Message ID | 1395270180-12701-1-git-send-email-arnout@mind.be |
---|---|
State | Accepted |
Headers | show |
Dear Arnout Vandecappelle (Essensium/Mind), On Thu, 20 Mar 2014 00:03:00 +0100, Arnout Vandecappelle (Essensium/Mind) wrote: > In 4268d39, a patch was added to update libtool.m4 for MIPS n64 targets. > However, this caused the configure script to be regenerated during the > build steps, which in turn caused build failures on Fedora 12 hosts. In > ed73d1d, this was fixed by patching the installed libtool.m4 file > instead of the source file. > > However, ed73d1d did not take into account that the target libtool will > also install libtool.m4 to staging, and that that version has > precedence over the one in HOST_DIR. > > Therefore, this patch takes a different approach: the source file is > patched in the usual way, and instead the reconfigure is avoided by > touching the generated files. > > On the target, we also cannot use the AUTORECONF mechanism because of > some mysterious Makefile.inc file. So just use the same approach and > wait for upstream to make a new release. Shouldn't we mark libtool on the target as something deprecated? We have already removed the support for autoconf and automake on the target, so having libtool seems quite useless, no? Thomas
On 20/03/14 00:25, Thomas Petazzoni wrote: > Dear Arnout Vandecappelle (Essensium/Mind), > > On Thu, 20 Mar 2014 00:03:00 +0100, Arnout Vandecappelle > (Essensium/Mind) wrote: >> In 4268d39, a patch was added to update libtool.m4 for MIPS n64 targets. >> However, this caused the configure script to be regenerated during the >> build steps, which in turn caused build failures on Fedora 12 hosts. In >> ed73d1d, this was fixed by patching the installed libtool.m4 file >> instead of the source file. >> >> However, ed73d1d did not take into account that the target libtool will >> also install libtool.m4 to staging, and that that version has >> precedence over the one in HOST_DIR. >> >> Therefore, this patch takes a different approach: the source file is >> patched in the usual way, and instead the reconfigure is avoided by >> touching the generated files. >> >> On the target, we also cannot use the AUTORECONF mechanism because of >> some mysterious Makefile.inc file. So just use the same approach and >> wait for upstream to make a new release. > > Shouldn't we mark libtool on the target as something deprecated? We > have already removed the support for autoconf and automake on the > target, so having libtool seems quite useless, no? Pulseaudio depends on libtool. I guess it actually depends on libltdl, though. But is precisely in libltdl that the problem occurs. Regards, Arnout
Dear Arnout Vandecappelle, On Thu, 20 Mar 2014 07:34:41 +0100, Arnout Vandecappelle wrote: > > Shouldn't we mark libtool on the target as something deprecated? We > > have already removed the support for autoconf and automake on the > > target, so having libtool seems quite useless, no? > > Pulseaudio depends on libtool. I guess it actually depends on libltdl, > though. But is precisely in libltdl that the problem occurs. But isn't the libltdl code normally "copied" into the project when autoreconf is made, so that there should in fact not be any dependency from pulseaudio to the target libtool? Thomas
On 20/03/14 15:54, Thomas Petazzoni wrote: > Dear Arnout Vandecappelle, > > On Thu, 20 Mar 2014 07:34:41 +0100, Arnout Vandecappelle wrote: > >>> Shouldn't we mark libtool on the target as something deprecated? We >>> have already removed the support for autoconf and automake on the >>> target, so having libtool seems quite useless, no? >> >> Pulseaudio depends on libtool. I guess it actually depends on libltdl, >> though. But is precisely in libltdl that the problem occurs. > > But isn't the libltdl code normally "copied" into the project when > autoreconf is made, so that there should in fact not be any dependency > from pulseaudio to the target libtool? I have no idea. Feel free to investigate. If we do deprecate libtool on target, then this patch is no longer needed to fix the autobuild issue. However, perhaps it would still be considered a cleaner way to handle things? Or perhaps another cleaner way is to use touch -t 201110171216 $(@D)/libltdl/m4/libtool.m4 in a post-patch hook. The -t option is POSIX so it should be OK to use. There are actually many more places where we could use that. Regards, Arnout
Dear Arnout Vandecappelle (Essensium/Mind), On Thu, 20 Mar 2014 00:03:00 +0100, Arnout Vandecappelle (Essensium/Mind) wrote: > In 4268d39, a patch was added to update libtool.m4 for MIPS n64 targets. > However, this caused the configure script to be regenerated during the > build steps, which in turn caused build failures on Fedora 12 hosts. In > ed73d1d, this was fixed by patching the installed libtool.m4 file > instead of the source file. > > However, ed73d1d did not take into account that the target libtool will > also install libtool.m4 to staging, and that that version has > precedence over the one in HOST_DIR. > > Therefore, this patch takes a different approach: the source file is > patched in the usual way, and instead the reconfigure is avoided by > touching the generated files. > > On the target, we also cannot use the AUTORECONF mechanism because of > some mysterious Makefile.inc file. So just use the same approach and > wait for upstream to make a new release. > > Fixes: > http://autobuild.buildroot.net/results/b86/b86a83c6549004f226e7255242e54ef4e50c5ec3/ > > Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be> Applied, thanks. Can you let me know if this patch was superseds http://patchwork.ozlabs.org/patch/331090/ ? Thanks! Thomas
