the fix to extendDerivation in #140051 unwittingly worsened eval performance by
quite a bit. set elements alone needed over 1GB extra after the change, which
seems disproportionate to how small it was. if we flip the logic used to
determine which outputs to install around and keep a "this one exactly" flag in
the specific outputs instead of a "all of them" in the root we can avoid most
of that cost.
llvm/gnu-install-dirs.patch: I've dropped most of the changes to
docs/CMake.rst as they aren't relevant for Nixpkgs and the restructuring
of that file makes it a bit annoying to resolve them via Git.
This also fixes the libunwind build (even with GCC 11 it fails with):
/build/source/libunwind/src/libunwind.cpp:19:5: warning: "__has_feature" is not defined, evaluates to 0 [-Wundef]
19 | #if __has_feature(address_sanitizer)
| ^~~~~~~~~~~~~
/build/source/libunwind/src/libunwind.cpp:19:18: error: missing binary operator before token "("
19 | #if __has_feature(address_sanitizer)
| ^
And the openmp build which failed with this error:
/nix/store/a4yw1svqqk4d8lhwinn9xp847zz9gfma-bash-4.4-p23/bin/bash: CLANG_TOOL-NOTFOUND: command not found
/nix/store/a4yw1svqqk4d8lhwinn9xp847zz9gfma-bash-4.4-p23/bin/bash: CLANG_TOOL-NOTFOUND: command not found
make[2]: *** [libomptarget/deviceRTLs/amdgcn/CMakeFiles/libomptarget-amdgcn-gfx906.dir/build.make:307: libomptarget/deviceRTLs/amdgcn/task.gfx906.bc] Error 127
make[2]: *** [libomptarget/deviceRTLs/amdgcn/CMakeFiles/libomptarget-amdgcn-gfx900.dir/build.make:307: libomptarget/deviceRTLs/amdgcn/task.gfx900.bc] Error 127
The version will initially remain the same so that no additional changes
to the packaging are required (i.e. this commit only includes the
required changes to build from the mono repository instead of individual
tarballs).
The purpose of this package is to continuously improve the LLVM
packaging in Nixpkgs without causing a lot of rebuilds and provide more
recent LLVM builds for users. For more details see:
https://github.com/NixOS/nixpkgs/issues/114828
