Closed pullmoll closed 2 years ago
Note that some of the packages fail only for i686-musl
which tells that there's something wrong with either the triplet i686-linux-musl
vs. i686-pc-linux-gnu
detection, or the compiler / linker for some reason does not link against libssp_nonshared.a
when it should.
Perhaps the triplet should be i686-pc-linux-musl
... i686 is a special case in this belong.
In some cases omitting -nostdlib
from LDFLAGS
solved the issues, in other cases explicitly listing LDFLAGS="-lssp_nonshared"
did help. In any case a package failing for i686-musl
does not necessarily mean it's a package problem.
wine
isn't on this list because the template doesn't try to build for i686-musl
, but it also fails. There are people on alpine trying to get it to build there, so I plan on trying to get a hold of their fixes.
@pullmoll Would you mind try building kexi
again. It's built fine over here.
@sgn will take some time since I'm restoring my repository files after accidental deletion :-/
I fixed some non-complicated packages. I'm not able to revisit this issue for a while.
New failure:
libclc-git
fails with llvm11It used to build before the update. Now I see this error:
LLVM-CC amdgcn--amdhsa/lib/math/fmax.cl.bc
LLVM-CC amdgcn--amdhsa/lib/math/fmin.cl.bc
clang-11: error: cannot find ROCm device library. Provide its path via --rocm-path or --rocm-device-lib-path
, or pass -nogpulib to build without ROCm device library.
clang-11: error: cannot find ROCm device library. Provide its path via --rocm-path or --rocm-device-lib-path
, or pass -nogpulib to build without ROCm device library.
LLVM-CC amdgcn--amdhsa/lib/math/ldexp.cl.bc
make: *** [Makefile:7927: amdgcn--amdhsa/lib/workitem/get_local_size.cl.bc] Error 1
make: *** Waiting for unfinished jobs....
make: *** [Makefile:7922: amdgcn--amdhsa/lib/workitem/get_global_size.cl.bc] Error 1
2 warnings generated.
Updating to latest git revision 9aa6f350a6ce0f2cfc7e489495af8899ca74e079 of 2019-10-27 doesn't change this.
Not sure what to do here. It's blocking my try to do (yet) another bootstrap as mesa
depends on it, and the chain back to cairo
, gobject-introspection
, libsecret
, git
, dtc
, qemu-user-static
, libgpg-error
, libgcrypt
, libmicrohttpd
, elfutils
, gdb
, atf
, xbps
, base-chroot
when trying to do a cross build for base-chroot
for i686
.
We probably should use the libclc from the llvm-project repo ( https://github.com/llvm/llvm-project/releases/download/llvmorg-11.0.0/libclc-11.0.0.src.tar.xz ). The libclc repo is seen as outdated and deprecated.
I also think we should do that. I'm a little busy with other stuff ATM though. Perhaps I can take a look tomorrow.
libgcrypt
fails to cross compile./usr/lib/gcc/x86_64-linux-musl/9.3.0/../../../../x86_64-linux-musl/bin/ld: ./.libs/libgcrypt.so: undefined reference to `_gcry_mpih_addmul_1'
/usr/lib/gcc/x86_64-linux-musl/9.3.0/../../../../x86_64-linux-musl/bin/ld: ./.libs/libgcrypt.so: undefined reference to `_gcry_mpih_rshift'
/usr/lib/gcc/x86_64-linux-musl/9.3.0/../../../../x86_64-linux-musl/bin/ld: ./.libs/libgcrypt.so: undefined reference to `_gcry_mpih_add_n'
/usr/lib/gcc/x86_64-linux-musl/9.3.0/../../../../x86_64-linux-musl/bin/ld: ./.libs/libgcrypt.so: undefined reference to `_gcry_mpih_mul_1'
/usr/lib/gcc/x86_64-linux-musl/9.3.0/../../../../x86_64-linux-musl/bin/ld: ./.libs/libgcrypt.so: undefined reference to `_gcry_mpih_submul_1'
/usr/lib/gcc/x86_64-linux-musl/9.3.0/../../../../x86_64-linux-musl/bin/ld: ./.libs/libgcrypt.so: undefined reference to `_gcry_mpih_lshift'
/usr/lib/gcc/x86_64-linux-musl/9.3.0/../../../../x86_64-linux-musl/bin/ld: ./.libs/libgcrypt.so: undefined reference to `_gcry_mpih_sub_n'
collect2: error: ld returned 1 exit status
Solution: add configure_args+=" ac_cv_sys_symbol_underscore=no"
for $CROSS_BUILD
, not just for $XBPS_TARGET_ARCH
in ppc|ppc-musl
.
RE libclc-git: arch went this route https://github.com/archlinux/svntogit-packages/commit/a5b389c616fc96fb7f32396adaaa2612fd14c19a
tried fixing k3s but had some issues. will take some time, especially with the holidays.
@fosslinux I have a PR building libclc-11.0.0 which seems to be good and waits for review.
