glibc-2.3.2-r10.ebuild

glibc-2.3.2-r10.ebuild (text/plain), 18.32 KB, created by Cory Visi (RETIRED) on 2004-05-26 08:25:32 UTC

# Copyright 1999-2004 Gentoo Technologies, Inc.
# Distributed under the terms of the GNU General Public License v2
# $Header: $

IUSE="nls pic build nptl erandom"

inherit eutils flag-o-matic gcc

filter-flags "-fomit-frame-pointer -malign-double"
filter-ldflags "-pie"

# Recently there has been a lot of stability problem in Gentoo-land.  Many
# things can be the cause to this, but I believe that it is due to gcc3
# still having issues with optimizations, or with it not filtering bad
# combinations (protecting the user maybe from himeself) yet.
#
# This can clearly be seen in large builds like glibc, where too aggressive
# CFLAGS cause the tests to fail miserbly.
#
# Quote from Nick Jones <[e-mail]>, who in my opinion
# knows what he is talking about:
#
#   People really shouldn't force code-specific options on... It's a
#   bad idea. The -march options aren't just to look pretty. They enable
#   options that are sensible (and include sse,mmx,3dnow when apropriate).
#
# The next command strips CFLAGS and CXXFLAGS from nearly all flags.  If
# you do not like it, comment it, but do not bugreport if you run into
# problems.
#
# <[e-mail]> (13 Oct 2002)
strip-flags

# Lock glibc at -O2 -- linuxthreads needs it and we want to be conservative here
export CFLAGS="${CFLAGS//-O?} -O2"
export CXXFLAGS="${CFLAGS}"
export LDFLAGS="${LDFLAGS//-Wl,--relax}"

BRANCH_UPDATE="20031115"

# Minimum kernel version for --enable-kernel
export MIN_KV="2.4.1"
# Minimum kernel version for enabling TLS and NPTL ...
# NOTE: do not change this if you do not know what
#       you are doing !
export MIN_NPTL_KV="2.6.0"

MY_PV="${PV/_}"
S="${WORKDIR}/${P%_*}"
DESCRIPTION="GNU libc6 (also called glibc2) C library"
SRC_URI="http://ftp.gnu.org/gnu/glibc/glibc-${MY_PV}.tar.bz2
	ftp://sources.redhat.com/pub/glibc/snapshots/glibc-${MY_PV}.tar.bz2
	http://ftp.gnu.org/gnu/glibc/glibc-linuxthreads-${MY_PV}.tar.bz2
	ftp://sources.redhat.com/pub/glibc/snapshots/glibc-linuxthreads-${MY_PV}.tar.bz2
	mirror://gentoo/${P}-branch-update-${BRANCH_UPDATE}.patch.bz2
	hppa? ( mirror://gentoo/${P}-hppa-patches-p1.tar.bz2 )"
HOMEPAGE="http://www.gnu.org/software/libc/libc.html"

KEYWORDS="x86"

SLOT="2.2"
LICENSE="LGPL-2"

# We need new cleanup attribute support from gcc for NPTL among things ...
DEPEND=">=sys-devel/gcc-3.2.3-r1
	x86? ( nptl? ( >=sys-devel/gcc-3.3.1-r1 ) )
	>=sys-devel/binutils-2.14.90.0.6-r1
	virtual/os-headers
	nls? ( sys-devel/gettext )"
RDEPEND="virtual/os-headers
	sys-apps/baselayout
	nls? ( sys-devel/gettext )"

PDEPEND="ppc? ( >=sys-kernel/linux-headers-2.4.22 )"

PROVIDE="virtual/glibc"


# Try to get a kernel source tree with version equal or greater
# than $1.  We basically just try a few default locations.  The
# version need to be that which KV_to_int() returns ...
get_KHV() {
	local headers=

	[ -z "$1" ] && return 1

	# - First check if linux-headers are installed (or symlink
	#   to current kernel ...)
	# - Ok, do we have access to the current kernel's headers ?
	# - Last option ... maybe its a weird bootstrap with /lib
	#   binded to the chroot ...

