Rework yum-minimal locale cleanup

It turns out our manual locale cleanup is causing issues (see
I54490b17a7f8b2f977369044fcc6bb49cc13768e).  Upon further
investigation, I think this is a better approach than manually
deleting repos.

glibc on Fedora obeys the %_install_langs macro for reducing the
installed locales (as mentioned in the comments, F24 has moved to
having different packages, but worry about that later).

So our existing clear-out is really only required for CentOS, whose
glibc does not have any way to indicate to build less locales.
However, %_install_langs is still correct there, as it restricts some
of the translation files and other things installed with the %lang
macro in spec files.

This is complicated by us having to set this at glibc-common install
time, which happens with the "yum" from outside the chroot (i.e. on
trusty).  Since this is too old to have flags to pass this, we need to
fiddle with rpmmacros.

I've tested this with fedora-minimal builds and the locales file is
about 2MiB, which is what it was after the cleanups, and the listed
locales are only those we expect (i.e. it appears to be working).

Change-Id: I528a68beeb7b2ceec25ccbec1900670501608158
This commit is contained in:
Ian Wienand 2016-05-31 04:53:05 +00:00
parent 5f4cac3303
commit f5dff9c52a
2 changed files with 77 additions and 51 deletions

View File

@ -23,58 +23,56 @@ set -o pipefail
# effectively: febootstrap-minimize --keep-zoneinfo --keep-rpmdb --keep-services "$target"
# Fedora 24 has a much better way to handle just installing some
# languages; see bug. We should support that at the right time.
if [[ $DISTRO_NAME == "fedora" && $DIB_RELEASE -gt 23 ]]; then
echo "Locale support for Fedora 24 is incomplete"
echo " see: https://bugs.launchpad.net/diskimage-builder/+bug/1571488"
die "Cannot cleanup locales on > Fedora 23"
# This is only required on CentOS ... see notes in
# root.d/08-yum-chroot about %_install_langs
if [[ $DISTRO_NAME != "fedora" ]]; then
# Stripping *all* locales is a bad idea. For now, we take the
# suggestion in [1] for reducing this
# [1] https://bugzilla.redhat.com/show_bug.cgi?id=156477
if [ ! -f /usr/lib/locale/locale-archive ]; then
die "locale-archive not found? Can not do cleanup."
fi
# now the archive has everything in it, and is about 100MiB. Strip it
# to just en_US (basically, this is the locale we support if you ssh
# in, other than POSIX)
localedef --delete-from-archive \
$(localedef --list-archive | grep -v '^en_US' | xargs)
# This removes the locales from the archive index but doesn't rebuild
# the file, so it is still the same size (maybe it is sparse?
# presumably as it's mmapped you don't want to fiddle with the offsets
# of locales in the archive on a live system. We are not live).
# build-locale-archive is a tool provided by the RH packaging of
# glibc. Documentation is scarce, but it takes the pre-built locales
# in the tmpl file and creates an archive. It seems originally the
# installer would set some flags to rpm to tell the package what
# languages to pick out of the template, but along the way, this was
# reverted to install them (size considered less important than locale
# support).
# We hack this by moving the locale-archive we've removed the extra
# locales from to the template, then re-run the build.
mv /usr/lib/locale/locale-archive /usr/lib/locale/locale-archive.tmpl
# rebuild archive from template
/usr/sbin/build-locale-archive
# leave empty template behind as package does. I think this stops
# upgrades redoing things
echo > /usr/lib/locale/locale-archive.tmpl
# remove the unnecessary source locale files and iconv files
pushd /usr/share/locale
find . ! -name 'locale.alias' -delete
popd
rm -rf {lib.lib64}/gconv
fi
# Stripping *all* locales is a bad idea. For now, we take the
# suggestion in [1] for reducing this
# [1] https://bugzilla.redhat.com/show_bug.cgi?id=156477
if [ ! -f /usr/lib/locale/locale-archive ]; then
die "locale-archive not found? Can not do cleanup."
fi
# now the archive has everything in it, and is about 100MiB. Strip it
# to just en_US (basically, this is the locale we support if you ssh
# in, other than POSIX)
localedef --delete-from-archive \
$(localedef --list-archive | grep -v '^en_US' | xargs)
# This removes the locales from the archive index but doesn't rebuild
# the file, so it is still the same size (maybe it is sparse?
# presumably as it's mmapped you don't want to fiddle with the offsets
# of locales in the archive on a live system. We are not live).
# build-locale-archive is a tool provided by the RH packaging of
# glibc. Documentation is scarce, but it takes the pre-built locales
# in the tmpl file and creates an archive. It seems originally the
# installer would set some flags to rpm to tell the package what
# languages to pick out of the template, but along the way, this was
# reverted to install them (size considered less important than locale
# support).
# We hack this by moving the locale-archive we've removed the extra
# locales from to the template, then re-run the build.
mv /usr/lib/locale/locale-archive /usr/lib/locale/locale-archive.tmpl
# rebuild archive from template
/usr/sbin/build-locale-archive
# leave empty template behind as package does. I think this stops
# upgrades redoing things
echo > /usr/lib/locale/locale-archive.tmpl
# remove the unnecessary source locale files and iconv files
pushd /usr/share/locale
find . ! -name 'locale.alias' -delete
popd
rm -rf {lib.lib64}/gconv
# docs
rm -rf /usr/share/{doc,info,gnome/help}
# cracklib

