Do not attempt to get xattrs from symlinks. Their extraction is not
supported either.

Do not use alloca.

At least in kernel packages for aarch64 we were running into
the 8000 file hard limit in linux-devel. Increase it until we
have a better solution in place.

nina authored 2 years ago and Timo Teräs committed 1 year ago

Fixes alpine/apk-tools#10826
Fixes alpine/apk-tools#10884

[TT: Remove ADBI_PKG_PASSWD for now.]

Timo Teräs authored 1 year ago and nina committed 1 year ago

There will be additional metadata in the package structure that
is user settable. Allow setting them via --info also.

Allow omitting loading of system installed database and system
repositories for the commands that operate on package indexes
only.

Handle file: URLs correctly, and don't do linkat() syscall if
it is a remote repository.

Also account the file size to progress always.

ref #10873

fixes #10871

This will also fix "search" to display one match from each principal
pkgname group. "search -e vim" will now show both vim and gvim.

fixes #10864
fixes #10777

To make the search output and experience more consistent.

elly authored 2 years ago and Timo Teräs committed 2 years ago

This change adds two manpages: apk-v2(5) and apk-v3(5). These pages
describe the v2 and v3 file formats respectively, as I currently
understand them.

Only the leading zeroes should be ignored. Handle properly if the
version component is actually zero.

fixes #10880

nina authored 2 years ago and Timo Teräs committed 2 years ago

In order to address the problem with early file ownership (before
passwd/group exists), do a few changes here:

1) For root-owned things, always fall back to fixed values, as
   those are the ones we always know and cannot change. Since
   the earliest packages only have root-owned files, this fixes
   the problem for those.
2) During file migration, if we encounter passwd/group and this
   is a newly committed file, reset the idcache, similarly to how
   it is done for scripts. This allows the next package to reload
   the mappings. Since those two files are usually installed as
   a part of the first package, the second package onwards should
   have a valid, complete mapping (reset only by scripts).

Sort the changes by package name before displaying each
package listing for easier reading by user.

fixes #10866

Omit separate static cache handling step if the explicit cache
is configured to the static cache directory.

fixes 609fd218 "cache: fix 'clean' to prune static cache always"

Document cache clean --purge better. Add a cache purge alias for
it. And also purge installed package copies on disk installs.

fixes #10857

Fix cache applet to prune the static cache which is used for index
files if explicit caching is not enabled.

fixes #10754

nina authored 2 years ago and Timo Teräs committed 2 years ago

This fixes undefined behavior at least in database, where BIT is
used with 31 (as APK_MAX_REPOS is 32) which is not representable
with a signed integer.

There are subtle issues where solving fails with --available,
and install_if rules if the repository indexes are not available.
Also it can be considered upgrade failure if index update failed.

Abort cache download, cache sync and upgrade operations early
in the above mentioned cases.

Also document side effects of --simulate that might affect upgrade.

fixes #10726, #10764

Track separately autoupdate failure, and opening errors. And print
only one error per index at maximum.

This makes update with or without --no-cache consistent.

Ignore errors from local paths as builders often refer to a local
path that is being built, but might be non-existent when building
the first package.

ref #10726

Force preference on the user specified action on commandline.
This allows upgrading packages which would otherwise be kept back
due to partial update preferring to keep top level dependency on
the installed version.

fixes #7531

The --no-chown flag is most useful when running apk as a regular
user, in which case we want to make sure that there are no issues
regarding permissions.

Fixes #10854

apk_db_close is not re-entrant, and would be executed by the signal
handler. It makes sense to just ignore the signal after applet
is complete as the clean up work has been started already.

fixes #10840

fixes #10848

fixes #10721

fixes commit 395e92b6 "io: formalize apk_ostream_write() always writing full data"

fixes #10650

It makes sense to not prioritize requirers anymore. It was
originally intended to select early packages with multiple
constraints seen. However, more important are the constraints
that actually limit the choise.

fixes #10843