On 17/04/14 23:38, Thomas Petazzoni wrote: > Dear Arnout Vandecappelle (Essensium/Mind), > > On Thu, 20 Mar 2014 00:03:00 +0100, Arnout Vandecappelle > (Essensium/Mind) wrote: >> In 4268d39, a patch was added to update libtool.m4 for MIPS n64 targets. >> However, this caused the configure script to be regenerated during the >> build steps, which in turn caused build failures on Fedora 12 hosts. In >> ed73d1d, this was fixed by patching the installed libtool.m4 file >> instead of the source file. >> >> However, ed73d1d did not take into account that the target libtool will >> also install libtool.m4 to staging, and that that version has >> precedence over the one in HOST_DIR. >> >> Therefore, this patch takes a different approach: the source file is >> patched in the usual way, and instead the reconfigure is avoided by >> touching the generated files. >> >> On the target, we also cannot use the AUTORECONF mechanism because of >> some mysterious Makefile.inc file. So just use the same approach and >> wait for upstream to make a new release. >> >> Fixes: >> http://autobuild.buildroot.net/results/b86/b86a83c6549004f226e7255242e54ef4e50c5ec3/ >> >> Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be> > > Applied, thanks. > > Can you let me know if this patch was superseds > http://patchwork.ozlabs.org/patch/331090/ ? Yes it does, I marked it as such. Regards, Arnout
Dear Arnout Vandecappelle, On Fri, 18 Apr 2014 10:12:24 +0200, Arnout Vandecappelle wrote: > > Can you let me know if this patch was superseds > > http://patchwork.ozlabs.org/patch/331090/ ? > > Yes it does, I marked it as such. Thanks! Thomas
diff --git a/package/libtool/libtool-mips64-n64-linking.post-install-patch b/package/libtool/libtool-01-mips64-n64-linking.patch similarity index 100% rename from package/libtool/libtool-mips64-n64-linking.post-install-patch rename to package/libtool/libtool-01-mips64-n64-linking.patch diff --git a/package/libtool/libtool.mk b/package/libtool/libtool.mk index 603f1f1..a039267 100644 --- a/package/libtool/libtool.mk +++ b/package/libtool/libtool.mk @@ -13,18 +13,25 @@ LIBTOOL_LICENSE_FILES = COPYING HOST_LIBTOOL_LIBTOOL_PATCH = NO -# libtool-mips64-n64-linking.post-install-patch is an upstream patch that -# fixes MIPS64 n64 link failures. However, because the patch touches an m4 -# file, applying it triggers a run of autoconf, automake, etc. This sometimes -# leads to build failures due to incompatible system autotools. We cannot -# simply set HOST_LIBTOOL_AUTORECONF = YES because that would create a -# circular dependency on host-libtool. Therefore, just apply the patch -# directly on the installed file. -define HOST_LIBTOOL_FIXUP_LIBTOOL_M4 - patch $(HOST_DIR)/usr/share/aclocal/libtool.m4 < \ - package/libtool/libtool-mips64-n64-linking.post-install-patch +# We have a patch that affects libtool.m4, which triggers an autoreconf +# in the build step. Normally we would set AUTORECONF = YES, but this +# doesn't work for host-libtool because that creates a circular +# dependency. Instead, touch the generated files so autoreconf is not +# triggered in the build step. Note that aclocal.m4 has to be touched +# first since the rest depends on it. Note that we don't need the changes +# in libtool.m4 in our configure script, because we're not actually +# running it on the target. +# For the target, we would normally be able to use AUTORECONF, but it +# fails on libltdl/Makefile.inc. Rather than trying to fix that failure, +# just use the same hack as on the host. +define LIBTOOL_AVOID_AUTORECONF_HOOK + find $(@D) -name aclocal.m4 -exec touch '{}' \; + find $(@D) -name config-h.in -exec touch '{}' \; + find $(@D) -name configure -exec touch '{}' \; + find $(@D) -name Makefile.in -exec touch '{}' \; endef -HOST_LIBTOOL_POST_INSTALL_HOOKS += HOST_LIBTOOL_FIXUP_LIBTOOL_M4 +LIBTOOL_PRE_CONFIGURE_HOOKS += LIBTOOL_AVOID_AUTORECONF_HOOK +HOST_LIBTOOL_PRE_CONFIGURE_HOOKS += LIBTOOL_AVOID_AUTORECONF_HOOK $(eval $(autotools-package)) $(eval $(host-autotools-package))
In 4268d39, a patch was added to update libtool.m4 for MIPS n64 targets. However, this caused the configure script to be regenerated during the build steps, which in turn caused build failures on Fedora 12 hosts. In ed73d1d, this was fixed by patching the installed libtool.m4 file instead of the source file. However, ed73d1d did not take into account that the target libtool will also install libtool.m4 to staging, and that that version has precedence over the one in HOST_DIR. Therefore, this patch takes a different approach: the source file is patched in the usual way, and instead the reconfigure is avoided by touching the generated files. On the target, we also cannot use the AUTORECONF mechanism because of some mysterious Makefile.inc file. So just use the same approach and wait for upstream to make a new release. Fixes: http://autobuild.buildroot.net/results/b86/b86a83c6549004f226e7255242e54ef4e50c5ec3/ Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be> --- ...l-patch => libtool-01-mips64-n64-linking.patch} | 0 package/libtool/libtool.mk | 29 ++++++++++++++-------- 2 files changed, 18 insertions(+), 11 deletions(-) rename package/libtool/{libtool-mips64-n64-linking.post-install-patch => libtool-01-mips64-n64-linking.patch} (100%)