[FFe] Please accept glibc 2.32-0ubuntu1 to Groovy

[Balint Reczey]

The major release upgrade is part of the toolchain updates planned for the 20.10 development cycle and it also contains numerous bug fixes:

Changes:
 glibc (2.32-0ubuntu1) groovy; urgency=medium
 .
   * Merge from Debian unstable
   * Refresh patches
   * debian/watch: Use HTTPS and download xz-compressed tarball
   * debian/watch: Use upstream's signing key to verify the tarball
   * Don't build and ship libnsl.a and Sun RPC static library and header files
   * XFAIL stdlib/tst-getrandom (LP: #1891403)
   * debian/symbols.wildcards: Update versions
   * Make libc-dev depend on rpcsvc-proto and libtirpc-dev.
     They replace development files for the Sun RPC library removed in glibc 32.
   * Make libc-dev depend on libnsl-dev.
     It replaces the development files for the new libnsl library replacing the
     one shipped by glibc.
   * Don't check symbols of libnss_nis.so and libnss_nisplus.so libraries.
     They are not shipped in glibc 2.32
   * Depend on libnss-nis and libnss-nisplus shipping the libraries dropped
     in 2.32
   * debian/patches/git-updates.diff: update from upstream stable branch:
     - x86-64: Fix FMA4 detection in ifunc
 .
 glibc (2.31-3) unstable; urgency=medium
 .
   [ Aurelien Jarno ]
   * debian/control.in/libc: add a Breaks: against openarena (<< 0.8.8+dfsg-4~)
     due to bug#966150.
   * debian/control.in/libc: add a Breaks: against ioquake3
     (<< 1.36+u20200211.f2c61c1~dfsg-2~) as previous versions are not correctly
     linked with libm.
   * debian/patches/git-updates.diff: update from upstream stable branch:
     - Fix an infinite loop in the iconv program (CVE-2016-10228). Closes:
       #856503.
     - debian/patches/any/submitted-selinux-deprecations.diff: upstreamed.
     - debian/patches/x32/submitted-fix-nptl-setgroups-x32.diff: upstreamed.
   * debian/rules.d/build.mk: install <finclude/math-vector-fortran.h> in the
     multiarch path. Closes: #962457.
 .
   [ Samuel Thibault ]
   * debian/libc0.3.symbols.hurd-i386: Fix dependency of __errno_location and
     __h_errno_location symbols in libpthread.
   * patches/hurd-i386/unsubmitted-sbrk-_end.diff: Fix _end symbol appearance
     by reworking sbrk.
   * patches/hurd-i386/unsubmitted-sched_sets.diff: Add sched_set/getscheduler.
   * patches/hurd-i386/git-pthread_atfork-hidden.diff: Hide pthread_atfork
     symbols imported to applications, to avoid leaking them.
 .
 glibc (2.31-2ubuntu1) groovy; urgency=medium
 .
   [ Michael Hudson-Doyle ]
   * Mark tst-getpw as XFAIL on arm64. (LP: #1869364)
 .
   [ Balint Reczey ]
   * Merge from Debian unstable
     - debian/patches/any/submitted-selinux-deprecations.diff: proposed patch to
       ignore the selinux deprecations introduced in libselinux (>= 3.1), fixing
       an FTBFS. (LP: #1887919)
     Remaining changes:
     - Enable libc6{,-dev}-armel on armhf and libc6{-dev}-armhf on armel.
     - Heavily mangle the way we do service restarting on major upgrades.
     - Build amd64 with -O3, and build ppc64 variants (both 64-bit and 32-bit)
       with -O3 -fno-tree-vectorize.
     - Build generic i386 flavour with -mno-tls-direct-seg-refs for Xen.
     - Drop the libc6-xen flavour, as the above change covers Xen's needs.
     - Enable systemtap support, which is currently disabled in Debian.
     - Don't build libc-l10n, its contents get stripped for language-packs.
     - Drop libc-bin manpages Recommends to Suggests to keep it in standard.
     - Revert dropping the ldconfig wrapper, as we still have a lot of
       packages that don't ship a trigger but instead call in postinst.
     - Use DH_COMPAT=8 for dh_strip to fix debug sections for valgrind.
     - Mangle locales package to support Ubuntu language packs seamlessly.
     - Relax some expected test failures for our infrastructure's quirks.
     - Let nptl/tst-stack4 fail, as it's been racing on several architectures.
     - Copy the fully conditionalized x86 variant for math-vector-fortran.h
       to /usr/include/finclude. On all architectures.
     - Backport x86 CET patches from the trunk.