	# We do not really support more than 2 arguments ...
	if [ -n "$2" ]
	then
		headers="$2"
	else
		# We try to find the current kernel's headers first,
		# as we would rather build against linux 2.5 headers ...
		headers="/lib/modules/`uname -r`/build/include \
		         ${ROOT}/lib/modules/`uname -r`/build/include \
				 /usr/src/linux/include \
				 ${ROOT}/usr/src/linux/include \
		         ${ROOT}/usr/include"
	fi

	for x in ${headers}
	do
		local header="${x}/linux/version.h"

		if [ -f ${header} ]
		then

			local version="`grep 'LINUX_VERSION_CODE' ${header} | \
				sed -e 's:^.*LINUX_VERSION_CODE[[:space:]]*::'`"

			if [ "${version}" -ge "$1" ]
			then
				echo "${x}"

				return 0
			fi
		fi
	done

	return 1
}

use_nptl() {
	# Enable NPTL support if:
	# - We have 'nptl' in USE
	# - We have linux-2.5 or later kernel (should prob check for 2.4.20 ...)
	if [ -n "`use nptl`" -a "`get_KV`" -ge "`KV_to_int ${MIN_NPTL_KV}`"  ]
	then
		# Enable NPTL support if:
		# - We have 'x86' in USE and:
		#   - a CHOST of "i486-pc-linux-gnu"
		#   - a CHOST of "i586-pc-linux-gnu"
		#   - a CHOST of "i686-pc-linux-gnu"
		# - Or we have 'alpha' in USE
		# - Or we have 'amd64' in USE
		# - Or we have 'mips' in USE
		# - Or we have 'ppc' in USE
		case ${ARCH} in
			"x86")
				if [ "${CHOST/-*}" = "i486" -o \
				     "${CHOST/-*}" = "i586" -o \
					 "${CHOST/-*}" = "i686" ]
				then
					return 0
				fi
				;;
			"alpha"|"amd64"|"mips"|"ppc"|"sparc")
				return 0
				;;
			*)
				return 1
				;;
		esac
	fi

	return 1
}

pkg_setup() {
	# We need gcc 3.2 or later ...
	if [ "`gcc-major-version`" -ne "3" -o "`gcc-minor-version`" -lt "2" ]
	then
		echo
		eerror "As of glibc-2.3, gcc-3.2 or later is needed"
		eerror "for the build to succeed."
		die "GCC too old"
	fi

	echo

	if use_nptl
	then
		# The use_nptl should have already taken care of kernel version,
		# arch and CHOST, so now just check if we can find suitable kernel
		# source tree or headers ....
		einfon "Checking for sufficient version kernel headers ... "
		if ! get_KHV "`KV_to_int ${MIN_NPTL_KV}`" &> /dev/null
		then
			echo "no"
			echo
			eerror "Could not find a kernel source tree or headers with"
			eerror "version ${MIN_NPTL_KV} or later!  Please correct this"
			eerror "and try again."
			die "Insufficient kernel headers present!"
		else
			echo "yes"
		fi
	fi

	if [ "$(KV_to_int $(uname -r))" -gt "`KV_to_int '2.5.68'`" ]
	then
		local KERNEL_HEADERS="$(get_KHV "`KV_to_int ${MIN_NPTL_KV}`")"

		einfon "Checking kernel headers for broken sysctl.h ... "
		if ! gcc -I"${KERNEL_HEADERS}" \
		         -c ${FILESDIR}/test-sysctl_h.c -o ${T}/test1.o &> /dev/null
		then
			echo "yes"
			echo
			eerror "Your version of:"
			echo
			eerror "  ${KERNEL_HEADERS}/linux/sysctl.h"
			echo
			eerror "is broken (from a user space perspective).  Please apply"
			eerror "the following patch:"
			echo
			eerror "*******************************************************"
			cat ${FILESDIR}/fix-sysctl_h.patch
			eerror "*******************************************************"
			die "Broken linux/sysctl.h header included in kernel sources!"
		else
			echo "no"
		fi
	fi

