Skip to content

Latest commit

 

History

History
626 lines (453 loc) · 27.6 KB

NEWS.md

File metadata and controls

626 lines (453 loc) · 27.6 KB
Raw

Packrat 0.6.0 (UNRELEASED)

  • BREAKING CHANGE: The default Packrat cache directory has changed, and now includes an R version suffix. This helps avoid populating a single cache with versions of packages compiled for different versions of R (as there is no guarantee that packages compiled for e.g. R 3.4.x will work with R 3.5.x). If you'd like to migrate your old Packrat cache directory, you can move the directory at dirname(packrat:::appDataDir()) to packrat:::appDataDir().

  • Packrat no longer attempts to unload packages loaded from the user library after calling packrat::init(), as this could fail in a myriad of cases. (To ensure proper isolation of the session, you should still restart R after invoking packrat::init()).

  • Packrat now properly annotates the RemoteType field when downloading packages from GitLab and Bitbucket. (#564)

  • Packrat no longer removes old source packages during a restore. (#560)

  • Added support for GitLab: packages downloaded from GitLab can now be restored by Packrat. (#562, @akgold)

  • Fixed an issue where tangled R code chunks containing invalid R code prevented Packrat from finding any dependencies. Packrat will now look for package dependencies within each code chunk independently. (#551)

  • Packrat no longer sets LC_ALL=C when building source packages, as this can lead to errors when building packages containing non-ASCII text. (#545)

  • Fixed an issue where ignored packages would still be queried by packrat::unused_packages(), which affected other APIs like packrat::clean(). (#525)

  • Fixed an issue where newly-added project options did not get their correct default value when no entry existed within the packrat.opts file. (#496)

Packrat 0.5.0

  • Packrat now supports both of BiocManager and BiocInstaller (as used for discovering the Bioconductor repositories active for the current project). BiocManager will be used for R >= 3.6.0; BiocInstaller will be used otherwise.

  • The R option packrat.dependency.discovery.disabled can be set to TRUE to disable dependency discovery in projects. This can be useful if you find Packrat's dependency discovery is slow (as it can be in projects containing a large number of R Markdown files). (#513, @ras44)

  • The scheme used for hashing packages that enter the Packrat cache has changed -- now, a defined ordering of fields is used when hashing a package's DESCRIPTION file. Note that this implies a package may need to be re-cached on restore, in the case that its hash has changed. This change should not affect any existing packages in the cache. (#505, @aronatkins)

  • packrat::with_extlib() now works with no packages provided; both with and without this option, the new behavior is that expr is executed in an environment where the original (not packrat) library search path is in place.

  • A project is now only considered 'packified' if it has both a Packrat lockfile as well as the associated autoloader in the project .Rprofile.

  • Calling packrat::init() on a project that already contains a Packrat lockfile no longer attempts to re-snapshot and restore the project.

  • Packrat now supports R packages available on BitBucket, courtesy of a PR from @mariamedp. (#481)

  • Added the project option symlink.system.packages: users can now configure whether base R packages from the system library are symlinked into a private library packrat/lib-R. Disabling this can be useful if you intentionally want Packrat to use packages that have been installed into the system library.

  • Fixed an issue where attempts to snapshot could fail when the R libraries live on a network drive.

Packrat 0.4.9-3

  • Adjusted unit tests to accommodate new CRAN package checks.

  • Fixed an issue where packrat::repos_upload() would fail to re-compress uploaded tarballs. (#474)

Packrat 0.4.9-2

  • Fix a regression where attempts to download packages from GitHub when devtools is not available could fail. (#464)

Packrat 0.4.9-1

  • Fix test errors on CRAN.

Packrat 0.4.9

  • Packrat now understands how to install R packages from private GitHub repositories. (The GITHUB_PAT environment variable should be set with an access token that provides access to the associated repositories.) (#449, #448, @ras44)

  • Packrat gained the get_lockfile_metadata() and set_lockfile_metadata() functions, for changing metadata associated with a particular Packrat project; e.g. the active R version, or the active set of repositories. (#429, @cderv)

  • Packrat now ignores all source files within a packrat/ directory when inferring dependencies, not just the packrat/ directory discovered at the top level. (#385)

  • Packrat no longer includes the .Rhistory file when creating bundles. (#401)

  • Packrat now properly handles the plain 'http' protocol when using versions R >= 3.2. (We now set options(download.file.method = "wininet") rather than options(download.file.method = "internal") in such cases.)

  • The infer.dependencies argument can now be used to switch off the scanning of code for dependencies when using packrat::init() and packrat::snapshot().

