Minor sync with upstream, adds a use flag we don't enable.
Updated our toolchain just in time, the -fuse-ld= option is now
supported and as of 218 systemd doesn't link with bfd any more so now is
a good time to re-enable the configure test to enable it. For reference
the guilty commit at fault is probably:
5f86c1f4c4
Or specifically:
```c
/* GCC maps this magically to the beginning and end of the BUS_ERROR_MAP section */
extern const sd_bus_error_map __start_BUS_ERROR_MAP[];
extern const sd_bus_error_map __stop_BUS_ERROR_MAP[];
```
... to which I say haha wtf... WELL WHAT ABOUT WHEN IT DOESN'T? :-P
The new dockerd wrapper script does its best to select between the btrfs
and overlay backends based on the filesystem mounted at /var/lib/docker.
The new 1.4 version will remain marked as ~amd64 for testing purposes
until we stabilize its dependencies, including Linux 3.18.x.
This is a temporary workaround to allow people to transistion into 1.3+
smoothly. This sets --insecure-registry=0.0.0.0/0 to maintain backward
compatibility.
Fixes commit 3f028060 which mistakenly added coreos-sources as both
build and run-time dependencies. This was missed initially because there
are exclude rules that remove /usr/src from production images but it
does needlessly slow down the build and pulls in extraneous kernel build
dependencies like perl.
Prune old releases, no need for them.
For anyone who wants to test OverlayFS. To use add the following to
`/build/amd64-usr/etc/portage/package.keywords/kernel`:
=sys-kernel/coreos-kernel-3.18.0 ~amd64
=sys-kernel/coreos-sources-3.18.0 ~amd64
At some point this variable stopped getting set properly in SDK ebuild
environments. It is currently used to filter out some build time files
from target images. Could be done cleaner but that can come later.
This cuts out the need for git for our kernel sources and the + and the
end of our versions is now gone. The build remains a bit weird in order
to bundle the initrd into the kernel image but eventually that will be
dropped, leaving the new coreos-kernel eclass pretty tiny.
Bonus: That + at the end of our versions is finally gone!
Providing the kernel source via git is overkill, we do not maintain any
patches on top of the mainline kernel any more and the repository is
huge. As an added benefit this installs the kernel source to the
traditional /usr/src/linux location that Gentoo expects. Hopefully that
will reduce the number of warnings many ebuilds tend to output.
Patches routinely modify the man page source XML, invalidating the
prebuilt man pages provided by release tarballs. Patch the live ebuild
too and require the docbook XML DTDs unconditionally like we did for 215
all the time going forward so this issue isn't forgotten again during
the next version bump.
Another day, another intermittent build failure because the local xml
docbook catalog didn't work right and the xml parser also failed to
fetch the schema over the network. Configure portage to use network
namespaces to prevent such situations from ever working in the first
place so at least the errors are unlikely to be intermittent.
For the sake of completeness use ipc namespaces too but that is less
likely to actually be a factor in our builds.
