TODO.md
In wnattt/renv: Project Environments

Downloads

[x] download.file(method = "wininet") appears to be unreliable for file URIs?

Dependencies

[x] User should be able to control whether errors encountered during depenency enumeration are fatal, reported, or ignored.
[x] renv::init() should report dependency errors, but otherwise initialize the project.
[x] renv APIs that might call dependencies() should avoid re-reporting dependency errors multiple times. Potential fix: renv_dependencies_scope() could call dependencies once at the top level, and then renv::dependencies() could re-use the result of that object when requested.
[x] Audit other internal usages of renv::snapshot() for the appropriate value of the errors argument.

Miscellaneous

[x] Packrat-style init() function, where we discover the R packages used within the project, and then initialize a new environment / library with those packages. Speed up the process by re-using installed packages in the user library if possible.
[x] Warn (error?) when attempting to snapshot a library with missing dependencies -- e.g. the library contains 'markdown' but not its dependency 'mime'.
[x] Properly handle broken links in the project library (e.g. if the cache moved or was mutated for some reason)
[x] rehash() -- tools for validating package hashes + their install location? rehash() would look at packages in the cache and update their cache location if the caching scheme changed.
[x] clean() function to remove ununsed packages from the library. (If the cache is enabled, they will remain in the cache).
[x] Allow users to override the repository used during restore of a particular package? (Setting options(repos) would suffice here I believe)
[ ] Use custom Makevars file and set some variables that certain packages need but don't properly declare? (E.g. older versions of the maps package need awk installed)
[x] restore() will attempt to repair the dependency tree during restore; e.g. dependent packages in the lockfile will be downloaded and installed as required. Should we prompt the user to snapshot() afterwards so that the newly-reinstalled dependencies can be captured in the lockfile?
[x] Take over the R_PROFILE, R_PROFILE_USER, R_ENVIRON and R_ENVIRON_USER environment variables? Needed for cases like https://github.com/rstudio/packrat/issues/526.
[x] Make it possible to snapshot arbitrary libraries. [DONE: call snapshot() on any project to generate a manifest for that project's state]
[x] hydrate() as a general function to discover dependencies and then install any missing packages into the active library. (Companion function to init())
[x] Should we provide for project-specific options? What about download method management? [Motivation: users might require curl + http_proxy settings to download packages from CRAN; authentication for GitHub]
[x] Allow users to ignore certain packages in a project (e.g. those that are not on CRAN)
[x] Consider .renvignore or similar for controlling what files / directories renv will crawl when discovering dependencies.
[x] Properly handle RENV_DEFAULTS case (they get lost if the user attempts to build and reload a package managed by renv, which screws up RStudio's notion of the library paths if the project is later deactivated)
[ ] Audit usages of renv_file_exists(). When do we care about broken symlinks? file.exists() returns FALSE for broken symlinks; renv_file_exists() returns TRUE.
[x] De-couple the 'retrieve' + 'install' steps during restore.
[x] Include recommended packages in the lockfile? (Since R installations on Linux may not have these packages available). Or at least confirm that packages which depend on recommended packages enter the lockfile.
[x] Use a single lockfile file, but provide a history() function to dig out old versions of the lockfile. (Make it clearer to users what needs to be committed)
[x] Handle NA mtime in filecached APIs.
[ ] Implement bundle() for packaging up a project, potentially with package sources, binaries, or even the library itself for restoration on a new machine (potentially lacking internet access).
[x] Provide tools for upgrading version of renv used to manage a project. [renv::upgrade()]
[ ] Think about how we might discover 'stale' entries in the cache.
[x] Handle renv::init() case when another project already active (want to make sure we use the user library rather than active library)
[x] renv::status() function reporting differences between lockfile + library.
[ ] Tools for analyzing the cache?
[x] Python: allow e.g. 'python = TRUE' for initializating a Python project.
[x] Python: properly handle conda environments vs. pip.
[x] Python: teach reticulate how to load virtual environments within renv dir.
[x] Warn when lockfile contains packages from unknown source?
[x] Automatically bootstrap an renv installation when loading a project if none is available locally?
[ ] Should renv::install() auto-update packages?
[x] Only report R package entries that have changed if they exist both before and after (ie: only report upgrades / downgrades / crossgrades)
[x] Include current session's R version when computing hash for DESCRIPTION
[ ] Install of keras failed to install reticulate?
[ ] API for activating project and upgrading renv, rather than using previously-recorded version of renv?
[ ] Treat user library as a source when restoring projects?
[ ] renv should build packages before install, especially for packages from GitHub.
[ ] Use pkgcache for package installation?
[ ] Use pkgdepends to validate dependency list before snapshot.
[ ] API to allow pak to use the cache?
[ ] Record usage time for packages (as alternate scheme for finding unused packages)
[x] renv::revert() should tell the user what it did.
[x] renv::revert() should have project argument last.
[x] Warn when installing a package currently loaded? (so that user knows they must restart R to get newly-installed package)