Skip to contentSkip to conversation

berkeleydb: fix static linking issues

PATCH New Waldemar Brodkorb 3 participants 5 messages Dec 4, 2016 – Dec 5, 2016

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 berkeleydb: fix static linking issues

Libtool by default drops -static, so force libtool with -all-static to do static linking.

Fixes:
http://autobuild.buildroot.net/results/f0a96f671644d9f9efcf245b354affdc84f7d7da

---
 package/berkeleydb/berkeleydb.mk | 8 ++++++++
 1 file changed, 8 insertions(+)
package/berkeleydb/berkeleydb.mk +8 -0
index e5c11bd..d09bece 100644
@@ -44,6 +44,10 @@ define BERKELEYDB_CONFIGURE_CMDS
 	)
 endef
 
+ifeq ($(BR2_STATIC_LIBS),y)
+BERKELEYDB_CONF_ENV += LDFLAGS="$(TARGET_LDFLAGS) -all-static"
+endif
+
 ifneq ($(BR2_PACKAGE_BERKELEYDB_TOOLS),y)
 
 define BERKELEYDB_REMOVE_TOOLS
@@ -60,4 +64,8 @@ endef
 
 BERKELEYDB_POST_INSTALL_TARGET_HOOKS += BERKELEYDB_REMOVE_DOCS
 
+define BERKELEYDB_BUILD_CMDS
+	$(TARGET_MAKE_ENV) $(TARGET_CONFIGURE_ARGS) $(MAKE) $(BERKELEYDB_CONF_ENV) -C $(@D)/build_unix
+endef
+
 $(eval $(autotools-package))
Signature · 1 line
2.1.4
#1 lore raw

PATCH 1/1 berkeleydb: fix static linking issues

Hello,

Quoted 2 lines · Waldemar Brodkorb

Libtool by default drops -static, so force libtool with -all-static to do static linking.

This needs a little bit more details, because libtool is used by a large fraction of our packages, we pass -static only, and it just works.

Why do we have a different situation here? I thought this undefined reference to `dl_iterate_phdr' issue was more a toolchain problem. It's affecting a number of different packages, and is apparently a new problem, so a fix at the package level looks a bit odd.

Thanks,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
#2 lore raw

PATCH 1/1 berkeleydb: fix static linking issues

Hi Thomas,
Thomas Petazzoni wrote,

Quoted 11 lines

Hello,

Quoted 2 lines · Waldemar Brodkorb

Libtool by default drops -static, so force libtool with -all-static to do static linking.

This needs a little bit more details, because libtool is used by a large fraction of our packages, we pass -static only, and it just works.

Why do we have a different situation here? I thought this undefined reference to `dl_iterate_phdr' issue was more a toolchain problem.

It is not a toolchain problem, it is more likely a change in toolchain behaviour with uClibc-ng >= 1.0.18.
We can not avoid the circular dependency between libc and libgcc, so every static compiled package must use -static while linking.

Quoted 2 lines

It's affecting a number of different packages, and is apparently a new problem, so a fix at the package level looks a bit odd.

In this case it is similar to autossh a package level problem.

All the libtool packages are working out-of-the box in buildroot with -static, because LIBTOOL_PATCH_HOOK patches libtool to behave like -all-static.

In this case the hook isn't working correctly, because ltmain.sh isn't found:

ESC[3m>>> berkeleydb 5.3.28 Patching libtoolESC[23m
for i in find /home/wbx/buildroot/output/build/berkeleydb-5.3.28/build_unix -name ltmain.sh; do ltmain_version=sed -n '/^[ \t]*VERSION=/{s/^[ \t]*VERSION=//;p;q;}' $i | sed -e 's/\([0-9]*\.[0-9]*\).*/\1/' -e 's/\"//'; ltmain_patchlevel=sed -n '/^[ \t]*VERSION=/{s/^[ \t]*VERSION=//;p;q;}' $i | sed -e 's/\([0-9]*\.[0-9]*\.*\)\([0-9]*\).*/\2/' -e 's/\"//'; if test ${ltmain_version} = '1.5'; then patch -i support/libtool/buildroot-libtool-v1.5.patch ${i}; elif test ${ltmain_version} = "2.2"; then patch -i support/libtool/buildroot-libtool-v2.2.patch ${i}; elif test ${ltmain_version} = "2.4"; then if test ${ltmain_patchlevel:-0} -gt 2; then patch -i support/libtool/buildroot-libtool-v2.4.4.patch
${i}; else patch -i support/libtool/buildroot-libtool-v2.4.patch ${i}; fi fi done
touch

wbx at helium:~/buildroot $ find /home/wbx/buildroot/output/build/berkeleydb-5.3.28/ -name ltmain.sh /home/wbx/buildroot/output/build/berkeleydb-5.3.28/dist/ltmain.sh /home/wbx/buildroot/output/build/berkeleydb-5.3.28/lang/sql/jdbc/ltmain.sh /home/wbx/buildroot/output/build/berkeleydb-5.3.28/lang/sql/odbc/ltmain.sh /home/wbx/buildroot/output/build/berkeleydb-5.3.28/lang/sql/sqlite/ltmain.sh

