Fix multilib install conflicts

From: Kai Kang <[email protected]>

gobject-introspection has changed installation paths of .gir files.

From: Kai Kang <[email protected]>

It fails to create image when install os-release and lib32-os-release both:

Quoted 2 lines

file /etc/os-release conflicts between attempted installs of os-release-1.0-r0.core2_64 and lib32-os-release-1.0-r0.x86

The /etc/os-release is a symlink link to ${libdir}/os-release. Actually
the content of files are identical and make /etc/os-release to be hard
link could fix the issue. But according to os-release (5), symlink link
is necessary for initrd environment such as dracut.

Signed-off-by Kai Kang <[email protected]>

---
 meta/recipes-core/os-release/os-release.bb | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

meta/recipes-core/os-release/os-release.bb +0 -2

index bf4f815a10..4d5487c06d 100644
@@ -1,10 +1,12 @@
-inherit allarch
-
 SUMMARY = "Operating system identification"
 DESCRIPTION = "The /usr/lib/os-release file contains operating

system identification data."
LICENSE = "MIT"
INHIBIT_DEFAULT_DEPS = "1"

From: Kai Kang <[email protected]>

update-alternatives is used to handle install file conflicts of multilib
packages in multilib_script.bbclass. And the default priority is used for the alternatives which causes QA warning:

Quoted 6 lines

WARNING: core-image-minimal-1.0-r0 do_rootfs: [log_check] core-
image-minimal: found 1 warning message in the logfile [log_check] Warn: update-alternatives: libtool has multiple providers with the same priority, please check .../core-image-minimal/1.0- r0/rootfs/usr/lib/opkg/alternatives/libtool for details

The same priorities of alternatives are by design, so eliminate the warning for alternatives from multilib packages.

Signed-off-by Kai Kang <[email protected]>

From: Kai Kang <[email protected]>

It has one line different which is a comment in check_stdint.h between libcheck multilib packages. And then causes install file conflict when install libcheck and lib32-libcheck at same time.

Remove the comment line from check_stdint.h to fix the issue.

Signed-off-by Kai Kang <[email protected]>

---
 meta/recipes-support/libcheck/libcheck_0.12.0.bb | 5 +++++
 1 file changed, 5 insertions(+)

meta/recipes-support/libcheck/libcheck_0.12.0.bb +2 -0

index e646d43968..9969e27054 100644
@@ -16,6 +16,11 @@ inherit autotools pkgconfig texinfo
 
 CACHED_CONFIGUREVARS += "ac_cv_path_AWK_PATH=${bindir}/gawk"
 
+do_install_append_class-target () {
+    # remove the only one line comment which causes multilib install

file conflict

From: Kai Kang <[email protected]>

There is a multilib install file conflict of nss:

Quoted 2 lines

file /etc/pki/nssdb/key4.db conflicts between attempted installs of lib32-nss-3.38-r0.corei7_32 and nss-3.38-r0.corei7_64

Move the creation of blank certificates to pkg_postinst. And check if certificates exist already, don't re-create them.

Signed-off-by Kai Kang <[email protected]>

---
 meta/recipes-support/nss/nss_3.38.bb | 32 +++++++++++++++++---------
--
 1 file changed, 20 insertions(+), 12 deletions(-)

From: Kai Kang <[email protected]>

The first line of config file man_db.conf is the package name. It causes
multilib install file conflict. So remove the first line.

Signed-off-by Kai Kang <[email protected]>

---
 meta/recipes-extended/man-db/man-db_2.8.3.bb | 3 +++
 1 file changed, 3 insertions(+)

meta/recipes-extended/man-db/man-db_2.8.3.bb +0 -0

index 97e5a3d6fb..da6d291011 100644
@@ -26,6 +26,9 @@ do_install() {
 	        install -d ${D}/etc/default/volatiles
 		install -m 0644 ${WORKDIR}/99_mandb

${D}/etc/default/volatiles

meta/recipes-extended/man-db/man-db_2.8.3.bb +2 -0

 	fi
+
+    # the 1st line of man_db.conf is package name which causes

multilib install file conflict
+ sed -i '1d' ${D}${sysconfdir}/man_db.conf
}

Quoted 23 lines · On Sat, 2018-09-29 at 13:43 +0800, [email protected] wrote:

From: Kai Kang <[email protected]>

The first line of config file man_db.conf is the package name. It causes
multilib install file conflict. So remove the first line.

