| Commit message (Collapse) | Author | Age | Files | Lines |
|
|
|
|
| |
With thin manifest, there's no need to have a Manifest file if there
are no DIST entries.
|
|
|
|
|
| |
With thin manifest, there's no need to have a Manifest file if there
are no DIST entries.
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
|
| |
This allows repos to disable manifest signatures, which is useful if
they want to prevent merge conflicts like those that thin-manifests is
designed to prevent.
|
| |
|
| |
|
| |
|
|
|
|
|
|
|
|
|
| |
For any repo that wants thin (just src_uri digests), they just need to add
thin-manifests = true
to their layout.conf. Again, this should only be used in repositories
were the backing vcs provides checksums for the ebuild data.
|
|
|
|
|
|
| |
'thin' is just distfiles. This is primarily useful when the ebuild
lives in a vcs- git for example, which already has it's own checksums
to rely on.
|
|
|
|
|
| |
This enables controling the behaviour (creation and validation) per
repo, and while mildly ugly, refactors in the right direction.
|
| |
|
| |
|
|
|
|
|
|
|
|
| |
This re-implements the fix from commit
21330075f07248765016e104b3ba8216903f1ecb, without introducing the
unwanted behavior reported in bug 382557. This involves checking the
direct dependencies of virtual slot updates to make sure they are all
visible, before pulling them in.
|
| |
|
|
|
|
|
|
| |
This reverts the behavior change from commit
21330075f07248765016e104b3ba8216903f1ecb, since it's too aggressive in
pulling in new virtual slots that may have masked dependencies.
|
|
|
|
|
|
|
| |
This re-implements the change from commit
21330075f07248765016e104b3ba8216903f1ecb in order to avoid executing
unnessary virtual slot expansion code when the given atom specifies a
slot or --update is enabled.
|
| |
|
|
|
|
| |
This ensures that USE deps and repo deps are preserved here.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Previously, we used sys.getfilesystemencoding() for the 'merge'
encoding, but that had various problems:
1) If the locale is ever changed then it can cause orphan files due
to changed character set translation.
2) Ebuilds typically install files with utf_8 encoded file names,
and then portage would be forced to rename those files to match
sys.getfilesystemencoding(), possibly breaking things.
3) Automatic translation between encodings can lead to nonsensical
file names when the source encoding is unknown by portage.
4) It's inconvenient for ebuilds to convert the encodings of file
names themselves, and upstreams typically encode file names with
utf_8 encoding.
So, instead of relying on sys.getfilesystemencoding(), we avoid the above
problems by using a constant utf_8 'merge' encoding for all locales, as
discussed in bug #382199 and bug #381509.
|
|
|
|
|
|
|
|
| |
Previously, the virtual cost minimization code from bug #141118 would
prevent virtual dependencies from pulling in new slots. That behavior
was not desired for --update, so now it's fixed to pull in the latest
slot available. This allows virtual/jdk-1.7.0 to be pulled in by
dependencies when --update is enabled.
|
|
|
|
| |
This will fix bug #382233.
|
|
|
|
|
|
| |
It should be enough that the automatic use of metadata-transfer is
documented in the man page, and allows us to avoid showing a useless
warning in common usage scenarios.
|
| |
|
|
|
|
| |
Special methods can't be assigned in the constructor.
|
| |
|
|
|
|
|
|
| |
This will avoid the "OSError: [Errno 2] No such file or directory" that
is triggered inside abssymlink if the merge encoding is not ascii or
utf_8, as shown in bug #382021.
|
| |
|
| |
|
| |
|
| |
|
| |
|
|
|
|
|
| |
This tests --update in any case, and --update-use-local-desc only if
python xml support is detected.
|
| |
|
| |
|
|
|
|
|
|
|
|
|
| |
This has been disabled since commit
c7faa634369e61b87a40172ceb0a5cb9494fd518, but the only reason cited
was to avoid permissions issues. So, go ahead and enable it, and handle
PermissionDenied just in case. NOTE: The NewsManager typically handles
permission errors by returning silently, so PermissionDenied won't
necessarily be raised even if we do trigger a permission error.
|
| |
|
| |
|
|
|
|
|
|
|
|
| |
This will fix bug #381629. The case is tested in tests/emerge. During
forced charset conversion, 'ascii' codec is forced, since otherwise we
somehow end up with the wrongly encoded file names when running the
test with Python 3.2 (which causes encoding problems for quickpkg when
it calls tarfile.gettarinfo).
|
| |
|
| |
|
| |
|
| |
|