  • Packrat no longer fails to download the current version of a package if the binary and source branches of the active repositories are out of sync.

  • Packrat now attempts to parse scripts using UTF-8 encoding in addition to the system encoding. This should primarily help users on Windows who (rightly) save their documents using UTF-8 encoding rather than the default system encoding.

  • Packrat now screens out empty package names discovered during package dependency discovery. (#314)

  • The Packrat global cache is now enabled on Windows. Junction points (rather than symbolic links) are used to populate entries in the private Packrat library.

  • The 'lib-R' and 'lib-ext' directories now use architecture-specific sub-directories for their libraries. This should further help in cases where multiple versions of R are operating within a Packrat project at the same time.

  • Packrat now better handles cases where non-symlink files find their way into the 'lib-R' and 'lib-ext' folders.

  • Packrat now better handles packages that contain trailing newlines in their DESCRIPTION file -- it now avoids inserting intervening blank lines between records, which should resolve the 'Error: contains a blank line' error that can occur during package installation.

  • Packrat now only checks for the 'BiocInstaller' package within the Packrat private library, when attempting to ascertain whether Bioconductor is used by a particular project.

  • Packrat no longer automatically restores projects on startup when Packrat is not detected within the library directory.

  • Packrat now more eagerly caches packages during packrat::restore() -- packages will be immediately cached following successful installation, rather than at the end of the restore process. (#324)

  • Packrat now better handles cases where multiple R processes attempt to write the same package to the cache at the same time. (#333)

  • Packrat now properly checks whether a package exists in the cache before attempting to copy / symlink that package to the active library directory. (#335)

  • Users can now control whether Packrat snapshots sources by setting the R_PACKRAT_SNAPSHOT_SOURCES environment variable to a 'truthy' value, or by setting the R option packrat.snapshot.sources.

  • Packrat now provides APIs for accessing the active paths to resource directories, with:

    • packrat::project_dir(project)
    • packrat::src_dir(project)
    • packrat::lib_dir(project)
    • packrat::bundles_dir(project)

    See ?packrat-resources for more details.

  • Packrat better preserves the pre-existing contents of ignore files. (#332)

Packrat 0.4.8

  • Packrat better handles UTF-8 encoded R Markdown documents on Windows. (#329, @jmcphers)

  • The source directory for a Packrat project can now be over-ridden either by setting the R_PACKRAT_SRC_DIR environment variable, or the 'packrat.src.dir' R option. Clever use of this should allow multiple projects to share a cache of downloaded package sources.

  • The library directory for a Packrat project can now be over-ridden either by setting the R_PACKRAT_LIB_DIR environment variable, or the packrat.lib.dir R option. This can be used if you'd like the library directory for a project to be stored in an alternate location -- this can be useful for Packrat projects stored in e.g. Dropbox folders. A similar treatment is supplied for the Packrat bundles directory, with the R_PACKRAT_BUNDLES_DIR environment variable and the packrat.bundles.dir project option.

  • Packrat now stores the active project as an environment variable, R_PACKRAT_PROJECT_DIR. This should help in situations where R sub-processes are launched while a packrat project is active, and those sub-processes need to access the active packrat library.

  • Packrat now respects the 'repos' field used in the Packrat lockfile when restoring a project. (#316)

  • Packrat better reports download failures, mentioning lack of internet connectivity as a potential culprit. (#306)

  • Packrat no longer treats YAML parse failures (when attempting to scour R Markdown documents for R package dependencies) as fatal errors. (#312)

  • Packrat now first attempts to move folders to the cache, and falls back to a directory copy when this fails. This should improve caching performance in the cases where the Packrat cache lies on the same volume as the active project.

  • packrat::set_opts() gains an argument, 'persist', controlling whether newly set options should be persisted. This allows session-temporary project options to be set.

  • Packrat now only updates ignore files (e.g., '.gitignore') when they have actually changed. (#303, @mdshw5)

  • Packrat now passes '-s' when invoking 'curl', thereby ensuring quiet output when downloading fails. This should resolve issues where attempts to use packrat without internet connectivity caused spurious messages of the form 'curl: (22) The requested URL returned error: 404 Not Found' to be printed.

  • Packrat now passes '-g' when invoking 'curl', thereby disabling curl's globbing parser. This should help ensure that URLs containing e.g. '[' and ']' characters can be successfully downloaded as expected.

  • Packrat now queries 'api.github.com', rather than 'www.github.com', when attempting to download package sources.

  • Packrat now properly respects the 'packrat::opts$ignored.packages()' project option when restoring a project.