Signed-off-by Kai Kang <[email protected]>

---
  meta/recipes-extended/man-db/man-db_2.8.3.bb | 3 +++
  1 file changed, 3 insertions(+)

meta/recipes-extended/man-db/man-db_2.8.3.bb +0 -0

index 97e5a3d6fb..da6d291011 100644
@@ -26,6 +26,9 @@ do_install() {
  	        install -d ${D}/etc/default/volatiles
  		install -m 0644 ${WORKDIR}/99_mandb

${D}/etc/default/volatiles
fi

meta/recipes-extended/man-db/man-db_2.8.3.bb +2 -0

+
+    # the 1st line of man_db.conf is package name which causes

Quoted 3 lines

multilib install file conflict
+ sed -i '1d' ${D}${sysconfdir}/man_db.conf
}

Please do this with a patch, not a sed expression.

The sed expressions are a maintenance nightmare since we don't know if/when the config file may change, or if/when this expression stops working.

Cheers,

Richard

Quoted 25 lines · On Sat, 2018-09-29 at 13:43 +0800, [email protected] wrote:

From: Kai Kang <[email protected]>

It has one line different which is a comment in check_stdint.h between libcheck multilib packages. And then causes install file conflict when install libcheck and lib32-libcheck at same time.

Remove the comment line from check_stdint.h to fix the issue.

Signed-off-by Kai Kang <[email protected]>

---
  meta/recipes-support/libcheck/libcheck_0.12.0.bb | 5 +++++
  1 file changed, 5 insertions(+)

meta/recipes-support/libcheck/libcheck_0.12.0.bb +2 -0

index e646d43968..9969e27054 100644
@@ -16,6 +16,11 @@ inherit autotools pkgconfig texinfo
 
  CACHED_CONFIGUREVARS += "ac_cv_path_AWK_PATH=${bindir}/gawk"
 
+do_install_append_class-target () {
+    # remove the only one line comment which causes multilib install

file conflict

meta/recipes-support/libcheck/libcheck_0.12.0.bb inherit autotools pkgconfig texinfo · line 22

Quoted 2 lines

meta/recipes-support/libcheck/libcheck_0.12.0.bb +2 -0

+    sed -i '/^\/\*/ d' ${D}${includedir}/check_stdint.h
+}

Please do this with a patch, not a sed expression.

The sed expressions are a maintenance nightmare since we don't know if/when the header may change, or if/when this expression stops working.

Cheers,

Richard

Quoted 34 lines · On Sat, 2018-09-29 at 13:43 +0800, [email protected] wrote:

From: Kai Kang <[email protected]>

It fails to create image when install os-release and lib32-os-release both:

Quoted 2 lines

file /etc/os-release conflicts between attempted installs of os-release-1.0-r0.core2_64 and lib32-os-release-1.0-r0.x86

The /etc/os-release is a symlink link to ${libdir}/os-release. Actually
the content of files are identical and make /etc/os-release to be hard
link could fix the issue. But according to os-release (5), symlink link
is necessary for initrd environment such as dracut.

Signed-off-by Kai Kang <[email protected]>

---
  meta/recipes-core/os-release/os-release.bb | 6 ++++--
  1 file changed, 4 insertions(+), 2 deletions(-)

meta/recipes-core/os-release/os-release.bb +0 -2

index bf4f815a10..4d5487c06d 100644
@@ -1,10 +1,12 @@
-inherit allarch
-
  SUMMARY = "Operating system identification"
  DESCRIPTION = "The /usr/lib/os-release file contains operating

system identification data."
LICENSE = "MIT"
INHIBIT_DEFAULT_DEPS = "1"

meta/recipes-core/os-release/os-release.bb line 8

Quoted 3 lines

meta/recipes-core/os-release/os-release.bb +3 -0

+inherit allarch multilib_script
+
+MULTILIB_SCRIPTS = "${PN}:${sysconfdir}/os-release"

This is exactly the kind of thing I did not want to see MUTLILIB_SCRIPTS used for. Its now being proposed everywhere you have an issue with conflicting files without consideration about what actually makes sense.

What might make more sense here is only to build os-release for the main namespace and not any of the multilibs. It could RPROVIDE if necessary.

