gcc11 / libgcc11: not building for 32bit SnowLeopard ld: illegal text-relocation to ___stack_chk_guard in /usr/lib/libSystem.dylib

[kencu (Ken)]

gcc11/libgcc11 is not presently building on SnowLeopard 32bit.

The first error is:

:info:build /opt/local/bin/clang++-mp-9.0 -arch i386 -std=c++11   -g  -DIN_GCC -fPIC   -fno-strict-aliasing -fno-exceptions -fno-rtti -fasynchronous-unwind-tables -W -Wall -Wno-narrowing -Wwrite-strings -Wcast-qual -Wno-error=format-diag -Wno-format -Wmissing-format-attribute -Woverloaded-virtual -pedantic -Wno-long-long -Wno-variadic-macros -Wno-overlength-strings -fno-common  -DHAVE_CONFIG_H  -DGENERATOR_FILE -fno-PIE -L/opt/local/lib -Wl,-headerpad_max_install_names -no-pie -o build/genhooks \
:info:build 	    build/genhooks.o build/errors.o ../build-i386-apple-darwin10/libiberty/pic/libiberty.a
:info:build clang: warning: argument unused during compilation: '-no-pie' [-Wunused-command-line-argument]
:info:build ld: illegal text-relocation to ___stack_chk_guard in /usr/lib/libSystem.dylib from __ZL18emit_documentationPKc in build/genhooks.o for architecture i386

that is a common error with 32bit builds, and can usually be overcome with this:

if {${configure.build_arch} eq "i386"} {
    configure.env-append                LDFLAGS=-Wl,-read_only_relocs,suppress
}

However, although that allows genhooks to be built, that leads to a bus error later when it is run:

:info:build build/genhooks -d \
:info:build 			/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_lang_gcc11/libgcc11/work/gcc-11.1.0/gcc/doc/tm.texi.in > tmp-tm.texi
:info:build /bin/sh: line 1: 53082 Bus error               build/genhooks -d /opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_macports_release_tarballs_ports_lang_gcc11/libgcc11/work/gcc-11.1.0/gcc/doc/tm.texi.in > tmp-tm.texi
:info:build make[3]: *** [s-tm-texi] Error 138
:info:build make[3]: *** Waiting for unfinished jobs....

so that will not be a solution.

Building with gcc-4.8 does work, although it is a bit tricky in our MacPorts configuration because gcc-4.8 is supposed to now use libgcc11 for it's libraries, and so that makes a catch-22, and you need to do some fancy footwork to do this:

$ port -v installed libgcc11
The following ports are currently installed:
  libgcc11 @11.1.0_2 (active) requested_variants='' platform='darwin 10' archs='i386' date='2021-07-01T17:55:22-0700'

What does seem to work when building with clang-9.0 is forcing pie off like this:

--- gcc/configure.old	2021-07-01 18:31:32.000000000 -0700
+++ gcc/configure	2021-07-01 18:31:54.000000000 -0700
@@ -31798,7 +31798,7 @@
 int main(void) {return 0;}
 _ACEOF
 if ac_fn_cxx_try_compile "$LINENO"; then :
-  gcc_cv_c_no_fpie=yes
+  gcc_cv_c_no_fpie=no
 else
   gcc_cv_c_no_fpie=no
 fi
@@ -31825,7 +31825,7 @@
 int main(void) {return 0;}
 _ACEOF
 if ac_fn_cxx_try_link "$LINENO"; then :
-  gcc_cv_no_pie=yes
+  gcc_cv_no_pie=no
 else
   gcc_cv_no_pie=no
 fi

Just investigating if that approach has any undesired side effects now.

[kencu (Ken)]

Upstream points this out to be a bug in the 32 bit clang implementation, which it indeed seems to be.

They have a more intricate patch that still allows no-pie to be used where it should be used and fixes the issue...testing now.

It will be harder to maintain that more intricate patch, but it's a better patch.

Older gcc versions may also need that fix to bootstrap with clang 32 bit. Or we can arrange to bootstrap clang with gcc, starting with gcc-4.2 (bootstraps up to gcc10), or gcc-4.8 or gcc-5.0 (bootstraps gcc11+ as well).

[kencu (Ken)]

In 7ffd82db0cba8a030346bc360bd0685c26c1de00/macports-ports (master):

gcc11: fix 32bit clang bootstrap

closes: #63161

uses upstream patch from gcc maintainer @iains

[catap (Kirill A. Korinsky)]

In 3833b50a3003539b5d0565a8cf7997f60405ce26/macports-ports (master):

gcc10-bootstrap: fixed build on 10.8..10.13

This is well know issue: #63161

Thus, it also adds --without-zstd as lucky guess.

[catap (Kirill A. Korinsky)]

In 2bf46bc6244cfa1f92e08da480f41395fd42dfbb/macports-ports (master):

gcc10-bootstrap: use flat patches

Idea to use muniversal PG and arch-specific patches wasn't good that
leads to broken build due to not applied patches on 10.8..10.13,
and may produce unpredicted result when ppc and i386 is mixed.

Here, patches a bit reorganized to be applied all together,
that allows to use the same sources to build it.

It also includes backported version a fix for
#63161 and all macOS 12-related patches
are put into one.