  • Packrat handles lockfiles with no packages available. (#294)

  • Fixed an issue where attempting to form junction points to separate drives on Windows could fail. (@raubreywhite, #288)

Packrat 0.4.7

  • Packrat now always uses R's internal tar method when bundling a project, for cross-platform consistency.

  • The 'use.cache' option is disabled on Windows -- this may be revisited in the future once we have a reliable mechanism for detecting whether a particular directory is a reparse point without the use of compiled code within packrat.

  • Packrat uses devtools + httr (when available) to download files and directories from GitHub URLs. This should enable users to allow packrat to access private GitHub repositories as long as the GITHUB_PAT environment variable is set to an appropriate private access token. See ?devtools::install_github for more details on setting up a private access token.

  • The install_github() shim from packrat has been removed -- please use devtools::install_github(), either by taking an explicit dependency on the devtools package, or by loading it from the user library with the packrat option packrat::opts$external.packages("devtools").

  • Packrat is now smarter when managing symlinks within the project library (when package caching is enabled). This should allow multiple R processes to use the same packrat project at the same time. (Previously, there was risk that one R session might clear / refresh symlinks while another process attempted to access them).

  • Packrat no longer erroneously generates recursive symlinks (and attempts to clean up any recursive symlinks discovered as appropriate).

  • Packrat now records the original library paths within its .onLoad() handler, and uses these library paths when attempting to load packages from the user library.

  • Fixed a bug where recursive hashing of a package's LinkingTo dependencies could fail.

  • Fixed a bug where with_extlib() could force a promise in the wrong environment.

Packrat 0.4.6

  • Packrat gains the option snapshot.recommended.packages(), to control whether Recommended packages detected in the system library should become part of the lockfile.

  • Silence noisy 'FAILED' messages that occurred when querying a repository for the existence of a package (in its archives) failed.

  • Fixed a regression where dependencies were not properly discovered when using (R <= 3.1).

  • Fixed an issue where the 'quiet.package.installation()' option could cause a restore failure, if it was not already set for the active project.

Packrat 0.4.5

  • Fixed an issue where the autoloader header in a project's .Rprofile could become duplicated.

  • Packrat now attempts to choose a secure download method when downloading files from https URLs, if a default download method is not already set.

  • The code used for detecting package dependencies has been re-written and refactored, and should properly avoid detecting x as a dependency in e.g. library(x, character.only = TRUE).

  • packrat::snapshot() now updates the active repositories in the lock file, even if no packages have changed.

  • Packrat no longer creates empty ignore files. (@aronatkins)

Packrat 0.4.4

  • Packrat now understands the pkgType = "both" option and can properly restore projects when that option is set.

  • The ignored.packages option has been added, allowing users to specify packages that should not be tracked by packrat. Such packaes will not enter the lockfile on packrat::snapshot() calls; nor will they be cleaned out on packrat::restore() calls.

  • Simple functions for interacting with the set of available repositories have been added. See ?repository-management for more details.

  • Facilities for interacting with local CRAN-like repositories have been added. This feature will eventually supersede the functionality offered by packrat's 'ad-hoc' local repositories. The functions packrat::repos_create() and packrat::upload_package() are the two main API functions currently exported for uploading (source) packages to a local CRAN-like repository. By using a local CRAN-like repository, pre-existing workflows using library() and install.packages() to install packages should 'just work', as long as the repos option is properly set.

  • The cache directory layout has been modified to ensure help (?) calls succeed. This is a breaking change with older versions of Packrat, and so newer versions of Packrat will use a new cache folder. (#194)

  • Packrat issues a warning on packrat::init() if it was unable to infer the source of a particular package on initialization and instead uses the latest CRAN version.

  • Packrat now infers whether a project implicitly depends on Shiny.

  • Packrat now properly infers whether a package is from a CRAN-like repository, by checking if the source field maps to one of the repository names in getOption('repos'). (#185)

Packrat 0.4.3

  • Packrat avoids superfluous calls to available.packages() when possible, to avoid unnecessary internet requests.

  • A bug where packrat::status() could cause the R session to hang was fixed. (#179)

  • Added the R option packrat.default.project.options, which are now respected on e.g. packrat::init(). You can set options(packrat.default.project.options = list(...)) to set your own default project options to use across Packrat projects.

Packrat 0.4.2

  • Packrat properly infers whether a project is an R package. A project with a DESCRIPTION file that has no Type: field, or has the Type: Package field, will be considered as an R package.

  • Custom library paths can be set through the R_PACKRAT_LIB_DIR environment variable, which can be useful when using Packrat for non-local dependency management or deployment.