The hook searches in /home/wbx/buildroot/output/build/berkeleydb-5.3.28/build_unix, but it seems berkeley db provides ltmain.sh in /home/wbx/buildroot/output/build/berkeleydb-5.3.28/dist/.

best regards
Waldemar

#3 lore raw

PATCH 1/1 berkeleydb: fix static linking issues

Quoted 12 lines · Waldemar Brodkorb

Hi Thomas,
Thomas Petazzoni wrote,

Quoted 9 lines

Hello,

Quoted 2 lines · Waldemar Brodkorb

Libtool by default drops -static, so force libtool with -all-static to do static linking.

This needs a little bit more details, because libtool is used by a large fraction of our packages, we pass -static only, and it just works.

Quoted 7 lines
Quoted 2 lines

Why do we have a different situation here? I thought this undefined reference to `dl_iterate_phdr' issue was more a toolchain problem.

It is not a toolchain problem, it is more likely a change in toolchain behaviour with uClibc-ng >= 1.0.18.
We can not avoid the circular dependency between libc and libgcc, so every static compiled package must use -static while linking.

Perhaps add -static in the toolchain wrapper? Which might make it possible to
remove it from the libtool patch as well.

Quoted 35 lines
Quoted 2 lines

It's affecting a number of different packages, and is apparently a new problem, so a fix at the package level looks a bit odd.

In this case it is similar to autossh a package level problem.

All the libtool packages are working out-of-the box in buildroot with

(quoted diff) +0 -2
-static, because LIBTOOL_PATCH_HOOK patches libtool to behave like
-all-static.

In this case the hook isn't working correctly, because ltmain.sh isn't found:

ESC[3m>>> berkeleydb 5.3.28 Patching libtoolESC[23m
for i in find /home/wbx/buildroot/output/build/berkeleydb-5.3.28/build_unix -name ltmain.sh; do ltmain_version=sed -n '/^[ \t]*VERSION=/{s/^[ \t]*VERSION=//;p;q;}' $i | sed -e 's/\([0-9]*\.[0-9]*\).*/\1/' -e 's/\"//'; ltmain_patchlevel=sed -n '/^[ \t]*VERSION=/{s/^[ \t]*VERSION=//;p;q;}' $i | sed -e 's/\([0-9]*\.[0-9]*\.*\)\([0-9]*\).*/\2/' -e 's/\"//'; if test ${ltmain_version} = '1.5'; then patch -i support/libtool/buildroot-libtool-v1.5.patch ${i}; elif test ${ltmain_version} = "2.2"; then patch -i support/libtool/buildroot-libtool-v2.2.patch ${i}; elif test ${ltmain_version} = "2.4"; then if test ${ltmain_patchlevel:-0} -gt 2; then patch -i support/libtool/buildroot-libtool-v2.4.4.patch
${i}; else patch -i support/libtool/buildroot-libtool-v2.4.patch ${i}; fi fi done
touch

wbx at helium:~/buildroot $ find /home/wbx/buildroot/output/build/berkeleydb-5.3.28/ -name ltmain.sh /home/wbx/buildroot/output/build/berkeleydb-5.3.28/dist/ltmain.sh /home/wbx/buildroot/output/build/berkeleydb-5.3.28/lang/sql/jdbc/ltmain.sh /home/wbx/buildroot/output/build/berkeleydb-5.3.28/lang/sql/odbc/ltmain.sh /home/wbx/buildroot/output/build/berkeleydb-5.3.28/lang/sql/sqlite/ltmain.sh

Quoted 4 lines

The hook searches in /home/wbx/buildroot/output/build/berkeleydb-5.3.28/build_unix, but it seems berkeley db provides ltmain.sh in /home/wbx/buildroot/output/build/berkeleydb-5.3.28/dist/.

Perhaps libtool patching should be done in $($(PKG)_DIR) instead of
$($(PKG)_SRCDIR), then we wouldn't have this issue. OTOH, only berkeleydb has this problem. Well, gcc also but that's only a host package, and glibc also but I guess glibc doesn't need any of the patching we do.

Regards,
Arnout

Quoted 8 lines

best regards
Waldemar
_______________________________________________
buildroot mailing list
buildroot at busybox.net
http://lists.busybox.net/mailman/listinfo/buildroot

Signature · 6 lines
Arnout Vandecappelle                          arnout at mind be
Senior Embedded Software Architect            +32-16-286500
Essensium/Mind                                http://www.mind.be
G.Geenslaan 9, 3001 Leuven, Belgium           BE 872 984 063 RPR Leuven
LinkedIn profile: http://www.linkedin.com/in/arnoutvandecappelle
GPG fingerprint:  7493 020B C7E3 8618 8DEC 222C 82EB F404 F9AC 0DDF
#4 lore raw

PATCH 1/1 berkeleydb: fix static linking issues

Hello,

Quoted 3 lines · Arnout Vandecappelle

Perhaps libtool patching should be done in $($(PKG)_DIR) instead of
$($(PKG)_SRCDIR), then we wouldn't have this issue. OTOH, only berkeleydb has this problem.

