This release overhauls the layout of the reproducibility report, adds support for GitLab, introduces some safety checks and warnings related to caching, provides more documentation, and more.
While it has always been possible to host workflowr projects using platforms other than GitHub, it was less convenient and not all the features were supported. With this release, a workflowr project hosted on GitLab has all the same features as a workflowr project hosted on GitHub, including links to past versions of the R Markdown, HTML, and figure files. It’s also possible to use workflowr with GitHub Enterprise or a custom GitLab instance, but you’ll have to coordinate with your administrators to make sure it’s possible to host the website.
domain
to wflow_git_remote()
to allow specifying any remote Git host, e.g. domain = "gitlab.com"
wflow_use_gitlab()
to automate GitLab setupThe layout of the reproducibility report and other content that workflowr inserts in the HTML output has been overhauled to be both highly informative but also collapsed by default. This way the information is there if you need it, but otherwise is minimally distracting.
wflow_use_github()
or wflow_use_gitlab()
to insert a link to your workflowr project into the navigation barA popular knitr/rmarkdown feature is caching slow-running chunks. This can be problematic for workflowr because it assumes that the results are newly created when wflow_publish()
publishes the results with a given version of the code. In this release, workflowr now provides warnings, safety checks, and some convience arguments for safely using caching.
cache=TRUE
) but is not set to re-run if its upstream chunks are changed (autodep=FALSE
)clean_fig_files
to wflow_build()
. The default for wflow_build()
is FALSE
, so that old figures are not removed. This is useful for iterative development when plots from cached chunks may not be regenerated during a build. However, clean_fig_files
is fixed to TRUE
for wflow_publish()
to ensure that the final results are produced during the build (suggested by @lazappi, #113)delete_cache
to wflow_build()
/wflow_publish()
. The defult is FALSE
, but if set to TRUE
it will delete the cache directory prior to building each R Markdown file. This helps ensure reproducibility of the published resultswflow_build()
send message about status of cache directoryIn addition to the new vignette on GitLab, this release has multiple other new vignettes plus updates to existing ones.
wflow_use_github()
fig_path_ext: false
in _workflowr.yml
removes the file extension from the figure subdirectories, allowing them to be viewed on Shiny Server (implemented by @Tutuchan, #119, #122)fig.path
, which workflowr ignores (idea from @lazappi, #114)disable_remote
to wflow_start()
. It creates a Git pre-push hook that disables the ability to push to a remote repository. Useful for confidential projects. Currently only available for Linux and macOS (suggested by @rgayler, #141)wflow_html()
to facilitate integrating workflowr features into other R Markdown output formats such as blogdown (suggested by @docmanny, #126)wflow_rename()
to rename files and directories, including committing the change with Gitwflow_publish()
/wflow_git_commit()
fail early if any of the files have merge conflictswflow_git_push()
set the upstream tracking branch by default (see git2r Issue 375)wflow_build()
report the current working directory. If the knit directory (where the code is executed) is different than the working directory, have wflow_build()
report where the code in each file is being executedFix unit test on CRAN Windows servers.
workflowr 1.1.0 is a maintenance release. It includes some minor new features, improved error handling, and bug fixes. Critically, it makes workflowr compatible with the latest release of git2r.
The most noticeable changes are 1) wflow_publish()
/wflow_status()
are now much faster, especially if your project has many R Markdown files and/or many Git commits, 2) the rmarkdown package is no longer loaded automatically when you load workflowr, and 3) the new function wflow_open()
(based on a previous version in workflowrBeta) to open new or existing R Markdown files.
wflow_open()
. Since there is no longer a workflowr template in workflowr 1.0+, it creates a file with a minimal YAML header (@xiangzhu, #107)verbose
to wflow_build()
/wflow_publish()
to display the build log in the R console as each file is built. Useful for monitoring long-running code chunks (idea from @pcarbo)dry_run
added to wflow_start()
wflow_start()
wflow_view()
:
recent
to latest
to display the most recently modified HTML filefiles
now requires correct paths to existing R Markdown or HTML files. Previously, wflow_view()
would guess the correct path. While convenient, it was also potentially confusing since it was inconsistent with the other workflowr functionslatest
is no longer mutually-exclusive with files
. If the most recently modified HTML is different than those specified by files
, they will all be opened for viewingbrowseURL()
if getOption("browser")
does not provide a default optionwflow_status()
/wflow_publish()
by using git2r::odb_blobs()
to obtain past commit times of files (if these functions are still slow for you, try running git gc
in the Terminal)include
/exclude
from the template _site.yml
. Also remove the file analysis/.nojekyll
(related to point above). Unlike rmarkdown:::default_site
, wflow_site()
does not copy the entire directory to the output directory, so these fields have no effect. See the R Markdown documentation on Included files for more informationFix Windows-specific bug in wflow_publish()
, wflow_remove()
, and wflow_git_commit()
. The bug prevented Windows users from running these functions from a subdirectory of a workflowr project
Restore all files in website directory if wflow_publish()
fails to build any of the files (reported by @pcarbo)
If a user has not set the Git configuration variables user.name
and user.email
, any workflowr function that creates a Git commit will throw an informative error message. Previously this was only done for wflow_start()
, but has been expanded to wflow_git_commit()
, wflow_publish()
, and wflow_remove()
(idea from @pcarbo)
Ensure that wflow_build()
creates a new .nojekyll
file if necessary and that wflow_publish()
commits it. This is most useful when changing the name of the output directory (#72)
Various minor changes to documentation, tests, and package infrastructure to prepare for CRAN submission.
