodejs, the libuv C library, which brings the benefit of extensive real-world use and rigorous cross-platform testing. The name, and some of the interface, is partially inspired by Rust’s fs module.
You can install the released version of fs from CRAN with:
And the development version from GitHub with:
fs functions smooth over some of the idiosyncrasies of file handling with base R functions:
Vectorization. All fs functions are vectorized, accepting multiple paths as input. Base functions are inconsistently vectorized.
Predictable return values that always convey a path. All fs functions return a character vector of paths, a named integer or a logical vector, where the names give the paths. Base return values are more varied: they are often logical or contain error codes which require downstream processing.
Explicit failure. If fs operations fail, they throw an error. Base functions tend to generate a warning and a system dependent error code. This makes it easy to miss a failure.
UTF-8 all the things. fs functions always convert input paths to UTF-8 and return results as UTF-8. This gives you path encoding consistency across OSes. Base functions rely on the native system encoding.
Naming convention. fs functions use a consistent naming
convention. Because base R’s functions were gradually added over
time there are a number of different conventions used (e.g.
fs functions always return ‘tidy’ paths. Tidy paths
/to delimit directories
Tidy paths are also coloured (if your terminal supports it) based on the
file permissions and file type. This colouring can be customised or
extended by setting the
LS_COLORS environment variable, in the same
output format as GNU
fs functions are divided into four main categories:
path_for manipulating and constructing paths
Directories and links are special types of files, so
will generally also work when applied to a directory or link.
library(fs)# Construct a path to a file with `path()`path("foo", "bar", letters[1:3], ext = "txt")#> foo/bar/a.txt foo/bar/b.txt foo/bar/c.txt# list files in the current directorydir_ls()#> CRAN-RELEASE DESCRIPTION LICENSE.md#> NAMESPACE NEWS.md R#> README.Rmd README.md _pkgdown.yml#> appveyor.yml bar check.R#> codecov.yml cran-comments.md docs#> example follow.R fs.Rcheck#> fs.Rproj fs_184.108.40.20600.tar.gz inst#> man man-roxygen script.R#> src tests# create a new directorytmp <- dir_create(file_temp())tmp#> /var/folders/dt/r5s12t392tb5sk181j3gs4zw0000gn/T/RtmpM84M38/file16f7126ece07c# create new files in that directoryfile_create(path(tmp, "my-file.txt"))dir_ls(tmp)#> /var/folders/dt/r5s12t392tb5sk181j3gs4zw0000gn/T/RtmpM84M38/file16f7126ece07c/my-file.txt# remove files from the directoryfile_delete(path(tmp, "my-file.txt"))dir_ls(tmp)#> character(0)# remove the directorydir_delete(tmp)
fs is designed to work well with the pipe, though because it is a minimal-dependency infrastructure package it doesn’t provide the pipe itself. You will need to attach magrittr or similar.
library(magrittr)paths <- file_temp() %>%dir_create() %>%path(letters[1:5]) %>%file_create()paths#> /var/folders/dt/r5s12t392tb5sk181j3gs4zw0000gn/T/RtmpM84M38/file16f713ca22ebf/a#> /var/folders/dt/r5s12t392tb5sk181j3gs4zw0000gn/T/RtmpM84M38/file16f713ca22ebf/b#> /var/folders/dt/r5s12t392tb5sk181j3gs4zw0000gn/T/RtmpM84M38/file16f713ca22ebf/c#> /var/folders/dt/r5s12t392tb5sk181j3gs4zw0000gn/T/RtmpM84M38/file16f713ca22ebf/d#> /var/folders/dt/r5s12t392tb5sk181j3gs4zw0000gn/T/RtmpM84M38/file16f713ca22ebf/epaths %>% file_delete()
Filter files by type, permission and size
dir_info("src", recursive = FALSE) %>%filter(type == "file", permissions == "u+r", size > "10KB") %>%arrange(desc(size)) %>%select(path, permissions, size, modification_time)#> # A tibble: 10 x 4#> path permissions size modification_time#> <fs::path> <fs::perms> <fs::bytes> <dttm>#> 1 src/RcppExports.o rw-r--r-- 655.5K 2018-05-20 17:39:19#> 2 src/dir.o rw-r--r-- 442.7K 2018-05-20 17:39:19#> 3 src/fs.so rwxr-xr-x 435.3K 2018-05-20 17:39:29#> 4 src/id.o rw-r--r-- 383.2K 2018-05-20 17:39:18#> 5 src/file.o rw-r--r-- 347.5K 2018-05-20 17:39:18#> 6 src/path.o rw-r--r-- 257.4K 2018-05-20 17:39:18#> 7 src/link.o rw-r--r-- 224.3K 2018-05-20 17:39:18#> 8 src/utils.o rw-r--r-- 117.9K 2018-05-20 17:39:18#> 9 src/error.o rw-r--r-- 17.3K 2018-05-20 17:39:15#> 10 src/RcppExports.cpp rw-r--r-- 11.4K 2018-05-20 17:39:14
Tabulate and display folder size.