I like this solution, it's actually the right thing to do. In fact $(PKG)_SRCDIR is badly named, because the "source directory" really is $(PKG)_DIR. $(PKG)_SRCDIR merely points to the sub-directory that contains the configure script.

I've tested changing PKG_SRCDIR to PKG_DIR, and indeed, the libtool patch gets properly applied to the various ltmain.sh in berkeleydb. I have not verified if it fixes the build, though.

Quoted 1 line
Quoted 1 line
Quoted 1 line

berkeleydb 5.3.28 Extracting

gzip -d -c /home/thomas/dl/db-5.3.28.NC.tar.gz | tar --strip-components=1 -C /home/thomas/projets/buildroot/output/build/berkeleydb-5.3.28 -xf -

Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

berkeleydb 5.3.28 Patching
berkeleydb 5.3.28 Updating config.sub and config.guess

for file in config.guess config.sub; do for i in $(find /home/thomas/projets/buildroot/output/build/berkeleydb-5.3.28 -name $file); do cp support/gnuconfig/$file $i; done; done

Quoted 1 line
Quoted 1 line
Quoted 1 line

berkeleydb 5.3.28 Patching libtool

patching file /home/thomas/projets/buildroot/output/build/berkeleydb-5.3.28/dist/ltmain.sh Hunk #1 succeeded at 1416 (offset -1 lines).
Hunk #2 succeeded at 2962 (offset -1 lines).
Hunk #3 succeeded at 5167 (offset -12 lines).
Hunk #4 succeeded at 5177 (offset -12 lines).
Hunk #5 succeeded at 5459 (offset -12 lines).
Hunk #6 succeeded at 5740 (offset -13 lines).
Hunk #7 succeeded at 6682 (offset -14 lines).
Hunk #8 succeeded at 9256 (offset -18 lines).
patching file /home/thomas/projets/buildroot/output/build/berkeleydb-5.3.28/lang/sql/jdbc/ltmain.sh Hunk #1 succeeded at 1048 (offset -166 lines).
Hunk #2 succeeded at 2018 (offset -221 lines).
Hunk #3 succeeded at 4117 (offset -315 lines).
Hunk #4 succeeded at 4127 (offset -315 lines).
Hunk #5 succeeded at 4404 (offset -320 lines).
Hunk #6 succeeded at 4669 (offset -325 lines).
Hunk #7 succeeded at 5411 (offset -327 lines).
Hunk #8 succeeded at 5569 (offset -327 lines).
Hunk #9 succeeded at 8057 (offset -321 lines).
patching file /home/thomas/projets/buildroot/output/build/berkeleydb-5.3.28/lang/sql/odbc/ltmain.sh Hunk #1 succeeded at 283 (offset 57 lines).
Hunk #2 succeeded at 419 (offset 171 lines).
Hunk #3 succeeded at 1471 (offset 227 lines).
Hunk #4 succeeded at 2441 with fuzz 2 (offset 288 lines).
Hunk #5 succeeded at 2587 with fuzz 1 (offset 302 lines).
Hunk #6 succeeded at 5654 (offset 472 lines).
Hunk #7 succeeded at 5979 (offset 475 lines).
patching file /home/thomas/projets/buildroot/output/build/berkeleydb-5.3.28/lang/sql/sqlite/ltmain.sh Hunk #1 succeeded at 1055 (offset -159 lines).
Hunk #2 succeeded at 2025 (offset -214 lines).
Hunk #3 succeeded at 4124 (offset -308 lines).
Hunk #4 succeeded at 4134 (offset -308 lines).
Hunk #5 succeeded at 4411 (offset -313 lines).
Hunk #6 succeeded at 4676 (offset -318 lines).
Hunk #7 succeeded at 5418 (offset -320 lines).
Hunk #8 succeeded at 5576 (offset -320 lines).
Hunk #9 succeeded at 8064 (offset -314 lines).

Waldemar, can you test this solution, and see if it works?

package/pkg-autotools.mk +1 -1
index d1cdb89..7e9eed4 100644
@@ -59,7 +59,7 @@ endef
 #
 define LIBTOOL_PATCH_HOOK
        @$(call MESSAGE,"Patching libtool")
-       $(Q)for i in `find $($(PKG)_SRCDIR) -name ltmain.sh`; do \
+       $(Q)for i in `find $($(PKG)_DIR) -name ltmain.sh`; do \
                ltmain_version=`sed -n '/^[ \t]*VERSION=/{s/^[ \t]*VERSION=//;p;q;}' $$i | \
                sed -e 's/\([0-9]*\.[0-9]*\).*/\1/' -e 's/\"//'`; \
                ltmain_patchlevel=`sed -n '/^[ \t]*VERSION=/{s/^[ \t]*VERSION=//;p;q;}' $$i | \

Thanks!

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com