gcc-10.1.0 (wait for LFS)

[Douglas R. Reno]

New major version

A year has lapsed away since the release of last major
GCC release, more than 33 years passed since the first
public GCC release and the GCC developers survived
repository conversion from SVN to GIT earlier this year.

Today, we are glad to announce another major GCC release, 10.1.

This release makes great progress in the C++20 language support,
both on the compiler and library sides [1], some C2X enhancements,
various optimization enhancements and bug fixes, several new
hardware enablement changes and enhancements to the compiler back-ends
and many other changes.  There is even a new experimental static
analysis pass [2].

Some code that compiled successfully with older GCC versions might require
source changes, see http://gcc.gnu.org/gcc-10/porting_to.html for
details.

See

  https://gcc.gnu.org/gcc-10/changes.html

for more information about changes in GCC 10.1.

This release is available from the FTP servers listed here:

 http://www.gnu.org/order/ftp.html

The release is in gcc/gcc-10.1.0/ subdirectory.

If you encounter difficulties using GCC 10.1, please do not contact me
directly.  Instead, please visit http://gcc.gnu.org for information about
getting help.

Driving a leading free software project such as GNU Compiler Collection
would not be possible without support from its many contributors.
Not to only mention its developers but especially its regular testers
and users which contribute to its high quality.  The list of individuals
is too large to thank individually!


[1] See https://gcc.gnu.org/projects/cxx-status.html,
    and https://gcc.gnu.org/onlinedocs/libstdc++/manual/status.html#status.iso.2020

[2] See https://developers.redhat.com/blog/2020/03/26/static-analysis-in-gcc-10

[Pierre Labastie]

Xi Ruoyao has found several issues because of the -fno-common default. See: ​http://linuxfromscratch.org/~xry111/fno-common-patches/

Also, on blfs-dev mailing list, from Ryan Marsaw:

The cpio package has breakage with GCC 10.  Here's a link to the issue
(Gentoo):

https://bugs.gentoo.org/705900

The fix is rather easy:

sed -i '/The name/,+2 d' src/global.c

Also the link to upstream fix: ​https://git.savannah.gnu.org/cgit/cpio.git/commit/?id=641d3f489cf6238bb916368d4ba0d9325a235afb

[Pierre Labastie]

Updated to gcc-10.1 at r23088. Now to applying fixes and testing.

[thomas]

ntp: There is a patch available on ntp.org which is scheduled for ntp-4.2.8p15 (​https://bugs.ntp.org/show_bug.cgi?id=3659). In the meanwhile, added -fcommon to CFLAGS, r23093

[Pierre Labastie]

sharutils: CFLAGS=-fcommon added at r23092

[thomas]

xf86-video-ati, xf86-vidoe-intel: CFLAGS=-fcommon added at r23094

[thomas]

QT: Not an -fcommon issue, but also caused by gcc-10. QT compilation terminates with errors like

g++ -o .obj/platform/default/bidi.o platform/default/bidi.cpp
agent:2020/03/12 11:42:37 build.go:253: platform/default/bidi.cpp: In member function ‘void mbgl::BiDi::mergeParagraphLineBreaks(std::set<long unsigned int>&)’:
agent:2020/03/12 11:42:37 build.go:253: platform/default/bidi.cpp:63:24: error: ‘runtime_error’ is not a member of ‘std’

According to the bugreport on qt.io (​https://bugreports.qt.io/browse/QTBUG-82850), it has something to do with standards of gcc. Cannot explain details. I reformatted the patch so it can be applied in the root source directory as usual.

Done in r23096

[Bruce Dubbs]

Postfix needed an update to version 3.5.1 for gcc10.

[Pierre Labastie]

Replying to thomas:

xf86-video-ati, xf86-vidoe-intel: CFLAGS=-fcommon added at r23094

In case this does not get fixed upstream, an alternative way is:

sed 's/const/extern &/' -i src/gen9_hevc_enc_kernels_binary.h

[edit]: see reply to this comment

[thomas]

qtwebengine: Added a patch which fixes issues with gcc-10. Also merged a patch (origin ArchLinux) to enable usage of system-icu again. (r23107)

[Pierre Labastie]

Replying to pierre.labastie:

Replying to thomas:

xf86-video-ati, xf86-vidoe-intel: CFLAGS=-fcommon added at r23094

In case this does not get fixed upstream, an alternative way is:

sed 's/const/extern &/' -i src/gen9_hevc_enc_kernels_binary.h

Sorry, this is for intel-vaapi-driver, not xf86-video-intel.