	if use_nptl
	then
		einfon "Checking gcc for __thread support ... "
		if ! gcc -c ${FILESDIR}/test-__thread.c -o ${T}/test2.o &> /dev/null
		then
			echo "no"
			echo
			eerror "Could not find a gcc that supports the __thread directive!"
			eerror "please update to gcc-3.2.2-r1 or later, and try again."
			die "No __thread support in gcc!"
		else
			echo "yes"
		fi

	elif use nptl &> /dev/null
	then
		echo
		# Just tell the user not to expect too much ...
		ewarn "You have \"nptl\" in your USE, but your kernel version or"
		ewarn "architecture does not support it!"
	fi

	echo
}

src_unpack() {

	unpack glibc-${MY_PV}.tar.bz2

	# Extract pre-made man pages.  Otherwise we need perl, which is a no-no.
	mkdir -p ${S}/man; cd ${S}/man
	use_nptl || tar xjf ${FILESDIR}/glibc-manpages-${MY_PV}.tar.bz2

	cd ${S}
	# Extract our threads package ...
	if (! use_nptl) && [ -z "${BRANCH_UPDATE}" ]
	then
		unpack glibc-linuxthreads-${MY_PV}.tar.bz2
	fi

	if [ -n "${BRANCH_UPDATE}" ]
	then
		epatch ${DISTDIR}/${P}-branch-update-${BRANCH_UPDATE}.patch.bz2
	fi

	if use_nptl
	then
		epatch ${FILESDIR}/2.3.2/${P}-redhat-nptl-fixes.patch
	else
		epatch ${FILESDIR}/2.3.2/${P}-redhat-linuxthreads-fixes.patch
	fi

	# To circumvent problems with propolice __guard and
	# __guard_setup__stack_smash_handler
	#
	#  http://www.gentoo.org/proj/en/hardened/etdyn-ssp.xml
	if [ "${ARCH}" != "hppa" -a "${ARCH}" != "hppa64" ]
	then
		cd ${S}
		epatch ${FILESDIR}/2.3.3/${P}-propolice-guard-functions-v3.patch
		cp ${FILESDIR}/2.3.3/ssp.c sysdeps/unix/sysv/linux
	fi

	# patch this regardless of architecture, although it's ssp-related
	epatch ${FILESDIR}/2.3.3/${PN}-2.3.3-frandom-detect.patch

	# With latest versions of glibc, a lot of apps failed on a PaX enabled
	# system with:
	#
	#  cannot enable executable stack as shared object requires: Permission denied
	#
	# This is due to PaX 'exec-protecting' the stack, and ld.so then trying
	# to make the stack executable due to some libraries not containing the
	# PT_GNU_STACK section.  Bug #32960.  <[e-mail]> (12 Nov 2003).
	epatch ${FILESDIR}/2.3.2/${PN}-2.3.2-dl_execstack-PaX-support.patch

	# This next patch fixes a test that will timeout due to ReiserFS' slow handling of sparse files
#	cd ${S}/io; epatch ${FILESDIR}/glibc-2.2.2-test-lfs-timeout.patch

	# This add back glibc 2.2 compadibility.  See bug #8766 and #9586 for more info,
	# and also:
	#
	#  http://lists.debian.org/debian-glibc/2002/debian-glibc-200210/msg00093.html
	#
	# We should think about remoing it in the future after things have settled.
	#
	# Thanks to Jan Gutter <[e-mail]> for reporting it.
	#
	# <[e-mail]> (26 Oct 2002).
	cd ${S}; epatch ${FILESDIR}/2.3.1/${PN}-2.3.1-ctype-compat-v3.patch

	# One more compat issue which breaks sun-jdk-1.3.1.  See bug #8766 for more
	# info, and also:
	#
	#   http://sources.redhat.com/ml/libc-alpha/2002-04/msg00143.html
	#
	# Thanks to Jan Gutter <[e-mail]> for reporting it.
	#
	# <[e-mail]> (30 Oct 2002).
	cd ${S}; epatch ${FILESDIR}/2.3.1/${PN}-2.3.1-libc_wait-compat.patch

	# One more compat issue ... libc_stack_end is missing from ld.so.
	# Got this one from diffing redhat glibc tarball .. would help if
	# they used patches and not modified tarball ...
	#
	# <[e-mail]> (7 Nov 2002).
#	cd ${S}; epatch ${FILESDIR}/2.3.1/${PN}-2.3.1-stack_end-compat.patch

