Cyclic dependencies (A depends on B, B depends on A) were not selected
because neither package could be removed individually, so
can_remove_package would always return false for both. By preselecting
all dependencies then filtering back out any dependencies still required
by any packages that will not be uninstalled, groups of unneeded cyclic
dependencies can be found.
Fixes FS#41031
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
When printing a list of URLs of packages to be updated, pacman was ignoring any
replacements that would be made in the update process.
Fixes FS#35812
Signed-off-by: Allan McRae <allan@archlinux.org>
Relying on localdb to determine which trigger operations should match is
completely broken for PostTransaction hooks because the localdb has
already been updated. Store a copy of the old version of any packages
being updated to use instead.
Fixes FS#47996
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
Check if we overwrote an exiting pacnew file before unlinking it.
Otherwise, updating to a version with an unchanged file would delete
existing pacnew files.
FS#47993
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
The test introduced herein illustrates a behavior that may be unexpected
to package writers.
It creates a package "pkg3" that is configured to depend on a
"dependency" which version is between 3 and 4, inclusive. Two other
packages are already present, providing "dependency" in version 2 and 5,
respectively. So, the situation looks roughly like this:
pkg1 pkg3 pkg2
provides depends on provides
| <------------> |
version __________2____________3____________4____________5___________...
This seems to be enough to satisfy pacman when installing "pkg3". From
an iterative standpoint, this is completely logical: First, the
requirement "dependency>=3" is checked. There is a package that
satisfies this restriction, it is called "pkg2". Afterwards,
"dependency<=4" is covered in the same way by "pkg1".
Nonetheless, what a package writer intends when specifying
depends=('dependency>=3' 'dependency<=4')
is most probably that pacman should only allow this package to be
installed when there indeed is a package present that provides a version
of "dependency" that lies _between_ 3 and 5.
Signed-off-by: Dominik Fischer <d dot f dot fischer at web dot de>
Signed-off-by: Allan McRae <allan@archlinux.org>
When replacing a file with a directory, any files inside the new
directory cannot possibly exist on the filesystem and can be skipped.
This allows cross-package symlink-to-directory transitions when there
are files with the same name under both the symlinked directory and the
new directory.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
When a symlink to a directory is changing to a directory, any package file
inside the new directory can create an unexpected conflict with the filesystem.
Reported by Neofytos and Luca from Chakra.
Signed-off-by: Allan McRae <allan@archlinux.org>
Newlines clutter tap output and can potentially confuse TAP parsers.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
Especially when maintaining local templates in addition to the ones
stored in /usr/share/makepkg-template, it can be useful to include
templates stored in multiple different locations into one PKGBUILD. This
patch makes this possible by allowing --template-dir to be specified
multiple times.
This also introduces a dedicated error message when a template cannot be
found, in contrast to the already existing "Couldn't detect version for
template '%s'".
If a template of the same name is present in more than one of the given
directories, the last one always takes precedence.
Neither the default behaviour without the option given, nor the handling
of a single template dir is changed.
Signed-off-by: Dominik Fischer <d.f.fischer@web.de>
Signed-off-by: Florian Pritz <bluewind@xinu.at>
Signed-off-by: Allan McRae <allan@archlinux.org>
Paths from noupgrade, the transaction skip_remove, and package backup
lists were combined into a single list matched using fnmatch causing
paths with glob characters to match unrelated files.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
LIBCURL was never set in the Makefile so XferCommand was always being
set in the test file. This removes the only substitution in our test
files which will prevent the TESTS file from being rebuilt every time
configure is run.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
Leave user files in place and save new config files with a .pacnew
extension. This reduces the complexity of file extraction and respects
the principle that pacman shouldn't modify files it didn't create.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Allows tap.sh to show the line number where the helper function was
called on failures.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
tap.sh is a reusable TAP library that handles test counting and provides
useful diagnostic messages on test failures.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
This moves most of the parsing work out of the sorting path. The explode
and splitfile functions now call input_new and append input_t structs
to the list of sort candidates instead of raw strings. This lets us
make smarter and easier decisions in the sorting callbacks, which are
now also split into the version and file comparison methods for clarity.
This fixes two bugs:
1) Incorrect ordering with filenames containing epoch in the pkgver
2) Incorrect ordering with package names which are substrings of
each other (e.g. "systemd" and "systemd-sysvcompat").
Performance of the --files mode degrades slightly as a result of this
change, but not unreasonably. Sorting with small inputs (5-10) doubles
in runtime, but larger inputs (4000+) only increase by 20%.
ref: https://bugs.archlinux.org/task/37631
Signed-off-by: Allan McRae <allan@archlinux.org>
--config does not respect root, causing pacman-db-upgrade to read the
local pacman.conf rather than the one in the test root.
Also add a rule to ensure the ALPM_DB_VERSION file is actually being
created.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
Adds a cmd property to tests (defaults to pacman) which is resolved
using directories specified with --bindir (defaults to PATH). The
ability to manually specify a particular binary is preserved in order to
allow running individual tests with differently named binaries such as
lt-pacman.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
--confirm cancels the effect of a previous --noconfirm.
This makes it easier for scripts to default to --noconfirm
but allow users to override it.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
This prevents an exception in the event backup entries are not in the
correct format and brings the test in line with alpm's backup parsing
which splits on the last tab rather than the first.
Signed-off-by: Andrew Gregory <andrew.gregory.8@gmail.com>
Signed-off-by: Allan McRae <allan@archlinux.org>
