erlang: bump to version 20.0

This change bumps the Erlang version from 19.3 to 20.0. As a new major release, there are a few incompatibilities. Details are here:
http://www.erlang.org/news/114

This should fix builds on ppc64le host machines:
http://autobuild.buildroot.net/results/407af9800313d4540643d7625e85acc0c2366892
http://autobuild.buildroot.net/results/2aca54ad48ecfd72242d97499e21a46c1b384b10

---
 package/erlang/erlang.hash | 4 ++--
 package/erlang/erlang.mk   | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

index dc12ccd062..cf820cec1b 100644
@@ -1,3 +1,3 @@
 # md5 from http://www.erlang.org/download/MD5, sha256 locally computed
-md5	a8c259ec47bf84e77510673e1b76b6db	otp_src_19.3.tar.gz
-sha256  fe4a00651db39b8542b04530a48d24b2f2e7e0b77cbe93d728c9f05325bdfe83	otp_src_19.3.tar.gz
+md5     2faed2c3519353e6bc2501ed4d8e6ae7        otp_src_20.0.tar.gz
+sha256  fe80e1e14a2772901be717694bb30ac4e9a07eee0cc7a28988724cbd21476811        otp_src_20.0.tar.gz

index f3684d0c91..733c1d5786 100644
@@ -5,7 +5,7 @@
 ################################################################################
 
 # See note below when updating Erlang
-ERLANG_VERSION = 19.3
+ERLANG_VERSION = 20.0
 ERLANG_SITE = http://www.erlang.org/download
 ERLANG_SOURCE = otp_src_$(ERLANG_VERSION).tar.gz
 ERLANG_DEPENDENCIES = host-erlang

@@ -19,7 +19,7 @@ ERLANG_AUTORECONF = YES
 
 # Whenever updating Erlang, this value should be updated as well, to the
 # value of EI_VSN in the file lib/erl_interface/vsn.mk
-ERLANG_EI_VSN = 3.9.3
+ERLANG_EI_VSN = 3.10
 
 # The configure checks for these functions fail incorrectly
 ERLANG_CONF_ENV = ac_cv_func_isnan=yes ac_cv_func_isinf=yes

2.14.1

Hi Johan,

Thanks for doing this. I had been holding on to a nearly identical patch, but it breaks several erlang packages currently in Buildroot. It didn't look easy to fix, and then I got busy. As I'm sure that you're aware, ejabberd builds fine. I don't think there are many of us Erlang/Buildroot users, so if this still breaks erlang packages that you don't use, I'm fine with removing them.

Two comments:

1. Do you want to bump to Erlang 20.1 since that's the latest?
2. Should we remove support for non-SMP builds now that it's deprecated? I've actually never used the option due to it disabling NIF features used by some libraries.

Having said that, I'd prefer this patch in than not since it moves Buildroot forward, so

Thanks!
Frank

Hello,

What are those packages ?

We definitely need to at least identify those packages. Is this a build time breakage ? Run-time breakage ?

Best regards,

Thomas

Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

Frank, All,

No, I tried it but it has a compiler bug that breaks some Erlang packages in Buildroot.
For example, I got the following error in erlang-p1-xmpp package:
https://bugs.erlang.org/browse/ERL-493

I think it's better to skip OTP 20.1 for now.

I'm not aware of Erlang packages that are broken by OTP 20.0. So same question as Thomas's one, what are those packages?

Excellent question. I haven't thought about it and I've never used this option neither.
Since it is now deprecated, I guess we can remove it. Should I remove it in another patch or revise this patch in a v2 that does both a bump to version 20.0 and a commit to remove the option to disable compilation of the SMP emulator?

Best,

Johan

Johan,

My mistake. I ran into the compiler bug too, but looking back at my experiment, it was OTP 20.1 and not OTP 20.0. I'd have to rerun everything to get the list of errors from 20.1. I agree with you about sticking with OTP 20.0 for now.

Frank

Hello,

Applied to master, thanks.

Thomas

Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

Hello,

This causes a bunch of build failures:

http://autobuild.buildroot.net/?reason=host-erlang-20.0

Seems like host-zlib is needed.

Could you have a look ?

Thanks!

Thomas

Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

Hi,

When trying to reproduce this build failure on my machine, I discovered an unrelated error in flann package:
ERROR: unsafe header/library path used in cross-compilation '/usr/include/hdf5/serial'

I skipped this error by adding this to FLANN_CONF_OPTS:
-DCMAKE_DISABLE_FIND_PACKAGE_HDF5=TRUE

As far as I understand it, this HDF5 dependency from flann is only here for testing purpose and can be disabled in buildroot build. Should I send a patch?

Unfortunately, I failed to reproduce the build failure on erlang but when looking at the log, it indeed failed to find libz even though the configure above seems to find it:
checking for zlib 1.2.5 or higher... yes
checking for zlib inflateGetDictionary presence... checking for library containing inflateGetDictionary... -lz

What bother me is that I can't find which changes from OTP 19.3 to OTP 20.0 could lead to this error.
Anyway, should I simply add host-zlib as an erlang dependency and see if the next build would pass?

Best,

Johan

Am Mon, 04 Dec 2017 17:32:31 +0100 schrieb Johan Oudinet:

Hi,

I think I found the commit, added with version 20.0, which causes the
error: https://github.com/erlang/otp/
commit/640c988fd41f9709b494554b2e5ef1f06f06957e

The autoconf function AC_SEARCH_LIBS adds "-llibrary to LIBS for the first library found to contain function" according to the autoconf manual:
https://www.gnu.org/software/autoconf/manual/autoconf-2.69/html_node/
Libraries.html

And this is exactly what happens:

host-erlang-20.0/erts/config.status contains

S["LIBS"]="-lz -lutil -ldl -lm "

while host-erlang-19.3/erts/config.status lacks -lz:

S["LIBS"]="-lutil -ldl -lm "

The content of this variable is used to build dyntrace.o with this
Makefile: lib/runtime_tools/c_src/Makefile.in, line 98.