Cheers,

Richard

Quoted 3 lines · On Sat, 2018-09-29 at 13:43 +0800, [email protected] wrote:

From: Kai Kang <[email protected]>

gobject-introspection has changed installation paths of .gir files.

gobject-introspection has not changed. *We* have changed it.

This is basically why I really don't want to start messing with upstream code like this as I'm guessing this will be the first of many such problems.

Has anyone discussed the gobject-introspection gir path issue with upstream? If data is arch specific, it shouldn't be going into datadir but into libexec iirc. I'd like to understand the upstream response before we start hacking around this in OE in a way which will generate a ton of patches and incompatibility.

Cheers,

Richard

Quoted 5 lines · On Sat, 2018-09-29 at 13:43 +0800, [email protected] wrote:

From: Kai Kang <[email protected]>

There is a multilib install file conflict of nss:

Quoted 2 lines

file /etc/pki/nssdb/key4.db conflicts between attempted installs of lib32-nss-3.38-r0.corei7_32 and nss-3.38-r0.corei7_64

Quoted 8 lines

Move the creation of blank certificates to pkg_postinst. And check if certificates exist already, don't re-create them.

Signed-off-by Kai Kang <[email protected]>

---
  meta/recipes-support/nss/nss_3.38.bb | 32 +++++++++++++++++---------
--
  1 file changed, 20 insertions(+), 12 deletions(-)

This does raise a question - why aren't the generated files the same? Is there a determinism problem here? This sounds like the image would change with each build and couldn't be reproduced so we have a bigger problem?

Cheers,

Richard

Quoted 19 lines · Richard Purdie

Quoted 11 lines · On Sat, 2018-09-29 at 13:43 +0800, [email protected] wrote:

From: Kai Kang <[email protected]>

There is a multilib install file conflict of nss:

Quoted 3 lines

file /etc/pki/nssdb/key4.db conflicts between attempted installs of
lib32-nss-3.38-r0.corei7_32 and nss-3.38-r0.corei7_64

Move the creation of blank certificates to pkg_postinst. And check if
certificates exist already, don't re-create them.

Quoted 6 lines

Signed-off-by Kai Kang <[email protected]>

---
 meta/recipes-support/nss/nss_3.38.bb | 32 +++++++++++++++++-----
----
--
 1 file changed, 20 insertions(+), 12 deletions(-)

Quoted 7 lines

This does raise a question - why aren't the generated files the same?
Is there a determinism problem here? This sounds like the image would
change with each build and couldn't be reproduced so we have a bigger
problem?

It calls certutil to create blank certificates:

certutil -N -d sql:${D}${sysconfdir}/pki/nssdb/ -f ./empty_password

It should be current time related that create blank certificates in current directory, the key4.db files are different:

kkang@msp-lpggp1:~/buildarea/bar-build

$ touch empty
kkang@msp-lpggp1:~/buildarea/bar-build
$ ./tmp/sysroots-components/x86_64/nss-native/usr/bin/certutil -N -d
sql:./ -f ./empty 
password file contains no data
kkang@msp-lpggp1:~/buildarea/bar-build
$ md5sum *.db
1de1260b3f38349a8633d33acd4e4de7  cert9.db
*7fea1d4dbc99db3ba1b72e30428eb5dc  key4.db*
kkang@msp-lpggp1:~/buildarea/bar-build
$ rm *.db

kkang@msp-lpggp1:~/buildarea/bar-build
$ ./tmp/sysroots-components/x86_64/nss-native/usr/bin/certutil -N -d
sql:./ -f ./empty 
password file contains no data
kkang@msp-lpggp1:~/buildarea/bar-build
$ md5sum *.db
1de1260b3f38349a8633d33acd4e4de7  cert9.db
*9fbbae3e2d65d29f51e357a2dc4650a2  key4.db*

Quoted 45 lines · Kang Kai

Quoted 17 lines · Richard Purdie

Quoted 10 lines · On Sat, 2018-09-29 at 13:43 +0800, [email protected] wrote:

From: Kai Kang <[email protected]>

There is a multilib install file conflict of nss:

Quoted 3 lines

file /etc/pki/nssdb/key4.db conflicts between attempted installs of
lib32-nss-3.38-r0.corei7_32 and nss-3.38-r0.corei7_64

Move the creation of blank certificates to pkg_postinst. And check if
certificates exist already, don't re-create them.

