Add patch to js-1.8.0 which fixes inline compilation error with gcc 4.7+

[dantealiegri]

Hey guys, this enables erlang_js to be compiled with gcc 4.7+, as gcc now wants inline on a function definition when always_inline is specified.

[Vagabond]

I can confirm that this does not harm compilation on Arch linux with gcc 4.8. However, my version of GCC did not need this patch to be able to compile erlang_js.

[cmeiklejohn]

@dantealiegri Can you provide the platform where this error is reproducible?

[dantealiegri]

@cmeiklejohn : it's on a linode machine running Fedora 17.

[dante@li549-157 ~]$ gcc --version gcc (GCC) 4.7.2 20120921 (Red Hat 4.7.2-2)

[dante@li549-157 ~]$ rpm -qa gcc gcc-4.7.2-2.fc17.i686

[cmeiklejohn]

I've just built a Fedora 17 VM and couldn't reproduce this using gcc-4.7.2-2.fc17.x86_64.

[dantealiegri]

Well that is exciting and awkward.

gcc -o Linux_All_OPT.OBJ/jslock.o -c -Wall -Wno-format -MMD -Os -DXP_UNIX -DSVR4 -DSYSV -D_BSD_SOURCE -DPOSIX_SOURCE -DHAVE_LOCALTIME_R -DHAVE_VA_COPY -DVA_COPY=va_copy -DPIC -fPIC -UDEBUG -DNDEBUG -UDEBUG_dante -DJS_THREADSAFE -DEDITLINE -ILinux_All_OPT.OBJ -I/home/dante/riak/riak-1.4.0/deps/erlang_js/c_src/system/include/nspr -DHAVE_VA_COPY -DVA_COPY=va_copy jslock.c jslock.c: In function âClaimTitleâ: jslock.c:389:14: warning: variable âstatâ set but not used [-Wunused-but-set-variable] jslock.c: In function âjs_SuspendThreadâ: jslock.c:905:14: warning: variable âstatâ set but not used [-Wunused-but-set-variable] jslock.c: In function âjs_ResumeThreadâ: jslock.c:935:14: warning: variable âstatâ set but not used [-Wunused-but-set-variable] jslock.c: At top level: jslock.c:997:1: warning: always_inline function might not be inlinable [-Wattributes] jslock.c:983:1: warning: always_inline function might not be inlinable [-Wattributes] jslock.c:130:1: warning: always_inline function might not be inlinable [-Wattributes] jslock.c: In function âjs_LockTitleâ: jslock.c:983:1: error: inlining failed in call to always_inline âjs_Lockâ: function body can be overwritten at link time jslock.c:1055:9: error: called from here jslock.c: In function âjs_UnlockTitleâ: jslock.c:997:1: error: inlining failed in call to always_inline âjs_Unlockâ: function body can be overwritten at link time jslock.c:1104:9: error: called from here jslock.c: In function âjs_TransferTitleâ: jslock.c:997:1: error: inlining failed in call to always_inline âjs_Unlockâ: function body can be overwritten at link time jslock.c:1178:5: error: called from here

Is what I am getting.

[cmeiklejohn]

Just tried again, no dice.

[root@feltmad erlang_js]# rpm -qa gcc
gcc-4.7.2-2.fc17.x86_64
[root@feltmad erlang_js]#  gcc --version
gcc (GCC) 4.7.2 20120921 (Red Hat 4.7.2-2)
Copyright (C) 2012 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

Are you building Riak or standalone erlang_js?

[cmeiklejohn]

I just realized you are running 32-bit and not 64-bit. I rebuilt my box and I'm still unable to reproduce this:

[root@feltmad erlang_js]# uname -a
Linux feltmad 3.3.7-1.fc17.i686.PAE #1 SMP Mon May 21 22:42:05 UTC 2012 i686 i686 i386 GNU/Linux
[root@feltmad erlang_js]# rpm -qa gcc
gcc-4.7.2-2.fc17.i686
[root@feltmad erlang_js]# gcc --version
gcc (GCC) 4.7.2 20120921 (Red Hat 4.7.2-2)
Copyright (C) 2012 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

[root@feltmad erlang_js]#

Could there be anything else unusual about your configuration? (are you building with any non-standard CFLAGS?)

[dantealiegri]

I'm building as part of riak 1.4. I'm going to avoid the pure insanity of some difference between the rpms, and I'm mostly sure this can't involve headers. The only other thing I can think of is that the flags to the compiler are different. Can you show me what the compilation line for that file is, on your working version?

[cmeiklejohn]

gcc -o Linux_All_OPT.OBJ/jslock.o -c -Wall -Wno-format -MMD -DGCC_OPT_BUG -Os -DXP_UNIX -DSVR4 -DSYSV -D_BSD_SOURCE -DPOSIX_SOURCE -DHAVE_LOCALTIME_R -DX86_LINUX -UDEBUG -DNDEBUG -UDEBUG_root -DJS_THREADSAFE -DEDITLINE -ILinux_All_OPT.OBJ -I/root/erlang_js/c_src/system/include/nspr -DHAVE_VA_COPY -DVA_COPY=va_copy jslock.c

[cmeiklejohn]

It appears the major differences here are related to -DPIC -fPIC. I've manually changed my CFLAGS to include these switches and I was able to reproduce the problem.

[cmeiklejohn]

However, using your branch I still run into the following compilation issue:

jslock.c: At top level:
jslock.c:997:1: warning: always_inline function might not be inlinable [-Wattributes]
jslock.c:983:1: warning: always_inline function might not be inlinable [-Wattributes]
jslock.c:130:1: warning: always_inline function might not be inlinable [-Wattributes]
jslock.c: In function ‘js_LockTitle’:
jslock.c:983:1: error: inlining failed in call to always_inline ‘js_Lock’: function body can be overwritten at link time
jslock.c:1055:9: error: called from here
jslock.c: In function ‘js_UnlockTitle’:
jslock.c:997:1: error: inlining failed in call to always_inline ‘js_Unlock’: function body can be overwritten at link time
jslock.c:1104:9: error: called from here
jslock.c: In function ‘js_TransferTitle’:
jslock.c:997:1: error: inlining failed in call to always_inline ‘js_Unlock’: function body can be overwritten at link time
jslock.c:1178:5: error: called from here
make[4]: *** [Linux_All_OPT.OBJ/jslock.o] Error 1
make[4]: Leaving directory `/root/erlang_js_fixed/c_src/js/src'
make[3]: *** [all] Error 2
make[3]: Leaving directory `/root/erlang_js_fixed/c_src/js/src'
make[2]: *** [/root/erlang_js_fixed/c_src/system/lib/libjs.a] Error 2
make[2]: Leaving directory `/root/erlang_js_fixed/c_src'
make[1]: *** [c_src] Error 2
make[1]: Leaving directory `/root/erlang_js_fixed'
ERROR: Command [compile] failed!
make: *** [all] Error 1

[dantealiegri]

is jstypes.h modified when that compile fails? I think configure needs to be re-run after the patch is applied.

[cmeiklejohn]

I've confirmed that the patch is applied, and the build still fails.

[dantealiegri]

gotcha. If you change

define JS_INLINE attribute((always_inline)) inline

to

error "Set Inline"

do you get an error on compile?

[cmeiklejohn]

Actually, I found the problem, the patch wasn't applying because my test VM didn't have 'patch'. Derp. Otherwise, this looks good to me, so I'm going to :+1: it.

[cmeiklejohn]

Thanks!