     - Ship arm64 variant with LSE support in libc6-lse
     - debian/sysdeps/{amd64/i386/x32}.mk: Enable Intel Control Flow
     Dropped changes:
     - Build i386 variants as -march=i686
     - debian/patches/git-updates.diff: update from upstream stable branch.
   * debian/gbp.conf: Add initial configuration
   * debian/debhelper.in/libc.preinst: Fix setting LDCONFIG_NOTRIGGER
     (LP: #1889190)
   * Build-depend again on g++-10-multilib on armel and armhf which were lost in the merge
     (LP: #1889069)
   * debian/control.in/main: Add Vcs-* pointing to Ubuntu packaging repository
   * Don't handle false positive stringop-overflow warnings as errors on ppc64el
   * Fall back to calling nanosleep syscall when __clock_nanosleep returns EINVAL
     (LP: #1871240)
   * debian/testsuite-xfail-debian.mk: XFAIL stdlib/tst-strtod-round on riscv64
 .
 glibc (2.31-2) unstable; urgency=medium
 .
   [ Aurelien Jarno ]
   * debian/control.in/libc: add a Breaks: against macs (<< 2.2.7.1-3~) due to
     bug #965073.
   * debian/patches/git-updates.diff: update from upstream stable branch:
     - Fix a signed comparison vulnerability in the ARMv7 memcpy and memmove
       functions (CVE-2020-6096). Closes: #961452.
   * debian/control.in/libc: do not limit the openssh-server breaks to 32-bit
     architectures, clock_nanosleep has to be allowed in addition to
     clock_gettime64. Closes: #965932.
   * debian/patches/any/submitted-selinux-deprecations.diff: proposed patch to
     ignore the selinux deprecations introduced in libselinux (>= 3.1), fixing
     an FTBFS. Closes: #965941.
   * debian/patches/x32/submitted-fix-nptl-setgroups-x32.patch: proposed patch
     to fix the setgroups functions in threaded applications on x32 (without
     the testsuite part). Closes: #965091.
 .
   [ Samuel Thibault ]
   * debian/patches/hurd-i386/local-tls-ie-align.diff: Fix TLS IE load with >=
     8 byte alignment.
   * debian/testsuite-xfail-debian.mk: Update backtrace result.
   * debian/patches/hurd-i386/git-fix-longjmp.diff: Fix longjmp from dl loader.
     Notably fixes calling setuid programs from eatmydata.
   * debian/control: Build-depend on gnumach-dev with userland driver RPC interface.
   * debian/libc0.3.symbols.hurd-i386: Add userland driver RPC interface stubs.
   * debian/patches/hurd-i386/local-clock_gettime_MONOTONIC.diff: Make
     clock_nanosleep accept CLOCK_MONOTONIC as well.
 .
 glibc (2.31-1) unstable; urgency=medium
 .
   [ Samuel Thibault ]
   * debian/control: Build-depend on gcc-10 version that defaults to i686 on
     hurd-i386.
   * debian/control: Build-depend on mig-for-host instead of mig.
   * debian/sysdeps/hurd-i386.mk: Drop adding -march=i686.
   * debian/hurd-i386/git-pselect.diff: Fix pselect atomicity.
   * debian/hurd-i386/git-fexecve.diff: Fix fexecve.
   * debian/hurd-i386/git-cond_destroy.diff: Fix cond_destroy synchronization
     with woken threads.
   * debian/hurd-i386/git-holes.diff: Fix detecting support for file holes.
   * debian/hurd-i386/local-clock_gettime_MONOTONIC.diff: Also fix clock_getres
     with CLOCK_MONOTONIC.
   * debian/hurd-i386/git-longjmp-onstack.diff: Fix longjmp-ing from altstack.
   * debian/hurd-i386/git-register-atfork2.diff: Fix register-atfork ordering.
   * debian/hurd-i386/git-intr-msg-unwind.diff: Fix unwinding over interruptible
     RPC.
   * debian/hurd-i386/git-strtol-test.diff: Fix testing strtol errors.
   * debian/testsuite-xfail-debian.mk: Update.
   * debian/debhelper.in/libc-udeb.install.hurd-i386: Add missing
     libmachuser/libhurduser.
   * debian/rules.d/debhelper.mk: Add dh_link pass to libc-udeb binaryinst.
   * debian/debhelper.in/libc-udeb.links.hurd-i386: Add missing ld.so link.
   * debian/rules: Clean links file.
 .
   [ Aurelien Jarno ]
   * debian/patches/git-updates.diff: update from upstream stable branch.
   * debian/patches/any/git-surplus-tls-accounting.diff: backport TLS surplus
     accounting from upstream. Closes: #964141.
   * debian/control.in/main: update breaks on cross-toolchain-base* as they
     will need changes to build with glibc 2.31.