  • A bug in the propagation of BioC repositories was fixed.

  • Symlinks to R packages are created and destroyed more conservatively; this should help prevent problems where multiple R processes are acting within a single Packrat project.

  • The autoloader was not properly created in rare cases (thanks, @krlmlr!)

  • install_local() now forces lib and repos to be passed as named arguments, to avoid insidious errors. (#162)

  • Packrat no longer removes the lockfile on disable; rather, it simply removes the autoloader. (#161)

  • Packrat projects can now be non-interactively bootstrapped using the command: R --vanilla -f packrat/init.R --args --bootstrap-packrat. (#158)

  • packrat::bundle() gains an 'omit.cran.src' argument, for ignoring package sources that are retrievable from CRAN. (#156)

  • Packrat now understands how to install packages from custom CRAN-like repositories. (#153)

  • Packrat now infers itself to be a git-managed project if any of its parent directories contains a .git folder. This ensures packrat projects included as sub-directories of a git-managed folder are properly understood as git-managed projects.

  • packrat::bundle() gains an 'include.vcs.history' argument, for specifying whether VCS history folders (e.g. .git/, .svn/) should be packaged as part of the bundle. (#159)

  • packrat::bundle() gains an 'include.bundles' argument, for specifying whether previously generated bundles are included as part of new bundles.

  • Packrat now properly bundles projects on Windows (files in the packrat folder were not properly bundled when using the internal version of R's tar) (#152)

  • Packrat gains the 'load.external.packages.on.startup' option, to control whether external packages are loaded on startup.

  • Allow users to specify packrat itself as an external package (experimental, #147)

  • The automatic snapshot mechanism now re-uses the available.packages() cache, to avoid unnecessary internet requests to CRAN.

  • packrat::status() no longer asks you to uninstall packages that are found within the Suggests: field of the project's R package dependency chain, but does not require you to track these packages in the lockfile either. (#109)

  • Downloads are now retried (up to a default of 5 times) on failure -- this should help with intermittent timeouts when e.g. downloading archives from GitHub. (#134)

  • Packrat now adds .Rprofile to the .Rbuildignore file used for package development. (#138)

Packrat 0.4.1

  • Coincides with first CRAN release of packrat.

  • Disabled migration warnings on Windows (as we now use junction points if possible to facilitate isolation)

  • Fix bug with tar and projects containing a large number of files (#129)

  • External packages (alongside their dependencies) are now symlinked to the packrat/lib-ext folder -- this allows external packages to function better alongside regular packrat workflows.

  • If package symlinking (junctioning) fails, we instead copy the package to the corresponding packrat lib* folder.

Packrat 0.4.0

  • The source.packages argument for init and snapshot has now been removed. In lieu, projects can now specify local repositories -- packages will be looked up in these directories when attempting to use a source package.

  • Added install_local for installing packages from local repositories. For example, install_local("abc") will attempt to find and install a package named abc from the local repositories as specified within packrat.opts.

  • Added option use.cache, for creating a persistent global cache of installed packages which can be easily symlinked and reused across multiple projects. When active, this will force installation of one version of a package only once -- thereafter, any projects requiring that package can symlink to the installed version in the cache.

  • Use junction points on Windows to enable caching of packages.

  • Added opts as an object for conveniently getting / setting specific options. For example, calling opts$auto.snapshot() returns the current value for the auto.snapshot option, while opts$auto.snapshot(TRUE) sets auto.snapshot on.

  • Fixed spurious download.file warnings on Windows.

  • Errors when installing packages on Windows are now captured and reported to the user.

Packrat 0.3.0

  • The packrat/bootstrap.R script has been renamed to packrat/init.R.

  • Fixed a bug in packrat::set_opts() where NULL entries were not handled correctly.

  • packrat::user_lib() and packrat::packrat_lib() can be used to query the locations for the user library and packrat library, respectively.

  • packrat::install_github() works as a dependency-free devtools::install_github(), and ensures Packrat projects using GitHub packages do not have an implicit dependency on devtools.

  • packrat::disable() can be used to disable Packrat on a project -- it removes the autoloader from the .Rprofile and turns off Packrat mode. (If you want to fully clean out the packrat library / sources, you will have to delete the packrat/ subdirectory explicitly)

  • packrat::packrat_mode() has been refactored into two separate functions: packrat::on() and packrat::off().

  • .Rmd files are now parsed for YAML dependencies (for rmarkdown dependencies).

  • Recommended packages (e.g. lattice) are now only taken as Packrat dependencies if explicitly installed by the user -- for example, if you want to include lattice (which is distributed with R, but updated on CRAN) in Packrat, you must explicitly call install.packages("lattice") to get that package into the private library.

