Bug #7485: ruby cannot build on mingw32 or FreeBSD 8 due to missing __sync_val_compare_and_swap

• Status: Closed
• Assignee: drbrain (Eric Hodel)
• Target version: 2.0.0
• ruby -v: trunk

[drbrain (Eric Hodel)]

mingw32 GCC has the other atomic builtins but not this one. The attached patch checks for the GCC builtins used in ruby_atomic.h but not checked for by configure.

[nobu (Nobuyoshi Nakada)]

__sync_or_and_fetch wasn't available in Apple derivative gcc.

[shyouhei (Shyouhei Urabe)]

@nobu (Nobuyoshi Nakada) Why do we have to bother old broken compiler?

[kosaki (Motohiro KOSAKI)]

I guess -march=i486 enable __sync_val_compare_and_swap.

[nobu (Nobuyoshi Nakada)]

Hmmm, if it can help, should add it by the default on mingw32?

[luislavena (Luis Lavena)]

Hello,

I just tested several native compilers on Windows 7 (without your patch) here are the results for those that define HAVE_GCC_ATOMIC_BUILTINS:

C:\Users\Luis\Code\ruby\ruby\build>gcc --version
gcc (tdm-1) 4.5.2

C:\Users\Luis\Code\ruby\ruby\build>cat .ext/include/i386-mingw32/ruby/config.h | grep ATOMIC
#define HAVE_GCC_ATOMIC_BUILTINS 1

C:\Users\Luis\Code\ruby\ruby\build>gcc --version
gcc (tdm-1) 4.6.1

C:\Users\Luis\Code\ruby\ruby\build>cat .ext/include/i386-mingw32/ruby/config.h | grep ATOMIC
#define HAVE_GCC_ATOMIC_BUILTINS 1

C:\Users\Luis\Code\ruby\ruby\build>gcc --version
gcc (rubenvb-4.7.2-release) 4.7.2

C:\Users\Luis\Code\ruby\ruby\build>cat .ext/include/i386-mingw32/ruby/config.h | grep ATOMIC
#define HAVE_GCC_ATOMIC_BUILTINS 1

In all scenarios it found the atomic builtins

On Windows XP (32bits), ATOMIC do get defined:

C:\Code\ruby\ruby\build>ver

Microsoft Windows XP [Version 5.1.2600]

C:\Code\ruby\ruby\build>gcc --version
gcc (rubenvb-4.7.2-release) 4.7.2

C:\Code\ruby\ruby\build>cat .ext\include\i386-mingw32\ruby\config.h | grep ATOMIC
#define HAVE_GCC_ATOMIC_BUILTINS 1

I was also able to compile Ruby without issues with "make all test"

I don't get the problem reported on this ticket on neither my systems.

I'm missing something?

Thank you.

[jonforums (Jon Forums)]

I agree with Luis' results and think we've just not fully debugged Eric's XP VM setup as in #7427.

With this snippet http://pastie.org/5460872 and Kosaki-san's feedback I see:

mingw-w64 32bit gcc 4.7.2

C:\sandbox>gcc --version
gcc (rubenvb-4.7.2-release) 4.7.2

C:\sandbox>gcc -Wall -o atomics.exe gcc_atomic.c

C:\sandbox>atomics.exe
Got 1

C:\sandbox>gcc -Wall -march=i386 -o atomics.exe gcc_atomic.c
C:\Users\Jon\AppData\Local\Temp\ccK9GPwk.o:gcc_atomic.c:(.text+0x2e): undefined reference
to `__sync_val_compare_and_swap_4'
collect2.exe: error: ld returned 1 exit status

mingw 32bit gcc 4.6.2

C:\sandbox>gcc --version
gcc (GCC) 4.6.2

C:\sandbox>gcc -Wall -o atomics.exe gcc_atomic.c
C:\Users\Jon\AppData\Local\Temp\ccpeH4Vt.o:gcc_atomic.c:(.text+0x2e): undefined reference
to `__sync_val_compare_and_swap_4'
collect2: ld returned 1 exit status

C:\sandbox>gcc -Wall -march={i486,i586,i686,native,core2} -o atomics.exe gcc_atomic.c

C:\Users\Jon\Documents\CDev\sandbox>atomics.exe
Got 1

Bottom line: even with these snippet results I see no build failures due to the current mingw32 build configuration.

[drbrain (Eric Hodel)]

• Subject changed from ruby cannot build on mingw32 due to missing __sync_val_compare_and_swap to ruby cannot build on mingw32 or FreeBSD 8 due to missing __sync_val_compare_and_swap

[drbrain (Eric Hodel)]

=begin
I'm also experiencing this on FreeBSD 8 with gcc 4.2.1:

$ gcc -v
Using built-in specs.
Target: i386-undermydesk-freebsd
Configured with: FreeBSD/i386 system compiler
Thread model: posix
gcc version 4.2.1 20070719 [FreeBSD]
$ make
linking miniruby
thread.o(.text+0x3f20): In function rb_threadptr_execute_interrupts': /data/home/drbrain/Work/svn/ruby/thread.c:1807: undefined reference to __sync_val_compare_and_swap_4'
*** Error code 1

Stop in /data/home/drbrain/Work/svn/ruby.

So it seems to be an old GCC issue, not a mingw-specific issue.

Is ruby 2.0 abandoning gcc 4.2.x?

=end

[kosaki (Motohiro KOSAKI)]

Hi

Does following patch fix your issue?

diff --git a/configure.in b/configure.in
index c916cd5..4a34334 100644
--- a/configure.in
+++ b/configure.in
@@ -1534,7 +1534,7 @@ AS_CASE(["$target_os"],
[])

-AS_CASE(["$target_cpu-$target_os"], [[i[3-6]86-linux*]], [
+AS_CASE(["$target_cpu-$target_os"], [[i[3-6]86-*]], [
AC_CACHE_CHECK([for __sync_val_compare_and_swap], [rb_cv_gcc_co
AC_TRY_LINK([unsigned long atomic_var;],
[

[drbrain (Eric Hodel)]

This patch works, thanks!

[kosaki (Motohiro KOSAKI)]

• Status changed from Open to Assigned
• Assignee changed from nobu (Nobuyoshi Nakada) to drbrain (Eric Hodel)

@drbrain (Eric Hodel), could you please commit the patch instead of me? I have no test environment and I don't like commit w/o make test-all.

[drbrain (Eric Hodel)]

I have confirmed it fixes mingw32 as well, I will commit it after lunch. Thanks for the patch!

[drbrain (Eric Hodel)]

• Status changed from Assigned to Closed
• % Done changed from 0 to 100

This issue was solved with changeset r38386.
Eric, thank you for reporting this issue.
Your contribution to Ruby is greatly appreciated.
May Ruby be with you.

---

• configure.in (HAVE_GCC_ATOMIC_BUILTINS): Set -march=i486 to enable
  __sync_val_compare_and_swap. Patch by KOSAKI Motohiro.
  [ruby-trunk - Bug #7485]

[drbrain (Eric Hodel)]

Please check my ChangeLog message for accuracy.

[kosaki (Motohiro KOSAKI)]

Please check my ChangeLog message for accuracy.

Looks correct. thank you!