[Bug 219779] devel/llvm40: Misconfiguration on ARM targets

[bugzilla-noreply at freebsd.org]

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=219779

            Bug ID: 219779
           Summary: devel/llvm40: Misconfiguration on ARM targets
           Product: Ports & Packages
           Version: Latest
          Hardware: arm
                OS: Any
            Status: New
          Severity: Affects Many People
          Priority: ---
         Component: Individual Port(s)
          Assignee: brooks at FreeBSD.org
          Reporter: mmel at FreeBSD.org
                CC: dim at FreeBSD.org, freebsd-arm at FreeBSD.org
                CC: freebsd-arm at FreeBSD.org
             Flags: maintainer-feedback?(brooks at FreeBSD.org)
          Assignee: brooks at FreeBSD.org

I found several configuration problems in all llvm ports:
1) In port's Makefile, the block marked as 
   ' # keep in sync with /usr/src/lib/clang/clang.build.mk'
   is significantly out of sync with clang.build.mk :)

2) For ARM, llvm's cmake/config.guess doesn't return proper ABI.

3) CONFIGURE_TARGET is not a passed to real llvm cmake at all. (Look to
   default triple for llvm – it doesn’t contain 'portbld')

All these issues causes that default llvm configuration is invalid (for ARM).

Fix for 1) is trivial.

Fix for 2) is also easy, but question is why we must explicitly specify ABI
for armv6-portbld-freebsd12.0 target. Dimitry, please, can you comment this? 
One can expect that pure triple 'armv6-portbld-freebsd12.0' is enough for
selecting right system's ABI (but I'm not sure).

Point 3) is more interesting. Seems that
'CMAKE_ARGS+= -DLLVM_DEFAULT_TARGET_TRIPLE=${CONFIGURE_TARGET}'
fixes default triple for normal (driver based) compilation. But it looks
like JIT compilers (an API based) doesn't uses it at all (is this bug or
feature?) and  
'CMAKE_ARGS+= -DLLVM_HOST_TRIPLE=${CONFIGURE_TARGET}'
is also necessary. At this point I'm simply not sure if this is expected
behavior – if yes, then fix is easy, but it depends on resolution of point 2).

Can you, please, help me with proper resolution of remaining questionable
points?

Thanks,
Michal

-- 
You are receiving this mail because:
You are the assignee for the bug.