Unsure when this became hard-required (may be from newer'ish kernel),
but happened to build against `make defconfig` kernel with DRM_QXL
enabled which didn't enable FBDEV_EMULATION then hit a implicit
declaration build failure and, looking at the kernel headers, that
function is only defined with FBDEV_EMULATION.
This been already suggested for a while so I assume most users have
this enabled either way at this point. The help text is not useful
anymore if it's to solve a build failure. In the event KMS_HELPER
is disabled, fbdev=1 also defaults to disabled, not that I test that
configuration much (realistically we need to require KMS_HELPER in
the future).
Signed-off-by: Ionen Wolkens <ionen@gentoo.org>
We moved where we set the default value for EGIT3_STORE_DIR, so at the
point `addwrite` is called, it isn't effective. Move it too.
Closes: https://bugs.gentoo.org/981310
Fixes: eae4d76cd2
Signed-off-by: Sam James <sam@gentoo.org>
Refactor the eclass logic to defer the 'git ls-remote' call used to
determine the object format until overrides are in place. Move it into
repository fetching loop so that mirrors and fallback URIs are respected
as well. While at it, fix duplicate directory entries in
EVCS_STORE_DIRS, and avoid creating an empty directory if 'git
ls-remote' fails.
For historical reasons, git-r3_set_gitdir function was responsible both
for setting the value of GIT_DIR and creating it if necessary.
To ensure that GIT_DIR was stable, the canonical URI was always used.
No network traffic was involved at this stage.
While adding sha256 support in 0477949bcf,
an additional 'git ls-remote' call was added to determine the remote
object format before creating GIT_DIR. We've missed that this call
would be performed early on, with no respect for repository URI
overrides, mirrors and fallback URIs. Most of the time this wasn't
a problem, at least as long as the canonical URI was reachable and had
the same object format.
It seems that the cleanest solution to this is to split setting GIT_DIR
from creating it. The former still needs to be done early, using
the canonical URI. The latter is best done in the fetching loop where
we are already iterating over the final list of URIs. This also permits
us to easily support fallback when the initial URI fails, much like we
do in the 'git fetch' invocation.
Closes: https://bugs.gentoo.org/979189
Signed-off-by: Michał Górny <mgorny@gentoo.org>
Part-of: https://codeberg.org/gentoo/gentoo/pulls/1708
Merges: https://codeberg.org/gentoo/gentoo/pulls/1708