This release brings some major changes to the way that Chromium
is packaged in Gentoo. After masking GCC support since it broke beyond
the capabilities of the maintainers to fix in 120, changes
in 127 have made dependencies on LLVM stronger: in addition to
requiring LLVM for Rust we now directly depend on LLVM libraries for
bindgen.
As a result we're experimenting with completely dropping GCC support
from the ebuild. This isn't a decision that was made lightly, however
the advantages to going down this path make it very appealing:
- Our builds are closer to upstream (and nominally supportable)
- We don't need to maintain so many out-of-tree patches
+ We may actually be able to drop them entirely for most arches
- Ebuild logic is significantly simplified, making it easier to review
- Decreased maintainer hair-pulling.
The removed GCC support could be revisited if there is sufficient
interest, and as an added benefit the new, streamlined, and better
documented ebuild should provide a better basis for maintaining it.
One of the outcomes of this decision to double-down on LLVM support has
been the opportunity to properly select an implementation via upgrading
to the `llvm-r1` eclass, which also enables us to start trying to make
the Rust->LLVM dependency a bit more clear for our ebuild.
Notable changes:
- `llvm-r1` eclass
- Dropped GCC support
- Always bundled libcxx for stdlib
- GN/bindgen wrapper updated to work with system bindgen
+ needs upstreaming
Signed-off-by: Matt Jolly <kangie@gentoo.org>
When AppArmor is enabled and sys-firmware/edk2-ovmf-bin is
installed then starting a guest under libvirt fails, because
libvirt assumed different paths for UEFI. A fix was merged
upstream so backport it.
Resolves: https://bugs.gentoo.org/911786
Signed-off-by: Michal Privoznik <michal.privoznik@gmail.com>
Signed-off-by: Sam James <sam@gentoo.org>
We filter-lto due to ODR violations. But only in src_configure, which is
common enough as that's where we set up the environment for configuring,
in general. However cmake is a bit special as we also need to bootstrap
it, and we do this in src_prepare(). We need to do that filtering there
as well.
While we are at it, the solaris linking issue is most probably relevant
the same way, although I have not actually tested this...
Bug: https://bugs.gentoo.org/858335
Signed-off-by: Eli Schwartz <eschwartz@gentoo.org>
Acked-by: Sam James <sam@gentoo.org>
This reverts commit fbe4bc7b9d.
The previous change does not install the NeoVim symlinks correctly as they
point to /usr/lib/tree-sitter/liblang.so instead of /usr/lib/liblang.so.
Also the query files are not installed in per-package directories and have
common names (e.g. highlight.scm) which is causing file collisions.
To be fixed in a follow up commit which will introduce
tree-sitter-grammar-r1.eclass, so that ebuilds can be rev-bumped so that
users can rebuild and get the changes.
Bug: https://bugs.gentoo.org/935631
Bug: https://bugs.gentoo.org/935641
Signed-off-by: Matthew Smith <matthew@gentoo.org>
Jakov Smolić
Matt Jolly
Miroslav Šulc
Sam James
Quincy Fleming
Yixun Lan
Michal Privoznik
Eli Schwartz
Matthew Smith