  • packrat::snapshot() now takes available.packages() as a default argument to available, assisting in dependency inference for packages inferred yet not installed.

  • packrat::snapshot() now takes the state of the library, along with dependencies inferred in the code, as 'truth' -- any packages installed or inferred will now be entered into the lockfile on a packrat::snapshot() call (with caveats for 'recommended' packages, e.g. lattice)

  • packrat::snapshot() no longer installs missing packages.

  • Failed packrat::inits now clean up any files / directories generated.

  • packrat::bootstrap has been renamed to packrat::init -- packrat::bootstrap remains as a (deprecated) alias.

  • Repositories are now stored in the lockfile as key-value pairs, and these repositories are used on startup. This mechanism will allow for more granular control over which repositories are valid, as well as for custom (e.g. non-CRAN) repositories.

  • Isolation is now achieved using a different mechanism on OS X. Because user libraries can be installed into the 'system' library on OS X, we work around this by symlinking all 'base' and 'recommended' libraries into a private packrat 'system' library, and using that instead.

  • Packrat gains project-specific options. packrat/packrat.opts is a DCF file of project-specific settings that can be queried and set through packrat::get_opts and packrat::set_opts. Please see ?"packrat-options" for more information.

  • Packrat can now handle source package tarballs, in addition to source folders.

  • API functions have been appropriately snake_cased and function arguments have been dot.cased. We apologize for any broken workflows here; but we imagine that most of use of packrat is done through calls to exported functions with no arguments passed, so this shouldn't disrupt most workflows.

  • Packrat has introduced support for R packages -- you should now be able to develop R packages using packrat to manage your dependencies. Work here is on-going and feedback is appreciated.

  • search_path() allows you to discover which packages are currently attached, and from which library each package has been loaded from.

  • status() now (invisibly) returns a data.frame outlining the current status of your project, in addition to printing information to the console.

  • packrat::on now attempts to clean the search path when entering packrat mode. Any packages loaded from the user library will be unloaded before entering packrat mode.

  • The init.R script has been updated to work better with bundle / unbundle: after bundleing a packrat project, one should be able to initialize a new project using a combination of unbundle and source('packrat/init.R').

  • Migration scripts for Windows users have been added to packrat, to migrate user libraries away from the system library, to provide the library separation that Packrat requires.

Packrat 0.2.0

There has been a change in the directory structure for packrat project files. If you'd like to migrate a 'packrat 0.1.0' project to the new format, please do the following:

  1. Navigate to your packrat project directory,

  2. Remove the project .Rprofile file,

  3. Start an R session in this folder,

  4. Run the following R code:

    if (!require("devtools")) install.packages("devtools") devtools::install_github("rstudio/packrat") packrat::migrate()

After this, you can restart your R session, and you should be good to go!

  • packrat_mode allows you to seamlessly step in and out of packrat mode, for when you would would like to manage or use external projects while working with a packrat project. In addition, once you have entered packrat mode, the project directory is remembered for all later calls to packrat functions, e.g. snapshot, clean, restore, and so on -- so calling these functions without arguments will use the project directory, even if you have navigated outside of the project or to a project sub-directory.

  • The packrat project files have been migrated into a single packrat/ folder -- this keeps pollution in the base directory of your project down. The remappings are:

    • packrat.sources -> packrat/src,
    • library -> packrat/lib,
    • packrat.lock -> packrat/packrat.lock, and
    • the .Renviron has been removed.

  • bundle and unbundle allow you to zip up your project as a tarball for easy sharing,

  • with_extlib and extlib allow you to (temporarily) load and use packages within the user library; this can be useful if you want to leverage another package (for example, devtools::install_github) while avoiding dependencies on devtools itself,

  • A number of bugs relating to status() output have been fixed and tweaked, to give better information to the user.

  • Projects in packrat mode will have automatic, asynchronous snapshots for safe actions: with automatic snapshots on, packrat will automatically upgrade packages that are out of date, or add packages that are new to the lock file (e.g. when seen by an install.packages() call). Downgrades, removals and 'crossgrades' will be ignored and will require you to take an appropriate action (guided by the information provided by status()).

  • Packrat will warn you if you have user libraries in the system library path. In addition, packrat supplies a script for OS X users, to assist in migrating user packages from the system library to a separate user library. (By default, R versions compiled for Mac OS X install all packages into the system library; for packrat to function correctly we require that user and system libraries be separate -- this script should help facilitate the process. If you have any problems with migration, please let us know!)