Hi, Bumping this thread. Is this the correct list to post this on?
Best regards, Matthew Mehrtens > On Feb 8, 2026, at 15:26, 'Matthew Mehrtens' via Open MPI users > <[email protected]> wrote: > > Hi, > > I'm getting a build failure when building Open MPI 5.0.9 with LLVM compilers > on macOS (Apple Silicon). Configure succeeds, but make fails when linking: > > ``` > flang-21: error: unknown argument: '-compatibility_version' > flang-21: error: unknown argument: '-current_version' > flang-21: error: no such file or directory: '81' > flang-21: error: no such file or directory: '81.3' > make[3]: *** [libmpi_usempif08.la] Error 1 > ``` > > I believe this error comes from config/ltmain_flang_darwin.diff. The patch > has three hunks that add flang-specific handling for Darwin linker flags. > Hunks 1 and 3 match against $CC directly using `case $CC in flang*)`, which > fails when the Fortran compiler is specified as a full path (e.g., > /opt/homebrew/opt/flang/bin/flang). Hunk 2 correctly uses `func_cc_basename > $CC` before matching, so it doesn't have this problem. > > Environment: > macOS 26.2 (Darwin 25.2.0), aarch64 (Apple M1 Pro) > Homebrew clang/clang++/flang 21.1.8 > Open MPI 5.0.9 > Configure command: > ``` > $ ./configure \ > CC=/opt/homebrew/opt/llvm/bin/clang \ > CPPFLAGS=-I/opt/homebrew/opt/llvm/include \ > CXX=/opt/homebrew/opt/llvm/bin/clang++ \ > CXXCPPFLAGS=-I/opt/homebrew/opt/llvm/include \ > FC=flang \ > LDFLAGS="-L/opt/homebrew/opt/llvm/lib -L/opt/homebrew/opt/llvm/lib/c++ > -L/opt/homebrew/opt/llvm/lib/unwind -lunwind" \ > --with-hwloc=/opt/homebrew/opt/hwloc \ > --with-libevent=/opt/homebrew/opt/libevent \ > --with-pmix=/opt/homebrew/opt/pmix > ``` > > Build output: > https://www.icloud.com/iclouddrive/00dSkBghSZdP0-MRsCPN7M7pw#ompi-output (let > me know if you cannot access this, and I'll resubmit using a different > service) > > Workaround: Use `FC=flang` (bare name) instead of the full path, with > /opt/homebrew/opt/flang/bin in $PATH. > > Suggested fix: Update hunks 1 and 3 in config/ltmain_flang_darwin.diff to use > `func_cc_basename $CC` and `$func_cc_basename_result` instead of matching $CC > directly, consistent with hunk 2. > > If this is indeed a bug and not intentional, I'd be happy to submit an > issue/PR on GitHub if you can advise on the preferred approach. I wasn't sure > whether to regenerate the diff against a specific libtool version or patch it > directly. > > Thanks, > Matthew Mehrtens > > To unsubscribe from this topic, visit > https://groups.google.com/a/lists.open-mpi.org/d/topic/users/4mIxq9op75k/unsubscribe. > To unsubscribe from this group and all its topics, send an email to > [email protected] > <mailto:[email protected]>. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected].