FYI for electron7: fsck it. I tried to patch it (bison3.7 and perfetto internal destructors problems) and got to some point. Now another update in the system makes it fail to build much earlier again. I'm fed and done with that one.
@sgn: kexi still fails here somewhere related to KReportGroupTracker
, probably coming from kreport-devel
. The difference may be my build of kreport
is rather new while the official build was last updated on 2020-07-08, i.e. before qt5
update to 5.15.1
.
Just a heads up on this issue: I moved the universe(s) rebuilds w/ musl-1.2.1 (and gcc-10.2.0 etc. pp.) to my main, experimental, repository at https://repo.voidlinux.de/ where the failing packages per arch are linked. The current numbers (as of 2021-01-13 20:15:02 CET) are: Architecture | packages built | bad |
---|---|---|
x86_64 | 11721 | 24 |
x86_64-musl | 11649 | 23 |
i686 | 11675 | 30 |
i686-musl | 11417 | 58 |
aarch64 | 10695 | 163 |
aarch64-musl | 10660 | 171 |
armv7l | 10565 | 240 |
armv7l-musl | 10653 | 165 |
armv6l | 10191 | 228 |
armv6l-musl | 10229 | 188 |
armv5tel | 10123 | 243 |
armv5tel-musl | 10085 | 250 |
mips-musl | 8858 | 309 |
mipshf-musl | 8941 | 225 |
mipsel-musl | 8889 | 284 |
mipselhf-musl | 8964 | 223 |
ppc | 9682 | 270 |
ppc-musl | 9242 | 568 |
ppcle | 6117 | 246 |
ppcle-musl | 6140 | 211 |
ppc64 | 10200 | 191 |
ppc64-musl | 10204 | 183 |
ppc64le | 10607 | 199 |
ppc64le-musl | 10590 | 198 |
Of course this is just a snapshot as I'm constantly trying to fix problems and then build pkgs for the 24 targets which sometimes takes its time to propagate to dependent pkgs.
I of course appreciate anyone looking into the bad logfiles and fixing the underlying issue(s). Some are simple to fix, some are complex, and some cannot be fixed at all.
I just tested some fresh built live ISOs and I have no network to external IP addresses running qemu-system-i386 and one of the Qt5 based desktops lxqt or kde. Booting the cinnamon or lxde live ISO in qemu-system-i386 I can connect to e.g. wikipedia with icecat. Also kde and falkon in qemu-system-x86_64 works.
So it seems to be some issue caused by or with Qt5 networking and i686 (both glibc and musl). No idea how that could prevent ping to external IP addresses in a terminal, though. I can ping 10.0.2.2 but nothing outside it seems.
The live ISOs will be up in about 2 hours or so from now.
Some errors when running xbulk
for armv5tel
I see here:
=> ERROR: tvbrowser-4.2.1_2: dependency 'java-runtime' does not exist!
=> ERROR: nvidia390-390.141_1: dependency 'libOpenCL' does not exist!
=> ERROR: nvidia-460.39_1: dependency 'libOpenCL' does not exist!
=> ERROR: libXpresent-1.0.0_2: dependency 'xproto' does not exist!
nvidia and nvidia-390 should be fixed, libOpenCL
isn't a package - #28503.
tvbrowser should move to the virtual - #28504.
libXpresent depends on xproto which is a non-existent package. Should be simple to fix, but I'm not sure if it should be replaced with something - #28505.
quota
: your musl is newer than our version, in that version, reallocarray
is available, we need to delete those lines:
https://github.com/void-linux/void-packages/blob/a0e654bf8237cc41dac600b3d1c35d0992b11793/srcpkgs/quota/template#L17-L20
sysprof
: delete https://github.com/void-linux/void-packages/blob/master/srcpkgs/sysprof/patches/musl.patch
@sgn Changed in my branch. Results after the next round of bulk builds.
I pushed the compatibility fix for both musl 1.1.24 and 1.2.2
For those build failure on cross in your repo, we need to remove implicit dependency on gobject-introspection
when its dependencies have gir
build options on.
sysprof
patch fixed to not disable the TEMP_FAILURE_RETRY
macro definition when HAVE_REALLOCARRAY
is defined.
gperftools for musl: https://github.com/gperftools/gperftools/pull/1250, I'll pull that patch in after discuss with upstream.
It seems ghc
for i686-musl
has issues with whatever setFileTimes()
maps to - utimes(2) perhaps because building stack
fails.
And i686
has problems with fdLock
for stack
packages.
python3-PyICU
: maybe related to #28531
xf86-video-mga
and xf86-video-cirrus
: VBE was disabled for all non-x86, @q66 said it's better to patch it out. VBE looks too old to me, too.
@sgn I have patches for them in my branch which disable vbe and int10 stuff for non x86. Not sure whether I should push the to master.
I think q66 has pushed the fix out.
Yup. I'll add the broken=…
for mips*
later.
Issues become stale 90 days after last activity and are closed 14 days after that. If this issue is still relevant bump it or assign it.
armv{6,7}l{,-musl}
. Date 21-03-03 10:27 Package name