Unfortunately I do not have more time to look for a fix today.

Regards, Bernd

Good catch. I tried the workaround of adding --enable-builtin-zlib in HOST_ERLANG_CONF_OPTS and it works.
So this is definitely an issue with how their use of AC_SEARCH_LIBS in erts/configure.in leaks to how dyntrace.o is linked in lib/runtime_tools/c_src/Makefile.in.

I don't think it is acceptable to use --enable-builtin-zlib in buildroot packaging of Erlang but we're making progress.

Johan

Hello,

Yes, please :)

Perhaps it's testing with the 64-bit version of the library, but then using the 32-bit version or the opposite. Can you build with V=1 (or whatever is needed in Erlang to get verbose details), and see what is the exact command that fails to link with zlib ?

Best regards,

Thomas

Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

Am Sun, 03 Dec 2017 10:59:05 +0100 schrieb Thomas Petazzoni:

Hi,

host-erlang depends on host-openssl which depends on host-zlib so this should not be the problem. Anyway I could reproduce the build error on a minimal Debian system which lacks a zlib-dev package, host-zlib was built by buildroot.

More investigation is needed on my part, but for today I have no time left for this, sorry.

Regards, Bernd

Hi,

Thanks, I did the same (i.e., run it from a fresh debian-testing-amd64 chroot) and manage to reproduce the -lz error:
/usr/bin/gcc -c -o ../priv/obj/x86_64-pc-linux-gnu/trace_ip_drv.o -DHAVE_CONFIG_H -Werror=undef -Werror=implicit -Werror=return-type -Wall -Wstrict-prototypes -Wmissing-prototypes -Wdeclaration-after-statement -DUSE_THREADS -D_THREAD_SAFE -D_REENTRANT -DPOSIX_THREADS -D_POSIX_THREAD_SAFE_FUNCTIONS -O2 -I/path/to/host/include -I/path/to/build/host-erlang-20.0/erts/x86_64-pc-linux-gnu -fno-tree-copyrename -I/path/to/host/include -D_GNU_SOURCE -fPIC -I./ -I/path/to/build/host-erlang-20.0/erts/emulator/beam -I/path/to/build/host-erlang-20.0/erts/include -I/path/to/build/host-erlang-20.0/erts/include/x86_64-pc-linux-gnu -I/path/to/build/host-erlang-20.0/erts/include/internal -I/path/to/build/host-erlang-20.0/erts/include/internal/x86_64-pc-linux-gnu -I/path/to/build/host-erlang-20.0/erts/emulator/sys/unix -I/path/to/build/host-erlang-20.0/erts/emulator/sys/common -I../priv/obj/x86_64-pc-linux-gnu -I/path/to/build/host-erlang-20.0/erts/emulator/x86_64-pc-linux-gnu trace_ip_drv.c
/usr/bin/gcc -shared -Wl,-Bsymbolic -o ../priv/lib/x86_64-pc-linux-gnu/dyntrace.so ../priv/obj/x86_64-pc-linux-gnu/dyntrace.o -lz -lutil -ldl -lm /usr/bin/ld: cannot find -lz
collect2: error: ld returned 1 exit status

It does have libz.so in the host so I think the issue is in one of Erlang's makefiles. Their build architecture is a real mess and I'm not sure how to fix it properly, but it clearly miss the HOST_LDFLAGS when linking lib/runtime_tools/c_src/dyntrace.c

I try to add LDFLAGS="$(HOST_LDFLAGS)" to HOST_ERLANG_MAKE_OPTS and it does solve this issue but fails later on:
/usr/bin/gcc -fno-PIE -Werror=undef -Werror=implicit -Werror=return-type -O3 -I/path/to/host/include -I/path/to/build/host-erlang-20.0/erts/x86_64-pc-linux-gnu -fno-tree-copyrename -I/path/to/host/include -D_GNU_SOURCE -DERTS_SMP -DHAVE_CONFIG_H -Wall -Wstrict-prototypes -Wmissing-prototypes -Wdeclaration-after-statement -DUSE_THREADS -D_THREAD_SAFE -D_REENTRANT -DPOSIX_THREADS -D_POSIX_THREAD_SAFE_FUNCTIONS
-DERTS_DIRTY_SCHEDULERS -Ix86_64-pc-linux-gnu/opt/smp -Ibeam
-Isys/unix -Isys/common -Ix86_64-pc-linux-gnu -Ipcre -Ihipe -I../include -I../include/x86_64-pc-linux-gnu -I../include/internal -I../include/internal/x86_64-pc-linux-gnu -c hipe/hipe_mkliterals.c -o obj/x86_64-pc-linux-gnu/opt/smp/hipe_mkliterals.o
/usr/bin/gcc -L/path/to/host/lib -Wl,-rpath,/path/to/host/lib -o /path/to/build/host-erlang-20.0/bin/x86_64-pc-linux-gnu/hipe_mkliterals.smp obj/x86_64-pc-linux-gnu/opt/smp/hipe_mkliterals.o -lz -lutil -ldl -lm /usr/bin/ld: obj/x86_64-pc-linux-gnu/opt/smp/hipe_mkliterals.o:
relocation R_X86_64_32 against `.rodata.str1.8' can not be used when making a shared object; recompile with -fPIC
/usr/bin/ld: final link failed: Nonrepresentable section on output
collect2: error: ld returned 1 exit status

I stop here for today.

Johan