The 1.0.0 release is a major change. Instead of relying on the external code chunks in chunks.R
to implement the reproducibility features, workflowr 1.0.0 replaces chunks.R
with the output format wflow_html()
and site generator wflow_site()
. This enables a much more informative reproducibility report to be created.
# Install from GitHub
devtools::install_github("jdblischak/workflowr")
# Start a new project to test out the new features
wflow_start("testproject")
# Learn about the new ways to customize your site
?wflow_html
# Update an existing project
wflow_update()
wflow_html()
and wflow_site()
to overhaul the reproduciblity features of workflowrwflow_commit()
-> wflow_git_commit()
wflow_remotes()
-> wflow_git_remote()
wflow_open()
and wflow_convert()
wflow_update()
to udpate a pre-1.0 workflowr project to a post-1.0 projectcreate_links_page()
(not widely used, if at all)wflow_build()
or wflow_publish()
is ignored if using wflow_html()
wflow_publish(republish = TRUE)
does not build HTML files that have unstaged/staged changeswflow_build()
reports the total number of files that will be builtwflow_start()
to set local Git user.name and user.email. Preferred method for most users is still to set global Git user.name and user.email with wflow_git_config()
.wflow_publish()
now views the website by default if run interactively (view = interactive()
) just like wflow_build()
Pin the dependency git2r to version 0.21.0 or lower because of the forthcoming breaking changes in the git2r API. This is not a great solution because install.packages()
always installs the latest version, which will cause an error in the installation. If you are having trouble with this, first install git2r with devtools::install_version("git2r", "0.21.0")
and then retry installing workflowr. If your machine is running macOS or Windows, you may need to run install_version("git2r", "0.21.0", type = "binary")
.
Fix minor bug that affected the error message produced for a failed push or pull using the SSH protocol
When wflow_git_push()
or wflow_git_pull()
fails for an unknown reason, the exact error message from git2r::push()
or git2r::pull()
is reported to faciliate troubleshooting
Multiple other internal changes to make workflowr more robust
Fix bug that caused wflow_status()
(and thus wflow_publish()
) to ignore R Markdown files with the all lowercase file extension .rmd
(reported by @LSun and @pcarbo). This was recently introduced in commit a8ce711, so it only affected versions 0.10.0 and 0.10.1.
Fix bug that prevented deleted files from being committed with wflow_commit()
or wflow_publish()
(reported by @pcarbo)
wflow_view()
(or wflow_build()
) now opens the website in the RStudio Viewer if run from within RStudiowflow_build()
automatically removes unused figure fileswflow_build()
argumentstempdir()
instead of to /tmp/workflowr
. This prevents failures due to permission issues when mutliple workflowr users try to use the same machine (e.g. a compute node on a HPC cluster) (#86)wflow_build()
to build the R Markdown files in an external R process, it now runs render_site()
in the global environment of the external R process. This better mimics the results of the RStudio “Knit” button.wflow_git_config()
sets the Git options user.name
and user.email
wflow_git_push()
pushes changes from the local computer to GitHubwflow_git_pull()
pulls the changes from GitHub to the local computerwflow_build()
automatically opens the website after building files, thus it’s not necessary to always run wflow_view()
manually afterwards. If one file is built, then that file is opened. If more than one file is built, then the index page is opened. (#41)wflow_build()
adds objects to the global environment when building files locally in the R console (i.e. local = TRUE
) (#53)wflow_remove()
removes R Markdown files and all their associated files ( e.g. HTML and figures)git rm
)wflow_remove()
also works on non-Rmd fileswflow_update()
and wflow_convert()
fail gracefully if the diff
utility is not available and inform the user to install RtoolsThis minor release fixes a bug in how filepaths are resolved so that workflowr can run on Windows.
