This is the mail archive of the
glibc-bugs@sourceware.org
mailing list for the glibc project.
glibc-2.11.2-build make install FAILS
- From: "John D. Hendrickson" <johnandsara2 at cox dot net>
- To: glibc-bugs at sourceware dot org
- Date: Tue, 06 Jan 2015 14:44:40 -0500
- Subject: glibc-2.11.2-build make install FAILS
- Authentication-results: sourceware.org; auth=none
- Authentication-results: cox.net; none
- References: <SRFg1p0102qVqVd01RFhL0>
i've had allot of work-arounds for circa 2010 toolchain (most are
further below) and not just with glibc
this is not advertised as needed, likely for a reason.
"make ; hack shell env ; make install"
after a successful build (i have used a similar previous build in chroot
seems ok).
MAKE INSTALL FAILS USING THE SAME ENVIRONMENT THAT MAKE SUCCEEDED ON
script that builds things... ran make already
case "$dir" in
glibc-2.11.2)
unset LDFLAGS
make install
;;
*)
make install
;;
esac
THE ISSUE (BUG REPORT) IS. it should install ok with same environment
it built in. i should have to: "make ; fix env ; make install"
# which then again FAILS trivially, (easy to see why)
test ! -x /usr/src/glibc-2.11.2-build/elf/ldconfig || LC_ALL=C LANGUAGE=C \
/usr/src/glibc-2.11.2-build/elf/ldconfig \
/usr/local2/lib /usr/local2/lib
/usr/src/glibc-2.11.2-build/elf/ldconfig: Can't open configuration file
/usr/local2/etc/ld.so.conf: No such file or directory
make[1]: Leaving directory `/usr/src/glibc-2.11.2'
########################3
pkg = glibc-2.11.2
PKG_CONFIG_PATH = /usr/local1/:/usr/local2/
LANGUAGE = C
LC_ALL = C
LANG = C
LOCALE = C
BMAKE =
BCONF =
LIBVERSION =
CPPFLAGS = -I/usr/local2/include -I/usr/local1/include -I/usr/include
-std=gnu99
CFLAGS = -march=pentium3 -I/usr/local2/include
-I/usr/local1/include -I/usr/include -O2 -std=gnu99
CXXFLAGS = -march=pentium3 -I/usr/local2/include
-I/usr/local1/include -I/usr/include -O2 -std=gnu++98
LDFLAGS = -L/usr/local2/lib -L/usr/local1/lib
opt = --prefix=/usr/local2 --exec-prefix=/usr/local2 --with-tls
--with-cpu=i386 --with-tune=pentium3 --enable-pt_chown
--with-headers=/usr/src/linux/usr/include --disable-sanity-checks
--enable-kernel=2.2.15
/------------------
| other things
\------------------
# no other "make" in the whole toochain runs ldconfig why does this ?
# this is the WORST of all places to run ldconfig without asking
# and the MOST probable pkg where ldconfig is irrelevant due to options
# /etc/ld.so.conf doesn't contain /usr/local2 quite intentionally
# only if --prefix were /lib would i want ldconfig forced to run
# why: i assume you know linux ld is very brain when it comes to
# ld.so.conf: it ALWAYS ignores any attemmpt to omit /lib /usr/lib
# and so LD_LIBRARY_PATH is also only a tool for adding, not setting
# (so, the above cannot be expected work, even if the file asked for
# were found, and were it to appease Makefile then it could break
# system - but can't have it both ways: linux ld)
also: glibc builds libssp though it is NOT enabled (NOT --enable-libssp)
though objdump does not say it's a depends. hope i dont have to
re-re-rebuild for that
also: instructions dont say build will fail if gcc opt std=gnu99 is not
used: it builds and takes a while then quits. do one or the other not
run and quit. i don't like forced -O2 because i want 3x faster
compiling on test runs, but whatever. if you can put a #if in a .h you
should be more able to do so in a configure.am (#if gnu99 or quit)
also: instructions were poor on saying with linux kernel headers and
what version. i had to run that multiple times. when i found what
worked only then understood the instructions were correct !
also: instructions are not clear (my own fault of being green) that cpu
must be i386 not i686 and i686 is detected that way but i386 is not
detected for i686. i actually changed source tree to so i686 could be
both cpu and tune (so i was sure i686 was used), only to find
sync_fetch_and_add_4 was still missing (because it was not in gcc-4.4.5)
and reverted. (it worked and was simpler though)
also: many autoconf projects say "it's wise to use ../pkg-build/../pkg
configure. however others do not force it. i dont like it - if i'm
repairing code i now have to be in two trees at the same time
hack: if glibc, runcheck=1
unlike any other: Makefile has not "^check:" entry (nor "make help")
this means a casual on-looker or idiot automaton can assume that "make
check" doesn't exist and would be unsafe to try
hack: rm -r * || true
there also is no "make clean" or "make diskclean"
there also is no "make uninstall"
it does seem (ie compared to Xfree X11R6: dl,make,install,use < 1 hr.
time) that today's toolchain is ever more allowing non-professionals
wide access to do anything they please: and that this is causing the
work-arounds to build the tool chain to be IN THE HUNDREDS to find solve
and apply, to me. i'm unsure i'd have to check an old release but i
dont think the old bsd "make World" or what (10,20,30 yrs ago) was so
"flakey"
---------------------------------
of course i'm very haapy to build the toolchain which has so many fixes
and good things it did not in the past