dir_info("src", recursive = TRUE) %>%group_by(directory = path_dir(path)) %>%tally(wt = size, sort = TRUE)#> # A tibble: 54 x 2#> directory n#> <fs::path> <fs::bytes>#> 1 src 2.86M#> 2 src/libuv 2.44M#> 3 src/libuv/src/unix 1.08M#> 4 src/libuv/autom4te.cache 1.08M#> 5 src/libuv/test 865.36K#> 6 src/libuv/src/win 683.14K#> 7 src/libuv/docs/src/static 328.32K#> 8 src/libuv/m4 319.95K#> 9 src/libuv/include 192.33K#> 10 src/libuv/docs/src/static/diagrams.key 184.04K#> # ... with 44 more rows
Read a collection of files into one data frame.
dir_ls() returns a named vector, so it can be used directly with
# Create separate files for each speciesiris %>%split(.$Species) %>%map(select, -Species) %>%iwalk(~ write_tsv(.x, paste0(.y, ".tsv")))# Show the filesiris_files <- dir_ls(glob = "*.tsv")iris_files#> setosa.tsv versicolor.tsv virginica.tsv# Read the data into a single table, including the filenamesiris_files %>%map_df(read_tsv, .id = "file", col_types = cols(), n_max = 2)#> # A tibble: 6 x 5#> file Sepal.Length Sepal.Width Petal.Length Petal.Width#> <chr> <dbl> <dbl> <dbl> <dbl>#> 1 setosa.tsv 5.1 3.5 1.4 0.2#> 2 setosa.tsv 4.9 3 1.4 0.2#> 3 versicolor.tsv 7 3.2 4.7 1.4#> 4 versicolor.tsv 6.4 3.2 4.5 1.5#> 5 virginica.tsv 6.3 3.3 6 2.5#> 6 virginica.tsv 5.8 2.7 5.1 1.9file_delete(iris_files)
We hope fs is a useful tool for both analysis scripts and packages. Please open GitHub issues for any feature requests or bugs.
In particular, we have found non-ASCII filenames in non-English locales on Windows to be especially tricky to reproduce and handle correctly. Feedback from users who use commonly have this situation is greatly appreciated.
This is a small bugfix only release.
file_move() now fall back to copying, then removing files when moving files
between devices (which would otherwise fail) (#131, https://github.com/r-lib/usethis/issues/438).
Fix for a double free when using
warn = TRUE (#132)
path_wd() generates paths from the current working directory (#122).
path_has_parent() determines if a path has a given parent (#116).
file_touch() used to change access and modification times for a file (#98).
file_info() gain a
fail parameter, to signal warnings rather than errors if they are called on
a path which is unavailable due to permissions or locked resources (#105).
path_tidy() now always includes a trailing slash for the windows root
path_ext_remove() now handle paths with
non-ASCII characters (#120).
fs_path objects now print (without colors) even if the user does not have
permission to stat them (#121).
compatibility with upcoming gcc 8 based Windows toolchain (#119)
Experimental support for
+ methods for
fs_path objects (#110).
dir_create() now take
..., which is passed to
path() to make construction a little nicer (#80).
path_ext_remove() now handle paths with
directories including hidden files without extensions (#92).
file_copy() now copies files into the directory if the target is a
fs_perm objects now use
so that S4 dispatch to their base classes works as intended (#91).
Fix allocation bug in
delete  when we should have
startargument, to specify where the absolute path should be calculated from (#87).
path_ext() now returns
character() if given 0 length inputs (#89)
Fix for a memory issue reported by ASAN and valgrind.
path_home() now use
HOMEPATH as the user home directory on Windows. This differs
from the definition used in
path.expand() but is consistent with
definitions from other programming environments such as python and rust. This
is also more compatible with external tools such as git and ssh, both of
which put user-level files in
USERPROFILE by default. To mimic R's (and
previous) behavior there are functions
Handling missing values are more consistent. In general
FALSE for missing values,
path_* functions always propagate
NA values (NA inputs become NA outputs) and
functions error with NA inputs.
fs functions now preserve tildes in their outputs. Previously paths were
always returned with tildes expanded. Users can use
path_expand() to expand
tildes if desired.
file_chmod()is now vectorized over both of its arguments (#71).
link_create()now fails silently if an identical link already exists (#77).
path_package()function created as an analog to
system.file()which always fails if the package or file does not exist (#75)
Tidy paths no longer expand
Filesystem modification functions now error for NA inputs. (#48)
path() now returns 0 length output if given any 0 length inputs (#54).
dir_delete() now correctly expands paths (#47).
dir_delete() now correctly deletes hidden files and directories (#46).
link_path() now checks for an error before trying to make a string,
avoiding a crash (#43).
libuv return paths now marked as UTF-8 strings in C code, fixing encoding issues on windows. (#42)
dir_copy() now copies the directory inside the target if the target is a
dir_copy() now works correctly with absolute paths and no longer removes
overwrite = TRUE.
NEWS.mdfile to track changes to the package.