The ?ignore parameter works correctly now.
Dependency resolution now does not fail if a package is not found.
pak can now install url:: remotes from
GitHub.
pak now does not fail when the package of a .tar.gz
GitHub snapshot is in a subdirectory, or in a subdirectory of a
subdirectory.
pak now errors early if it cannot deduce the name of the package
from a Remotes or Config/Needs/*
entry.
Solver failures now include details in some cases where previously they did not.
pak can now update packages in Docker containers where the old version was installed in the different Docker later (#251)
pak errors are now user friendlier and better formatted.
pak now does not load tibble and its dependencies in the pak subprocess, so their dlls are not locked by the pak subprocess on Windows.
pak now does not fail when installing a package that uses a non-UTF-8 encoding on R 4.3.x and later (#404).
pak functions that used to return tibbles return data frames now.
While data frames and tibbles are very similar, they are not completely
compatible. To convert the outputs of pak functions to tibbles call the
tibble::as_tibble() function on them. If the pillar package
is loaded, it improves the printing of the returned data frames.
Relatedly, pak::pak_install_extra() installs pillar now,
instead of tibble.
pak now supports file:// repositories.
pak now uses HTTP 1.1 to download packages on Linux, in addition to macOS. This fixes HTTP issues with some servers (#358).
New ?ignore-before-r parameter to ignore optional
dependencies that need a newer R version
(https://github.com/r-lib/pkgdepends/issues/243).
New ?ignore parameter to ignore an optional
dependency.
Allow specifying downstream package parameters with the
package=?param syntax.
lockfile_install() now works better for
any:: refs, and pak always install the version it has
planned for.
System requirement installation is now more robust and works for Unix shell expressions (#347).
CRAN-like resolution is more robust now if a repository is missing the usual metadata.
The lock file is pretty JSON now.
pak now handles all version requirement types properly: ‘<’,
‘<=’, ==, >=,
>.
The dependency solver now uses better heuristics and does not (effectively) freeze if multiple repositories have multiple versions of the same packages (e.g. RSPM and CRAN) (https://github.com/r-lib/pkgdepends/pull/277)
No user visible changes.
Lots of news, too much to list. This is a completely new package now.
First version on CRAN.