This release changes the interface of some of the main workflowr functions. The functions wflow_publish()
and wflow_status()
are introduced, and the previous functions wflow_build()
and wflow_commit()
are re-designed.
New function wflow_status()
reports which analysis files require user attention (inspired by git status
)
Defines R Markdown files whose HTML has been committed as part of the Git repo as “Published”, R Markdown files which have been committed to the Git repo but not their HTML as “Unpublished”, and R Markdown files that are untracked by Git as “Scratch”. Furthermore, previously published files that have been subsequently edited are classified as “Modified”. See the man page ?wflow_status
for more details.
This new function replaces the previous functionality of wflow_commit()
. The basic interface is much simpler.
wflow_publish("analysis/file.Rmd")
will 1) commit analysis/file.Rmd
, 2) build analysis/file.Rmd
in its own separate R session with set.seed(12345)
, and 3) commit docs/file.html
and any generated figures. These 3 steps are referred to as “publishing a file”.
wflow_publish(all = TRUE)
will publish all tracked analysis files, analogous to git commit -a
.
To change the theme or make some other change to the entire site, run wflow_publish("analysis/_site.yml", republish = TRUE)
, which will 1) commit the configuration file analysis/_site.yml
, 2) re-build all the previously published analysis files using the new configuration options, and 3) commit the re-built HTML files.
By default, wflow_build()
runs in “Make”-mode, only building R Markdown files that have been updated more recently than their corresponding HTML files. If instead files are specfically stated, those files will be built.
By default, R Markdown files are now built each in their own separate R session (similar in function to the “Knit HTML” button in RStudio). This prevents undesirable behavior like sharing variables and loaded packages across separate files. Set local = TRUE
to build the files in the local R console (useful for debugging, but otherwise discouraged).
By default, the seed for random number generation is set to the arbitrary number 12345
(using set.seed()
). This ensures that any code that produces random numbers will be reproducible.
wflow_commit()
has been demoted to only being a wrapper for the equivalent functionality of git add
and git commit
. This can be useful for committing non-analysis files or R Markdown files that you aren’t ready to publish yet. However, you should use wflow_publish()
for the standard workflow.
Set all = TRUE
to run the equivalent of git commit -a
.
wflow_update()
to better explain its functionality. It will attempt to convert all R Markdown files present to use the latest version of the workflowr R Markdown template; however, it will only commit R Markdown files that are tracked by Git.All workflowr functions should now accept the file extesion .rmd
in addition to .Rmd
(Issue #10)
Replaced the shared argument path
with project
to clarify that this argument specifies the path to a directory in the workflowr project
wflow_start()
now includes docs/.nojekyll
Change dependency to R >= 3.2.5 (Issue #32)
Change stringr dependency (>= 1.1.0)
Started a vignette with Frequently Asked Questions
Added sections to README (Quick start, Upgrading, Related work, and Citation)
wflow_start()
API has been simplified. Now the first (and only required) argument is the directory to be created. By default, the name of the project will be determined from the directory name. For example, wflow_start("~/projects/proj-x")
will create a workflowr project with the name “proj-x”. To specify a different name, provide the argument name
change_wd = TRUE
changes the working directory to the newly created project.existing = FALSE
is a safety feature so that by default workflowr projects are only created in new directorieswflow_start()
no longer adds “BuildType: Website” to RStudio project file. Users should run wflow_build()
instead. wflow_update()
removes the build specification for existing workflowr projectswflow_convert()
to convert an R Markdown file based on the previous template to use the latest versionstandalone = TRUE
for wflow_open()
)path = NULL
wflow_update()
function updates workflowr projects to use the latest version of the templatewflow_convert()
converts an existing R Markdown file to use the workflowr template (can also update a file that uses a previous version of the template)wflow_view()
opens the website in the browserwflow_remotes()
manages remote Git repositories on GitHub. It can add a remote, remove a remote, or update the URL for an existing remoteSecond alpha release. Major API overhaul.
Improved naming of functions: start_project
-> wflow_start
, open_rmd
-> wflow_open
, build_site
-> wflow_build
, commit_site
-> wflow_commit
, create_results
-> create_links_page
wflow_commit
can optionally add and commit provided files (argument is commit_files
) before re-building website and commiting HTML files
wflow_open
accepts multiple filenames
Both wflow_build
and wflow_commit
have the argument all
to optionally build all the HTML files
First alpha release. Contains enough functions to create workflowr project from scratch.
Main functions: start_project
, open_rmd
, build_site
, commit_site
, create_results