[Pierre Labastie]

intel-vaapi-driver fixed at r23108

[Pierre Labastie]

freeglut needs a fix: ​https://sourceforge.net/p/freeglut/code/1863/

Maybe just wait for next version and add CFLAGS=-fcommon for now...

Added -DCMAKE_C_FLAGS=-fcommon at r23109

[Pierre Labastie]

cpio fixed at r23110.

[Pierre Labastie]

p7zip needs a fix: ​https://src.fedoraproject.org/rpms/p7zip/c/c3938a10a21b4eda3cc002c5e0c65ac7191d8b6b?branch=master

[Douglas R. Reno]

Valgrind needs a fix:

gcc -Winline -Wall -Wshadow -Wno-long-long -g -fno-stack-protector   -m64     -o libvexmultiarch_test libvexmultiarch_test-libvex_test.o ../../VEX/libvexmultiarch-amd64-linux.a ../../VEX/libvex-amd64-linux.a  
/usr/bin/ld: ../../VEX/libvex-amd64-linux.a(libvex_amd64_linux_a-guest_s390_helpers.o):/sources/valgrind-3.15.0/valgrind-3.15.0/VEX/priv/guest_s390_defs.h:289: multiple definition of `s390x_vec_op_t'; ../../VEX/libvexmultiarch-amd64-linux.a(libvexmultiarch_amd64_linux_a-multiarch_main_main.o):/sources/valgrind-3.15.0/valgrind-3.15.0/VEX/priv/guest_s390_defs.h:289: first defined here
/usr/bin/ld: ../../VEX/libvex-amd64-linux.a(libvex_amd64_linux_a-guest_s390_toIR.o):/sources/valgrind-3.15.0/valgrind-3.15.0/VEX/priv/guest_s390_defs.h:289: multiple definition of `s390x_vec_op_t'; ../../VEX/libvexmultiarch-amd64-linux.a(libvexmultiarch_amd64_linux_a-multiarch_main_main.o):/sources/valgrind-3.15.0/valgrind-3.15.0/VEX/priv/guest_s390_defs.h:289: first defined here
/usr/bin/ld: ../../VEX/libvex-amd64-linux.a(libvex_amd64_linux_a-host_s390_defs.o):/sources/valgrind-3.15.0/valgrind-3.15.0/VEX/priv/guest_s390_defs.h:289: multiple definition of `s390x_vec_op_t'; ../../VEX/libvexmultiarch-amd64-linux.a(libvexmultiarch_amd64_linux_a-multiarch_main_main.o):/sources/valgrind-3.15.0/valgrind-3.15.0/VEX/priv/guest_s390_defs.h:289: first defined here
/usr/bin/ld: ../../VEX/libvex-amd64-linux.a(libvex_amd64_linux_a-host_s390_isel.o):/sources/valgrind-3.15.0/valgrind-3.15.0/VEX/priv/guest_s390_defs.h:289: multiple definition of `s390x_vec_op_t'; ../../VEX/libvexmultiarch-amd64-linux.a(libvexmultiarch_amd64_linux_a-multiarch_main_main.o):/sources/valgrind-3.15.0/valgrind-3.15.0/VEX/priv/guest_s390_defs.h:289: first defined here
collect2: error: ld returned 1 exit status

Note that I originally noticed this while running the regression test. That makes me wonder about the functionality of the program itself.

Looking for a fix...

[Douglas R. Reno]

Valgrind:

It looks like we'll need to change an 'enum' to a 'typedef enum' in VEX/priv/guest_s390_defs.h. There are multiple 'enum's though, and only one of them are affected. The sed I found is:

sed -i '267 s/enum/typedef enum/' VEX/priv/guest_s390_defs.h

Testing now

[Douglas R. Reno]

Valgrind

Attempting to run Valgrind on a program may result in a segfault because GCC-10 changed the way that assembly was produced in a way that Valgrind's developers did not account for (from my understanding, this is due to -fno-common, but I'm not sure). The tests show this with:

