JuliaLang
Fix `JULIA_CPU_TARGET` being propagated to workers precompiling stdlib pkgimages
Apparently (thanks ChatGPT) each line in a makefile is executed in a separate shell so adding an export line on one line does not propagate to the next line.
Is there any way we can make sure that this errors if we fail to propagate this again in the future?
I think the canonical way is something like https://github.com/JuliaLang/julia/commit/f106bd9708fd61735f658e3e27ee1f227b829736
Question, why not set the env variable "globally" together with JULIA_DEPOT_PATH and JULIA_LOAD_PATH?
I'm ok with this but if someone has a better or more canonical way feel free to push to this branch. https://github.com/JuliaLang/julia/pull/54093#issuecomment-2058889142 seems easiest to me IIUC though?
With a build from this PR I get
julia> Base.parse_image_targets(Base.parse_cache_header("share/julia/compiled/v1.12/Pkg/tUTdb_fyEGR.ji")[7])
1-element Vector{Base.ImageTarget}:
generic; flags=0; features_en=(cx16)
which seems to suggest JULIA_CPU_TARGET is still not being effectively propagated where needed?
The command to compile the pkgimage seems to include the CPU target properly:
julia_cmd(; cpu_target) = `julia-master/usr/bin/julia -C 'generic;sandybridge,-xsaveopt,clone_all;haswell,-rdrnd,base(1);x86-64-v4,-rdrnd,base(1)' -Jjulia-master/usr/lib/julia/sys.dylib -g1 --startup-file=no`
but yet
julia> cd(joinpath(Sys.BINDIR, "../share/julia/compiled/v1.12/Test"))
julia> Base.parse_image_targets(Base.parse_cache_header("JfdTE_tFidc.ji")[7])
1-element Vector{Base.ImageTarget}:
generic; flags=0; features_en=(cx16)
is Base.precompilepkgs ignoring this? It seems to take in a CacheFlags argument that is empty
is Base.precompilepkgs ignoring this? It seems to take in a CacheFlags argument that is empty
CacheFlags are AFAIU not related to the CPU target and even if it was, my latest commit prints the julia command which does include the target.
I just ran into this while cross-compiling, where the -C native figured out from qemu resulted in illegal instructions when deploying the build. I confirmed that this change now correctly forwards my configured JULIA_CPU_TARGET, so let's merge this after CI passes.
