| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
|
| |
This slightly breaks the API in a few places, namely in
pkgcore.config.domain if the installed_repositories and
source_repositories attrs are used externally.
|
|
|
|
|
|
| |
This is the beginning of refactoring domain into separate pieces bit by
bit. It would have been cleaner if the visibility module was written but
that will come later.
|
|
|
|
|
| |
This simplifies things as they appeared to be used interchangeably in
certain places.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Everything is now anchored under /var/gentoo, note that
/var/cache/gentoo wasn't chosen since various data might not be able to
be regenerated, e.g. certain binary pkgs in PKGDIR or locally modified
ebuilds that aren't committed anywhere else.
The defaults are now as follows:
DISTDIR=/var/gentoo/dist
PKGDIR=/var/gentoo/pkgs
PORTDIR=/var/gentoo/repos/gentoo
Note that PORTDIR isn't actually supported anymore but all repos will be
available under /var/gentoo/repos/<repo_name> where repo_name should
generally be the matching name from the repo metadata.
|
| |
|
|
|
|
|
|
| |
All regular, nonhidden files under make.conf will be parsed in
alphabetical order meaning that later settings will override earlier
ones.
|
|
|
|
|
| |
Mostly the same thing is being used in a several places throughout
the code.
|
|
|
|
|
| |
Only the file provided by pkgcore (usually installed at
/usr/share/pkgcore/config/make.globals) is used for global settings.
|
| |
|
|
|
|
|
|
|
| |
The simpler install method doesn't appear to work quite right with the
travis-ci setup yet.
This reverts commit e080b28a84bf237ec72a8da4a5685123169ff939.
|
|
|
|
| |
The warnings and errors for the missing uids/gids was dropped recently.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
This should fix installing pkgcore via pip which uses setuptools
internally and passes options to the install method that distutils
doesn't support so our custom, distutils-based install method breaks.
However, this still doesn't allow just doing `pip install pkgcore` or
similar to fully work since pkgcore's setup.py depends on snakeoil
modules so pip is unable to figure out the dependency and install it on
new setups. In other words, you have to do `pip install snakeoil` first
manually before installing pkgcore in new environments.
|
|
|
|
|
| |
It's not worth demandloading modules that will almost always be needed.
Some of these are always needed now that we require repos.conf to exist.
|
| |
|
|
|
|
|
|
| |
We're not quite at a fully unpinned repos setup without any underlying
default so either the default must be set or the gentoo repo must be
configured in repos.conf.
|
| |
|
| |
|
|
|
|
|
| |
We don't need a multiplex setup for the base case where only the default
repo exists.
|
|
|
|
|
| |
This is what the previous code did and is the easiest fall back until we
fix masters handling.
|
|
|
|
|
|
|
|
|
| |
This fixes things like slaved repos properly inheriting mirrors from the
master repo.
Note that more work needs to be done since this doesn't fully handle
multiple masters yet or response the masters field from repos.conf
settings.
|
| |
|
|
|
|
|
|
| |
Strip leading slash from the world file path passed to os.path.join(),
as that causes it to ignore preceding parameters and therefore fail to
prepend root.
|
|
|
|
|
| |
tox>=2 is broken for various documented dir substitutions (see
https://bitbucket.org/hpk42/tox/issue/246/envbindir-substitution-key-not-found).
|
|
|
|
| |
It can be found/revived from the history if needed.
|
|
|
|
| |
Syncers should always exist under the new repos.conf only system.
|
| |
|
|
|
|
| |
This mirrors portage's current usage and docs.
|
|
|
|
|
| |
Use 'bzr branch' rather than 'bzr get' since the latter is deprecated
and no longer works correctly.
|
| |
|
|
|
|
|
| |
It's cleaner to just use __inject_phase_funcs directly in only the
EAPI=0 lib.
|
| |
|
|
|
|
|
|
|
|
|
|
|
| |
These types of messages spew forth annoyingly on non-Gentoo systems
where pkgcore or pkgcheck work fine running as root or another user
(e.g. on travis-ci) or in a prefix setting once support is finished for
it.
Really it seems like it would better to document such requirements for
certain use cases and error out if absolutely necessary in the code
somewhere else.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
| |
The only one left in the code is required to be exported to the ebuild
environment according to PMS and should really be deprecated and dropped
from a future EAPI.
Besides that, references are scattered throughout the docs and may be
cleaned up in the future.
|
|
|
|
| |
Only /etc/portage/make.conf is used now.
|
| |
|
| |
|
| |
|
|
|
|
| |
Mostly for clarification and simplification purposes.
|
|
|
|
|
| |
Various simplifications and explanations hopefully to make the code a
bit clearer.
|
| |
|
|
|
|
|
|
|
|
|
| |
Now pkgcore only supports configuring repos via repos.conf, PORTDIR and
PORTDIR_OVERLAY settings in make.conf are no longer respected.
Also, all references to portdir and overlays are slowly getting purged
from pkgcore. All that is seen are repos that are sorted via their
priority settings in repos.conf.
|
|
|
|
|
| |
The official gentoo repo has properly unset masters in its layout.conf
file for a while now.
|
| |
|
|\
| |
| | |
fixed function usev() in bash/eapi/common.lib
|
|/ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
'clean' should reuse livefs repos. With replace=True, livefs repos are
excluded by the _vdb_restrict in ebuild/resolver.py, causing the
resolver to generate a packages list that is not consistent with the
installed packages.
The error can be shown by running "pmerge --clean" then "pmerge -uD
@world". Some packages (automake-1.15 in my case) removed by 'clean'
are reinstalled by 'update'.
Setting replace=False fixed the error. We don't want to replace
anything by 'pmerge --clean' afterall.
|