== 711 tests, 11 stderr failures, 5 stdout failures, 0 stderrB failures, 0 stdoutB failures, 0 post failures ==
memcheck/tests/amd64/shr_edx             (stdout)
memcheck/tests/amd64/shr_edx             (stderr)
memcheck/tests/overlap                   (stderr)
helgrind/tests/hg05_race2                (stderr)
helgrind/tests/tc09_bad_unlock           (stderr)
drd/tests/sem_open2                      (stderr)
drd/tests/sem_open_traced                (stderr)
none/tests/amd64/fxtract                 (stdout)
none/tests/amd64/fxtract                 (stderr)
none/tests/amd64/jrcxz                   (stdout)
none/tests/amd64/jrcxz                   (stderr)
none/tests/amd64/sbbmisc                 (stdout)
none/tests/amd64/sbbmisc                 (stderr)
none/tests/amd64/shrld                   (stdout)
none/tests/amd64/shrld                   (stderr)
exp-sgcheck/tests/stackerr               (stderr)

Some of the commits that fix this are (these are all related to GCC-10):

​https://sourceware.org/git/?p=valgrind.git;a=commit;h=fe6805efc1b6db0cfa5f1cd5fb1854775cbfa31a ​https://sourceware.org/git/?p=valgrind.git;a=commit;h=2dab3249867615055a680cad7fccb22411587579 ​https://sourceware.org/git/?p=valgrind.git;a=commit;h=4c39fd9a0098648e19ab6e436a515dc0d9fd387e ​https://sourceware.org/git/?p=valgrind.git;a=commit;h=ee18ab11ddf41a7a484fea01e8738c32bae630d9

I'm going to roll up a patch for this.

[Pierre Labastie]

openjdk needs a fix. Looks like --with-extra-cflags=-fcommon should do

[Douglas R. Reno]

Valgrind fixed at r23114

[Pierre Labastie]

Replying to pierre.labastie:

openjdk needs a fix. Looks like --with-extra-cflags=-fcommon should do

Unfortunately not. There are also segfaults, which may be due to too aggressive optimization, according to bug reports (for other versions of openjdk: java-12 does not seem to be in wide use).

[Bruce Dubbs]

Replying to pierre.labastie:

p7zip needs a fix: ​https://src.fedoraproject.org/rpms/p7zip/c/c3938a10a21b4eda3cc002c5e0c65ac7191d8b6b?branch=master

p7zip fixed at revision 23116.

[Pierre Labastie]

texlive-2020 does not need a fix.

[Pierre Labastie]

xine-lib needs:

sed 's/^const/extern &/' -i src/libreal/real_common.h

[Douglas R. Reno]

xf86-video-amdgpu has multiple definition errors

[Douglas R. Reno]

xf86-video-amdgpu needs:

sed -i 's/^miPointerSpriteFuncRec/extern &/' src/drmmode_display.h

[Douglas R. Reno]

xf86-video-amdgpu fixed at r23123

[Pierre Labastie]

xine-lib fixed at r23125

[Pierre Labastie]

lxtask has "multiple definitions" errors. Found this patch: ​https://gitweb.gentoo.org/repo/gentoo.git/tree/lxde-base/lxtask/files/lxtask-0.1.7-fix-no-common.patch?id=a6e8c6ce860d627a878d8957c81c6991860aed67

Works with:

configure --prefix=/usr CFLAGS=-fcommon

[Pierre Labastie]

menu-cache has "multiple definitions" errors. Pass by adding CFLAGS=-fcommon to configure.

[Pierre Labastie]

dhcp has errors. There is patch by xry111. But since it removes completely the faulty variables, I am not sure about it...

[Bruce Dubbs]

tcsh needs -fcommon

CFLAGS=-fcommon ./configure --prefix=/usr --bindir=/bin

Fixed at revision 23128.

[Bruce Dubbs]

jfsutils needs -fcommon

CFLAGS=-fcommon ./configure

Fixed at revision 23129.

[Douglas R. Reno]

Sawfish needs a patch:

​https://github.com/SawfishWM/sawfish/commit/d4364707a58d335dab4d6ca11b761fd91ef2434d

CFLAGS=-fcommon didn't work for me. Going to start work on this now

[Douglas R. Reno]

Sawfish fixed at r23130

[Pierre Labastie]

menu-cache and lxtask fixed at r23133

[Pierre Labastie]

Replying to pierre.labastie:

dhcp has errors. There is patch by xry111. But since it removes completely the faulty variables, I am not sure about it...

I'm now convinced xry111's patch is the way to go. Can be applied with two seds:

sed -i -r '/u.*(local|remote)_port/d' client/dhclient.c relay/dhcrelay.c
sed -i '/o.*dhcp_type/d' server/mdb.c

[Pierre Labastie]

transcode has "multiple definitions" errors. The symbol audio is defined in several .c files and no .h. Better use CFLAGS...

[Pierre Labastie]

vinagre now...