The binary packages are staged in https://bileto.ubuntu.com/#/ticket/4249 because the new new upstream release does not ship libnss_nis*.so libraries which are now provided by the libnss-nis and libnss-nisplus dependencies and bootstrapping was required to break the build-dependency cycle.
The autopkgtests are still running for the ticket.

The new glibc dependencies, libnsl, libnss-nis and libnss-nisplus need to be included in main, I've filed the MIR bug for that: LP: #1894620.

The glibc 2.32 package still ships the libnsl.so.1 library, but a transition with no-change rebuilds is desired to let reverse dependencies pick up the new libnsl.so.2 by building packages against the new libnsl-dev package which is a new dependency of libc6-dev.

Test rebuild of packages that ship binaries linked to libnsl.so.1 was performed in:
https://bileto.ubuntu.com/#/ticket/4247 .

Glibc 2.32 stopped shipping the rpcgen command and the RPC development headers which are now shipped in the rpcsvc-proto package except for rpc/types.h. The new libc-dev package depends on the rpcsvc-proto package and the libtirpc-dev package to allow easy transition for packages supporting TI RPC, but many packages need sourceful updates.

Test rebuild of packages that FTBFS [1] with glibc 2.31 dropping the libnsl and RPC development files was performed in:
https://bileto.ubuntu.com/#/ticket/4232 .

(The test rebuilds were performed against the #4017 ticket's ppa, with packages almost identical to the ones in #4249.)

[1] https://people.debian.org/~aurel32/glibc-rpc-nsl-removal/

[Matthias Klose (doko)]

Tested the binaries from a PPA on amd64, doing a toolchain build.

[Iain Lane (laney)]

Finding it a bit hard to assess the impact of the transitions from looking at the tickets, sorry. Could you please state here what the transitions look like? i.e. I see quite a number of FTBFS there - do you have a plan for them?

I'm guessing without having checked(!) that there would be entanglement with libffi, so we should get that through first.

[Balint Reczey (rbalint)]

@laney Since librpcsvc is statically linked it does not need a transition, just introduces a lot of FTBFS-s (https://launchpad.net/~ci-train-ppa-service/+archive/ubuntu/4232/+packages) from which some are the symbol deprecations from https://sourceware.org/glibc/wiki/Release/2.32 .

The libnsl one is smoother:

https://launchpad.net/%7Eci-train-ppa-service/+archive/ubuntu/4247/+packages?batch=75&memo=75&start=75

Some existing FTBFSs, and the binary packages will still work without the rebuilds because libc still ships libnsl.so.1, thus the transition does not have to be completed.

I'm pretty enanglement with libffi is granted, and getting that through first would be ideal.

[Steve Langasek (vorlon)]

I understand that there will be some impact on buildability of the archive as a result of the transitions, but frankly, regressing buildability is nothing new with new upstream versions of glibc anyway. FFe approved.

[Iain Lane (laney)]

> regressing buildability is nothing new with new upstream versions of glibc anyway

Maybe not, but it is a consideration for doing so after feature freeze.