	# The mathinline.h header omits the middle term of a ?: expression.  This
	# is a gcc extension, but since the ISO standard forbids it, it's a
	# GLIBC bug (bug #27142).  See also:
	#
	#   http://bugs.gentoo.org/show_bug.cgi?id=27142
	#
	cd ${S}; epatch ${FILESDIR}/${PV}/${P}-fix-omitted-operand-in-mathinline_h.patch

	# We do not want name_insert() in iconvconfig.c to be defined inside
	# write_output() as it causes issues with PaX.
	cd ${S}; epatch ${FILESDIR}/${PV}/${P}-iconvconfig-name_insert.patch

	# Fix broken reverse resolving for IPv6 addresses, bug #42492.
	cd ${S}; epatch ${FILESDIR}/${PV}/${P}-ipv6-bytestring-fix.patch

	# A few patches only for the MIPS platform.  Descriptions of what they
	# do can be found in the patch headers.
	# <[e-mail]> thx <[e-mail]> (11 Jan 2003)
	# <[e-mail]> remove tst-rndseek-mips & ulps-mips patches
	if [ "${ARCH}" = "mips" ]
	then
		cd ${S}
		epatch ${FILESDIR}/2.3.1/${PN}-2.3.1-fpu-cw-mips.patch
		epatch ${FILESDIR}/2.3.1/${PN}-2.3.1-libgcc-compat-mips.patch
		epatch ${FILESDIR}/2.3.1/${PN}-2.3.1-librt-mips.patch
		epatch ${FILESDIR}/2.3.2/${P}-mips-add-n32-n64-sysdep-cancel.patch
		epatch ${FILESDIR}/2.3.2/${P}-mips-configure-for-n64-symver.patch
		epatch ${FILESDIR}/2.3.2/${P}-mips-pread-linux2.5.patch
	fi

	if [ "${ARCH}" = "alpha" ]
	then
		cd ${S}
		# Fix compatability with compaq compilers by ifdef'ing out some
		# 2.3.2 additions.
		# <[e-mail]> (14 Jun 2003).
		epatch ${FILESDIR}/2.3.2/${P}-decc-compaq.patch

		# Fix compilation with >=gcc-3.2.3 (01 Nov 2003 agriffis)
		epatch ${FILESDIR}/2.3.2/${P}-alpha-pwrite.patch
		epatch ${FILESDIR}/2.3.2/${P}-alpha-crti.patch
	fi

	if [ "${ARCH}" = "amd64" ]
	then
		cd ${S}; epatch ${FILESDIR}/2.3.2/${P}-amd64-nomultilib.patch
	fi

	if [ "${ARCH}" = "ia64" ]
	then
		# The basically problem is glibc doesn't store information about
		# what the kernel interface is so that it can't efficiently set up
		# parameters for system calls.  This patch from H.J. Lu fixes it:
		#
		#   http://sources.redhat.com/ml/libc-alpha/2003-09/msg00165.html
		#
		cd ${S}; epatch ${FILESDIR}/2.3.2/${P}-ia64-LOAD_ARGS-fixup.patch
	fi

	if [ "${ARCH}" = "hppa" ]
	then
		cd ${WORKDIR}
		unpack ${P}-hppa-patches-p1.tar.bz2
		cd ${S}
		EPATCH_EXCLUDE="010* 020* 030* 040* 050* 055*"
		for i in ${EPATCH_EXCLUDE}
		do
			rm -f ${WORKDIR}/${P}-hppa-patches/$i
		done
		for i in ${WORKDIR}/${P}-hppa-patches/*
		do
			einfo Applying `basename $i`...
			patch -p1 < $i
		done
		einfo Applying glibc23-07-hppa-atomicity.dpatch...
		patch -p 1 < ${FILESDIR}/2.3.1/glibc23-07-hppa-atomicity.dpatch
	fi

	# Fix permissions on some of the scripts
	chmod u+x ${S}/scripts/*.sh
}

setup_flags() {
	# -freorder-blocks for all but ia64 s390 s390x
	use ppc || append-flags "-freorder-blocks"