[Pierre Labastie]

vinagre needs an extern in a .h file, and to define the variable in the corresponding .c file. No confirmation upstream (no bugs, no commit, no PR, as far as I can see). Will use CFLAGS.

[Pierre Labastie]

pipewire (in xry111 patches) has been fixed for version 0.3.4.

[Pierre Labastie]

dhcp, vinagre, transcode fixed at r23134.

[Pierre Labastie]

libreoffice has a "format-overflow" false positive... [Edit] patch: ​https://cgit.freedesktop.org/libreoffice/core/commit/?id=047e8ae5d189f030d565b13f97a4d6a45b00e6be

[Pierre Labastie]

libreoffice: the patch fixes the "format-overflow" issue, but I have another one:

CXX      OutputShape.lo
In file included from /usr/include/boost/format/alt_sstream.hpp:174,
                 from /usr/include/boost/format/internals.hpp:24,
                 from /usr/include/boost/format.hpp:38,
                 from PMDExceptions.h:15,
                 from OutputShape.h:16,
                 from OutputShape.cpp:10:
/usr/include/boost/format/alt_sstream_impl.hpp: In instantiation of ‘boost::io::
basic_altstringbuf<Ch, Tr, Alloc>::int_type boost::io::basic_altstringbuf<Ch, Tr
, Alloc>::overflow(boost::io::basic_altstringbuf<Ch, Tr, Alloc>::int_type) [with Ch = char; Tr = std::char_traits<char>; Alloc = std::allocator<char>; boost::io::basic_altstringbuf<Ch, Tr, Alloc>::int_type = int]’:
/usr/include/boost/format/alt_sstream_impl.hpp:227:9:   required from here
/usr/include/boost/format/alt_sstream_impl.hpp:261:45: error: no matching function for call to ‘std::allocator<char>::allocate(std::size_t&, char*)’
  261 |                     newptr = alloc_.allocate(new_size, is_allocated_? oldptr : 0);
      |                              ~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from /usr/include/c++/10.1.0/memory:64,
                 from OutputShape.h:13,
                 from OutputShape.cpp:10:
/usr/include/c++/10.1.0/bits/allocator.h:167:7: note: candidate: ‘constexpr _Tp* std::allocator< <template-parameter-1-1> >::allocate(std::size_t) [with _Tp = char; std::size_t = long unsigned int]’
  167 |       allocate(size_t __n)
      |       ^~~~~~~~
/usr/include/c++/10.1.0/bits/allocator.h:167:7: note:   candidate expects 1 argument, 2 provided
make[5]: *** [Makefile:493: OutputShape.lo] Error 1
make[4]: *** [Makefile:381: all-recursive] Error 1
make[3]: *** [Makefile:490: all-recursive] Error 1
make[2]: *** [Makefile:401: all] Error 2
make[1]: *** [/sources/libreoffice/libreoffice-6.4.3.2/external/libpagemaker/ExternalProject_libpagemaker.mk:24: /sources/libreoffice/libreoffice-6.4.3.2/workdir/ExternalProject/libpagemaker/build]

will try disabling system boost

[Pierre Labastie]

libreoffice: there is a patch (to modify bundled boost, actually) ​https://cgit.freedesktop.org/libreoffice/core/commit/?id=677c8de4fa79cd9b278b142013ba7f1c9e4e41c3

there is a PR at boost: ​https://github.com/boostorg/format/pull/70

[Pierre Labastie]

libreoffice builds with a patch to libreoffice, and a patch to boost.

[Pierre Labastie]

rox-filer has "multiple definition" error.

[Pierre Labastie]

Replying to pierre.labastie:

rox-filer has "multiple definition" error.

Seems to be fixed with:

sed -i 's/gboolean/extern &/' src/session.h

[Pierre Labastie]

vsftpd has a conversion error:

sysstr.c:77:40: error: implicit conversion from ‘enum EVSFSysStrOpenMode’ to ‘enum EVSFSysUtilOpenMode’ [-Werror=enum-conversion]
   77 |   enum EVSFSysUtilOpenMode open_mode = kVSFSysStrOpenUnknown;
      |                                        ^~~~~~~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors

Can be fixed with:

sed -i 's/kVSFSysStrOpenUnknown;/0;/' sysstr.c

[Pierre Labastie]

vsftpd, libreoffice, rox-filer fixed at r23151. Boost patched at the same revision. Closing.

[Bruce Dubbs]

Milestone renamed

[Bruce Dubbs]

Milestone renamed