Quoted 6 lines

Signed-off-by Kai Kang <[email protected]>

---
  meta/recipes-support/nss/nss_3.38.bb | 32 +++++++++++++++++-----
----
--
  1 file changed, 20 insertions(+), 12 deletions(-)

Quoted 7 lines

This does raise a question - why aren't the generated files the same?
Is there a determinism problem here? This sounds like the image would
change with each build and couldn't be reproduced so we have a bigger
problem?

It calls certutil to create blank certificates:

certutil -N -d sql:${D}${sysconfdir}/pki/nssdb/ -f ./empty_password

It should be current time related that create blank certificates in current directory, the key4.db files are different:

kkang@msp-lpggp1:~/buildarea/bar-build

$ touch empty
kkang@msp-lpggp1:~/buildarea/bar-build
$ ./tmp/sysroots-components/x86_64/nss-native/usr/bin/certutil -N -d
sql:./ -f ./empty
password file contains no data
kkang@msp-lpggp1:~/buildarea/bar-build
$ md5sum *.db
1de1260b3f38349a8633d33acd4e4de7  cert9.db
*7fea1d4dbc99db3ba1b72e30428eb5dc  key4.db*
kkang@msp-lpggp1:~/buildarea/bar-build
$ rm *.db

Quoted 8 lines

kkang@msp-lpggp1:~/buildarea/bar-build
$ ./tmp/sysroots-components/x86_64/nss-native/usr/bin/certutil -N -d
sql:./ -f ./empty
password file contains no data
kkang@msp-lpggp1:~/buildarea/bar-build
$ md5sum *.db
1de1260b3f38349a8633d33acd4e4de7  cert9.db
*9fbbae3e2d65d29f51e357a2dc4650a2  key4.db*

Can we generate them with a known standard time then? Is there some way to specify that or can we add one?

Cheers,

Richard

Quoted 45 lines · Kang Kai

Quoted 17 lines · Richard Purdie

Quoted 10 lines · On Sat, 2018-09-29 at 13:43 +0800, [email protected] wrote:

From: Kai Kang <[email protected]>

There is a multilib install file conflict of nss:

Quoted 3 lines

file /etc/pki/nssdb/key4.db conflicts between attempted installs of
lib32-nss-3.38-r0.corei7_32 and nss-3.38-r0.corei7_64

Move the creation of blank certificates to pkg_postinst. And check if
certificates exist already, don't re-create them.

Quoted 6 lines

Signed-off-by Kai Kang <[email protected]>

---
  meta/recipes-support/nss/nss_3.38.bb | 32 +++++++++++++++++-----
----
--
  1 file changed, 20 insertions(+), 12 deletions(-)

Quoted 7 lines

This does raise a question - why aren't the generated files the same?
Is there a determinism problem here? This sounds like the image would
change with each build and couldn't be reproduced so we have a bigger
problem?

It calls certutil to create blank certificates:

certutil -N -d sql:${D}${sysconfdir}/pki/nssdb/ -f ./empty_password

It should be current time related that create blank certificates in current directory, the key4.db files are different:

kkang@msp-lpggp1:~/buildarea/bar-build

$ touch empty
kkang@msp-lpggp1:~/buildarea/bar-build
$ ./tmp/sysroots-components/x86_64/nss-native/usr/bin/certutil -N -d
sql:./ -f ./empty
password file contains no data
kkang@msp-lpggp1:~/buildarea/bar-build
$ md5sum *.db
1de1260b3f38349a8633d33acd4e4de7  cert9.db
*7fea1d4dbc99db3ba1b72e30428eb5dc  key4.db*
kkang@msp-lpggp1:~/buildarea/bar-build
$ rm *.db

Quoted 8 lines

kkang@msp-lpggp1:~/buildarea/bar-build
$ ./tmp/sysroots-components/x86_64/nss-native/usr/bin/certutil -N -d
sql:./ -f ./empty
password file contains no data
kkang@msp-lpggp1:~/buildarea/bar-build
$ md5sum *.db
1de1260b3f38349a8633d33acd4e4de7  cert9.db
*9fbbae3e2d65d29f51e357a2dc4650a2  key4.db*

Can we generate them with a known standard time then? Is there some way to specify that or can we add one?

Cheers,

Richard