	# Sparc/Sparc64 support
	if use sparc
	then

		# Both sparc and sparc64 can use -fcall-used-g6.  -g7 is bad, though.
		replace-flags "-fcall-used-g7" ""
		append-flags "-fcall-used-g6"

		# Sparc64 Only support...
		if [ "${PROFILE_ARCH}" = "sparc64" ]
		then

			# Get rid of -mcpu options, the CHOST will fix this up
			replace-flags "-mcpu=ultrasparc" ""
			replace-flags "-mcpu=v9" ""

			# Get rid of flags known to fail
			replace-flags "-mvis" ""

			# Setup the CHOST properly to insure "sparcv9"
			# This passes -mcpu=ultrasparc -Wa,-Av9a to the compiler
			[ "${CHOST}" == "sparc-unknown-linux-gnu" ] && \
				export CHOST="sparcv9-unknown-linux-gnu"
		fi
	fi
}

src_compile() {
	local myconf=
	local myconf_nptl=

	setup_flags

	# These should not be set, else the
	# zoneinfo do not always get installed ...
	unset LANGUAGE LANG LC_ALL

	use nls || myconf="${myconf} --disable-nls"

	use erandom || myconf="${myconf} --disable-dev-erandom"

	if use_nptl
	then
		local kernelheaders="$(get_KHV "`KV_to_int ${MIN_NPTL_KV}`")"

		# NTPL and Thread Local Storage support.
		myconf="${myconf} --with-tls --with-__thread \
		                       --enable-add-ons=nptl \
		                       --enable-kernel=${MIN_NPTL_KV} \
		                       --with-headers=${kernelheaders}"
	else
		myconf="${myconf} --without-__thread \
		                  --enable-add-ons=linuxthreads"

		# If we build for the build system we use the kernel headers from the target
		# We also now set it without "build" as well, else it might use the
		# current kernel's headers, which might just fail (the linux-headers
		# package is usually well tested...)
#		( use build || use sparc ) \
#			&& myconf="${myconf} --with-headers=${ROOT}usr/include"
		myconf="${myconf} --with-headers=${ROOT}usr/include"

		# If kernel version and headers in ${ROOT}/usr/include are ok,
		# then enable --enable-kernel=${MIN_KV} ...
		if [ "`get_KV`" -ge "`KV_to_int ${MIN_KV}`" -a \
		     -n "$(get_KHV "`KV_to_int ${MIN_KV}`" "${ROOT}/usr/include")" ]
		then
			myconf="${myconf} --enable-kernel=${MIN_KV}"
		else
			myconf="${myconf} --enable-kernel=2.2.5"
		fi
	fi

	einfo "Configuring GLIBC..."
	rm -rf ${S}/buildhere
	mkdir -p ${S}/buildhere
	cd ${S}/buildhere
	../configure --build=${CHOST} \
		--host=${CHOST} \
		--with-gd=no \
		--without-cvs \
		--disable-profile \
		--prefix=/usr \
		--mandir=/usr/share/man \
		--infodir=/usr/share/info \
		--libexecdir=/usr/lib/misc \
		${myconf} || die

	einfo "Building GLIBC..."
	cd ${S}/buildhere
	make PARALLELMFLAGS="${MAKEOPTS}" || die
#	einfo "Doing GLIBC checks..."
#	make check
}

src_install() {
	local buildtarget="buildhere"

	setup_flags

	# These should not be set, else the
	# zoneinfo do not always get installed ...
	unset LANGUAGE LANG LC_ALL

	einfo "Installing GLIBC..."
	make PARALLELMFLAGS="${MAKEOPTS}" \
		install_root=${D} \
		install -C ${buildtarget} || die

