Only force -arch x86_64 on Intel macOS

[BenjaminDEMAILLE]

Fixes #695

Problem

The external_libs Makefiles append -arch x86_64 on any Darwin host:

UNAME_S := $(shell uname -s)
ifeq ($(UNAME_S),Darwin)
 CXXFLAGS += -arch x86_64
endif

On Apple Silicon the helper libraries are therefore cross-compiled for x86_64 while src/ is compiled for arm64, and the final link fails:

ld: warning: ignoring file external_libs/qf/qf.a, building for macOS-arm64
    but attempting to link with file built for macOS-x86_64
Undefined symbols for architecture arm64:
  "_qf", referenced from:
      get_davies_pv(double const&, Eigen::Ref<...>, bool const&) in SKAT.cpp.o
ld: symbol(s) not found for architecture arm64

This affects mvtnorm, pgenlib, qf, quadpack and remeta, so an arm64 macOS build of v4.1.3 cannot complete without editing the Makefiles.

Fix

Guard the flag with uname -m. Intel macOS builds are unchanged; arm64 builds now keep the host architecture. The flag is redundant on Intel too (the compiler already targets the host), but keeping it preserves the existing behaviour for anyone relying on it.

Testing

macOS 26.6 on arm64 (Apple Silicon), Homebrew clang, gfortran from GCC 15, BGEN v1.1.7:

BGEN_PATH=... HAS_BOOST_IOSTREAM=1 cmake -S . -B build && cmake --build build
file build/regenie
  build/regenie: Mach-O 64-bit executable arm64
build/regenie --version
  v4.1.3.gz

Step 1 on the bundled example data completes normally:

regenie --step 1 --bed example/example --covarFile example/covariates.txt \
        --phenoFile example/phenotype_bin.txt --bsize 100 --bt --lowmem --out fit
  0.99  : Rsq = 0.448642, MSE = 0.0981588, -logLik/N = 0.318482
  List of blup files written to: [fit_pred.list]

This came up while packaging regenie for Homebrew, where builds run natively on arm64 runners.

🤖 Generated with Claude Code

[joellembatchou]

Good catch thanks!

[BenjaminDEMAILLE]

One question while this is open: once it is merged, do you have a rough timeline for the next tagged release?

I ask because Homebrew builds from release tarballs rather than from a branch, so the fix only reaches users there when it is in a tag. Until then the formula has to carry the equivalent one-line patch itself, which reviewers understandably ask about: Homebrew/homebrew-core#302933

No pressure on the timing, I just want to know whether to keep the local patch or wait for the tag.