Build failures with Xcode CLT 15.0: scipy (via meson), ecm

[mkoeppe]

SciPy build failure when homebrew's gfortran is used (via meson), reported in https://groups.google.com/g/sage-release/c/1wOBmhvNJqc/m/dT-6HGAjCAAJ

• see also macOS: LD version detection is unable to detect linker for compiler gfortran with Xcode CLT 15.0 mesonbuild/meson#12282

ECM build failure reported in https://groups.google.com/g/sage-release/c/dvPti2UkyjQ/m/Nj4e5zApCQAJ:

 checking what assembly label suffix to use... :
checking if globals are prefixed by underscore... configure: error: Test program links neither with nor without underscore.
********************************************************************************
Error configuring ecm-7.0.5 See the file
    /Users/iohara/Desktop/sage-10.2.beta4/local/var/tmp/sage/build/ecm-7.0.5/src/config.log
for details.

This looks similar to https://gitlab.inria.fr/zimmerma/ecm/-/issues/21856 @dimpase @zimmermann6

Our latest ECM upgrade was made in #31325.

[mkoeppe]

@zimmermann6 GMP seems to have moved away from using a test program for GMP_ASM_UNDERSCORE in acinclude.m4; their current version (6.3.0) says:

dnl  GMP_ASM_UNDERSCORE
dnl  ------------------
dnl  Determine whether global symbols need to be prefixed with an underscore.
dnl  The output from "nm" is grepped to see what a typical symbol looks like.
dnl
dnl  This test used to grep the .o file directly, but that failed with greps
dnl  that don't like binary files (eg. SunOS 4).
dnl
dnl  This test also used to construct an assembler file with and without an
dnl  underscore and try to link that to a C file, to see which worked.
dnl  Although that's what will happen in the real build we don't really want
dnl  to depend on creating asm files within configure for every possible CPU
dnl  (or at least we don't want to do that more than we have to).
dnl
dnl  The fallback on no underscore is based on the assumption that the world
dnl  is moving towards non-underscore systems.  There should actually be no
dnl  good reason for nm to fail though.

[zimmermann6]

does the following patch solve the issue?

diff --git a/acinclude.m4 b/acinclude.m4
index 5d168b6..64cdf6c 100644
--- a/acinclude.m4
+++ b/acinclude.m4
@@ -329,7 +329,10 @@ else
   if test $tmp_result = yes; then
     gmp_cv_asm_underscore=no
   else
-    AC_MSG_ERROR([Test program links neither with nor without underscore.])
+dnl AC_MSG_ERROR([Test program links neither with nor without underscore.])
+dnl In GMP 6.3.0, there fallback is on no underscore, based on the assumption
+dnl that the worldis moving towards non-underscore systems. We do the same.
+    gmp_cv_asm_underscore=no
   fi
 fi
 rm -f conftes1* conftes2* a.out

[roed314]

I ran into the scipy build error; for me I think it's due to my build environment being messed up due to a MacOS upgrade. The things that fixed it were

xcode-select --install
sudo xcode-select --switch /Library/Developer/CommandLineTools
brew install gcc

Now I'm stuck on a build error in sagelib, but that's hopefully unrelated...

[kcrisman]

For what it's worth, same problem. @roed314 solution did not work for me (for one thing, I already had brew installed gcc). I had to install gfortran for it to work.