[RFC] musl update

[dkwo]

Has a musl update been planned?
My reason for asking is I'm trying to build flint and it throws me an error unknown type name 'cpu_set_t', while I noticed Alpine Linux packages it.
Otherwise, does anyone understand what should be patched in our musl to build it?

Thanks.

[abenson]

What architecture are you on? I was able to build it for x86_64-musl.

index: added `flint2-2.7.1_1' (x86_64-musl).

[dkwo]

Thanks for helping.
I'm on the same architecture.
If I create a template, it works for me as well (previously, I was just using cmake on another machine).
Do you think this looks fine?

# Template file for 'flint2'
pkgname=flint2
version=2.7.1
revision=1
wrksrc=flint-${version}
build_style=cmake
configure_args="-DCMAKE_BUILD_TYPE=None -DCMAKE_INSTALL_PREFIX=/usr -DCMAKE_INSTALL_LIBDIR=lib -DBUILD_TESTING=ON"
makedepends="cmake gmp-devel mpfr-devel blas-devel python3"
short_desc="C library in support of computations in number theory"
maintainer="..."
license="LGPL-2.1-or-later"
homepage="https://www.flintlib.org"
distfiles="https://www.flintlib.org/flint-${version}.tar.gz"
checksum=186e2fd9ab67df8a05b122fb018269b382e4babcdb17353c4be1fe364dca481e
distfiles="https://www.flintlib.org/flint-${version}.tar.gz"
checksum=186e2fd9ab67df8a05b122fb018269b382e4babcdb17353c4be1fe364dca481e

[Johnnynator]

You need to.most be some makedpends (cmake, python3) to hostmakedepends.
Furthermore everything in configure_args but BUILD_TESTING can be dropped, it is already defined by the build_style

[Logarithmus]

What about musl update though? It's been half a year since 1.2 was released, yet we are still on 1.1. glibc got updated, but musl didn't. Alpine somehow managed to update, what are the blocking issues for Void?

[sgn]

On 2021-03-21 04:54:52-0700, Artur Sinila ***@***.***> wrote: What about musl update though? It's been half a year since 1.2 was released, yet we are still on 1.1. Alpine somehow managed to update, what are the blocking issues for Void?

Rebuild the whole world for 32-bit architecture. And force update the whole world. For that, we need to add more feature to xbps.

[Logarithmus]

Somehow we've managed to rebuild the whole world for new glibc. How's musl different in this aspect? Also AFAIK there's no i686-musl variant, so i686 is out of scope of the question.

…

On March 21, 2021 3:05:17 PM GMT+03:00, "Đoàn Trần Công Danh" ***@***.***> wrote: On 2021-03-21 04:54:52-0700, Artur Sinila ***@***.***> wrote: > What about musl update though? It's been half a year since 1.2 was > released, yet we are still on 1.1. Alpine somehow managed to update, > what are the blocking issues for Void? Rebuild the whole world for 32-bit architecture. And force update the whole world. For that, we need to add more feature to xbps.

-- Sent from my Android device with K-9 Mail. Please excuse my brevity.

[Duncaen]

Somehow we've managed to rebuild the whole world for new glibc. How's musl different in this aspect? Also AFAIK there's no i686-musl variant, so i686 is out of scope of the question.

No we didn't, glibc update doesn't require a rebuild. armv6l-musl and armv7l-musl require a whole rebuild because of the 64bit time changes.

http://musl.libc.org/time64.html

[ericonr]

The XBPS feature proposal is void-linux/xbps#331 , at least AFAIK.

Fixes for the flint template have been suggested, so I'm closing this.

[dkwo]

@ericonr is it ok to briefly reopen this?
I tried to build musl 1.2.2 by myself on x86_64, but firefox stops working:

(/usr/lib/firefox/firefox:2087): Gtk-WARNING **: 08:51:38.988: Failed to parse /home/dkwo/.config/gtk-3.0/settings.ini: Permission denied
Crash Annotation GraphicsCriticalError: |[C0][GFX1]: no fonts - init: 1 fonts: 80 loader: 0 (t=0.371393) [GFX1]: no fonts - init: 1 fonts: 80 loader: 0
JavaScript error: moz-extension://3be9b177-ed03-40c3-9f1f-25790023c3e7/content/staticNS.js, line 77: ReferenceError: policy is not defined
###!!! [Parent][MessageChannel] Error: (msgtype=0x5D0014,name=PHttpChannel::Msg_DeleteSelf) Channel error: cannot send/recv
Sandbox: seccomp sandbox violation: pid 2128, tid 2128, syscall 16, args 2 21523 140731621541608 0 1 2128.

Do you know what should be changed?
I deleted

srcpkgs/musl/patches/0001_reorder_thread_list_unlink_in_pthread_exit_after_all.patch,
0002_don_t_use_libc_threads_minus_1_as_relaxed_atomic_for.patch,
0003_cut_down_size_of_some_libc_struct_members.patch,
0004_restore_lock_skipping_for_processes_that_return_to_s.patch,
CVE-2020-28928.patch b/srcpkgs/musl/patches/CVE-2020-28928.patch,
aarch64-fregs.patch b/srcpkgs/musl/patches/aarch64-fregs.patch,
ppc-pt_regs.patch b/srcpkgs/musl/patches/ppc-pt_regs.patch,
ppc64-fpregset_t.patch b/srcpkgs/musl/patches/ppc64-fpregset_t.patch,
ppcle.patch b/srcpkgs/musl/patches/ppcle.patch

and edited

/srcpkgs/musl/template
 # Template file for 'musl'
 pkgname=musl
-reverts="1.2.0_1"
-version=1.1.24
-revision=6
+version=1.2.2
+revision=1
 archs="*-musl"
 bootstrap=yes
 build_style=gnu-configure
@@ -11,8 +10,9 @@ short_desc="Musl C library"
 maintainer="Enno Boland <[email protected]>"
 license="MIT"
 homepage="http://www.musl-libc.org/"
-distfiles="http://www.musl-libc.org/releases/musl-${version}.tar.gz"
-checksum=1370c9a812b2cf2a7d92802510cca0058cc37e66a7bedd70051f0a34015022a3
+# distfiles="${homepage}/releases/musl-${version}.tar.gz"
+distfiles="https://git.musl-libc.org/cgit/musl/snapshot/musl-${version}.tar.gz"
+checksum=aec1ca7128576753111a1e66a0404127fff07683f8884f1677cd29aa4074e640
 
 nostrip_files="libc.so"
 shlib_provides="libc.so"

Thanks.

[Duncaen]

Firefoxs sandbox needs to be patched to allow the syscall or the the specific arguments. There is also a whitelist option in about:config as workaround.

[dkwo]

@Duncaen Thanks, do you refer to this by any chance?

[Duncaen]

Maybe that too later, but your error uses syscall number 16, which is ioctl on x86_64, the sandbox will filter ioctl syscalls based on the arguments.

[ericonr]

iirc the issue was to do with O_LARGEFILE now being defined to a proper value, which, unless Firefox is built with such headers, leads to a sandbox violation. I am running 1.2.1 myself with some patches on top because of that.