	# If librt.so is a symlink, change it into linker script (Redhat)
	if [ -L "${D}/usr/lib/librt.so" -a "${LIBRT_LINKERSCRIPT}" = "yes" ]
	then
		local LIBRTSO="`cd ${D}/lib; echo librt.so.*`"
		local LIBPTHREADSO="`cd ${D}/lib; echo libpthread.so.*`"

		rm -f ${D}/usr/lib/librt.so
		cat > ${D}/usr/lib/librt.so <<EOF
/* GNU ld script
	librt.so.1 needs libpthread.so.0 to come before libc.so.6*
	in search scope.  */
EOF
		grep "OUTPUT_FORMAT" ${D}/usr/lib/libc.so >> ${D}/usr/lib/librt.so
		echo "GROUP ( /lib/${LIBPTHREADSO} /lib/${LIBRTSO} )" \
			>> ${D}/usr/lib/librt.so

		for x in ${D}/usr/lib/librt.so.[1-9]
		do
			[ -L "${x}" ] && rm -f ${x}
		done
	fi

	if [ -z "`use build`" ]
	then
		einfo "Installing Info pages..."
		make PARALLELMFLAGS="${MAKEOPTS}" \
			install_root=${D} \
			info -C ${buildtarget} || die

		einfo "Installing Locale data..."
		make PARALLELMFLAGS="${MAKEOPTS}" \
			install_root=${D} \
			localedata/install-locales -C ${buildtarget} || die

		# Compatibility hack: this locale has vanished from glibc,
		# but some other programs are still using it.
		keepdir /usr/lib/locale/ru_RU/LC_MESSAGES

		einfo "Installing man pages and docs..."
		# Install linuxthreads man pages
		use_nptl || {
			dodir /usr/share/man/man3
			doman ${S}/man/*.3thr
		}

		# Install nscd config file
		insinto /etc
		doins ${FILESDIR}/nscd.conf

		dodoc BUGS ChangeLog* CONFORMANCE COPYING* FAQ INTERFACE \
			NEWS NOTES PROJECTS README*
	else
		rm -rf ${D}/usr/share ${D}/usr/lib/gconv

		einfo "Installing Timezone data..."
		make PARALLELMFLAGS="${MAKEOPTS}" \
			install_root=${D} \
			timezone/install-others -C ${buildtarget} || die
	fi

	if [ "`use pic`" ]
	then
		find ${S}/${buildtarget}/ -name "soinit.os" -exec cp {} ${D}/lib/soinit.o \;
		find ${S}/${buildtarget}/ -name "sofini.os" -exec cp {} ${D}/lib/sofini.o \;
		find ${S}/${buildtarget}/ -name "*_pic.a" -exec cp {} ${D}/lib \;
		find ${S}/${buildtarget}/ -name "*.map" -exec cp {} ${D}/lib \;
		for i in ${D}/lib/*.map
		do
			mv ${i} ${i%.map}_pic.map
		done
	fi

	# Is this next line actually needed or does the makefile get it right?
	# It previously has 0755 perms which was killing things.
	fperms 4755 /usr/lib/misc/pt_chown

	# Currently libraries in  /usr/lib/gconv do not get loaded if not
	# in search path ...
#	insinto /etc/env.d
#	doins ${FILESDIR}/03glibc

	rm -f ${D}/etc/ld.so.cache

	# Prevent overwriting of the /etc/localtime symlink.  We'll handle the
	# creation of the "factory" symlink in pkg_postinst().
	rm -f ${D}/etc/localtime

	# Some things want this, notably ash.
	dosym /usr/lib/libbsd-compat.a /usr/lib/libbsd.a
}

pkg_postinst() {
	# Correct me if I am wrong here, but my /etc/localtime is a file
	# created by zic ....
	# I am thinking that it should only be recreated if no /etc/localtime
	# exists, or if it is an invalid symlink.
	#
	# For invalid symlink:
	#   -f && -e  will fail
	#   -L will succeed
	#
	if [ ! -e ${ROOT}/etc/localtime ]
	then
		echo "Please remember to set your timezone using the zic command."
		rm -f ${ROOT}/etc/localtime
		ln -s ../usr/share/zoneinfo/Factory ${ROOT}/etc/localtime
	fi

	if [ -x ${ROOT}/usr/sbin/iconvconfig ]
	then
		# Generate fastloading iconv module configuration file.
		${ROOT}/usr/sbin/iconvconfig --prefix=${ROOT}
	fi

	# Reload init ...
	if [ "${ROOT}" = "/" ]
	then
		/sbin/init U &> /dev/null
	fi
}