View File

@ -128,6 +128,29 @@ function _install_pkg_manager {
flock -w 1200 9 || die "Can not lock .rpmmacros"
echo "%_dbpath /var/lib/rpm" >> $HOME/.rpmmacros
# Fedora 24 has a much better way to handle just installing some
# languages; see bug. We should support that at the right time.
if [[ $DISTRO_NAME == "fedora" && $DIB_RELEASE -gt 23 ]]; then
echo "Locale support for Fedora 24 is incomplete"
echo " see: https://bugs.launchpad.net/diskimage-builder/+bug/1571488"
die "Cannot cleanup locales on > Fedora 23"
fi
# _install_langs is a rpm macro that limits the translation
# files, etc installed by packages. For Fedora 23 [1], the
# glibc-common package will obey this to only install the
# listed locales, keeping things much smaller (we still have
# to clean up locales manually on centos7). We install just
# en_US because people often ssh in with that locale, but
# leave out everything else. Note that yum has an option to
# set this from the command-line [2], but the yum in trusty we
# are using is too old to have it. So we set it directly in
# the macros file
#
# [1] http://pkgs.fedoraproject.org/cgit/rpms/glibc.git/commit/glibc.spec?h=f23&id=91764bd9ec690d4b8a886c0a3a104aac12d340d2
# [2] http://yum.baseurl.org/gitweb?p=yum.git;a=commit;h=26128173b362474456e8f0642073ecb0322ed031
echo "%_install_langs C:en_US:en_US.UTF-8" >> $HOME/.rpmmacros
sudo -E yum -y \
--setopt=cachedir=$YUM_CACHE/$ARCH/$DIB_RELEASE \
--setopt=reposdir=$TARGET_ROOT/etc/yum.repos.d \
@ -137,7 +160,8 @@ function _install_pkg_manager {
# We modified the base system - make sure we clean up always!
rm $HOME/.rpmmacros.dib.lock
sed -i '$ d' $HOME/.rpmmacros
# sed makes it easy to remove last line, but not last n lines...
sed -i '$ d' $HOME/.rpmmacros; sed -i '$ d' $HOME/.rpmmacros;
if [ $rc != 0 ]; then
die "Initial yum install to chroot failed! Can not continue."
fi
@ -193,6 +217,10 @@ else
sudo -E chroot $TARGET_ROOT rpm --rebuilddb
sudo -E chroot $TARGET_ROOT ${YUM} clean all
# populate the lang reduction macro in the chroot
echo "%_install_langs C:en_US:en_US.UTF-8" | \
sudo tee -a $TARGET_ROOT/etc/rpm/macros.langs > /dev/null
# bootstrap the environment within the chroot; bring in new
# metadata with an update and install some base packages we need.
sudo -E chroot $TARGET_ROOT ${YUM} -y update