Refreshing eclasses and profiles can potentially introduce surprises but
the only notable change I see are fixing the sse2 flag for OpenSSL and
changing SLOTs for cross compilers.
Until recently Gentoo used stand-alone USE flags like `sse2` to toggle
optional CPU features. This has changed to USE_EXPAND style flags like
`cpu_flags_x86_sse2`. Our OpenSSL ebuild pulled in this change in commit
a9285ef530f88a1373c1459c4e20ca1065b24a91 but at the time the profiles
weren't updated to match, resulting in turning off sse2 for our OpenSSL
build.
Cross compiler ebuilds now always act as if the `multislot` USE flag was
on, allowing minor versions to be installed in parallel. e.g.:
- cross-*/gcc-4.8.3: SLOT changes from `4.8` to `4.8.3`
- cross-*/binutils-2.24-r3: SLOT changes from `0` to `2.24`
The new slot for binutils change may require adding a call to
`binutils-config`, similar to `gcc-config`, the next time we upgrade
binutils but for now nothing needs to change.
Packages updated:
eclass
licenses
profiles
scripts
Update eselect and all packages depending on eselect modules using the
old app-admin/eselect-* names for eselect modules. Upstream has moved
the eselect modules to a new category: app-eselect/*
The only impact this has on production images is upgrading
app-crypt/pinentry from 0.8.2 to 0.9.0.
Packages updated:
app-admin/eselect
app-crypt/pinentry
app-editors/emacs
app-shells/bash-completion
dev-lang/python
dev-lang/ruby
dev-util/ctags
Pull in newer EAPI=5 ebuilds, the upstream python eclass has deprecated
EAPI=4 and reports an error if util-linux-9999 isn't using EAPI=5
The version we ship in CoreOS doesn't change.
Update from vim 7.3.762 to 7.4.273. The 7.4.x ebuilds no longer use the
vim.eclass which has been removed upstream. Updating vim in preparation
for syncing the eclasses with upstream.
Packages updated:
app-editors/vim
app-editors/vim-core
scons has been replaced by a standard autotools build system. Beyond
that a few things are different:
- remove GCC 4.8 workaround, the build system handles this now.
- instead of trying to make executing code in a cross-compiled $SYSROOT
work just avoid running the test suite for board builds. The tests
can still run for the native SDK build.
- `make install` now handles at least part of the install process.
The Xen framebuffer driver, on HVM EC2 instances, fails to connect to
device/vfb/0 on the xenbus. This causes a 30-second boot delay on
certain architectures since the support is compiled in.
Demoting to a module increases boot speed by 30 seconds.
Update linux-headers from 3.13 to 3.18 for some definitons needed by arm64
that were not in older header versions.
Signed-off-by: Geoff Levand <geoff@infradead.org>
