Patchwork [1/1] update-rc.d.bbclass: change RRECOMMENDS to RDEPENDS

login
register
mail settings
Submitter Qi.Chen@windriver.com
Date Sept. 12, 2013, 9:43 a.m.
Message ID <4802511117bc648d9cd10d8262645069d9db1317.1378978986.git.Qi.Chen@windriver.com>
Download mbox | patch
Permalink /patch/57881/
State New
Headers show

Comments

Qi.Chen@windriver.com - Sept. 12, 2013, 9:43 a.m.
From: Chen Qi <Qi.Chen@windriver.com>

If a package inherits update-rc.d.bbclass, it actually needs it to
update the symlinks under /etc/rc?.d/ directories. This is a runtime
dependency. For example, if we try to upgrade the dropbear on target,
its postinst process will need the update-rc.d utility available,
otherwise, the upgrade process will fail.

Besides, if we only recommend update-rc.d, the do_rootfs process will
fail once NO_RRECOMMENDATIONS is enabled. Because in OE, we try to do
the postinsts at rootfs time.

This patch fixes this problem by changing the RRECOMMENDS to RDEPENDS
to let the packages actually rdepend on update-rc.d.

[YOCTO #5169]

Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
---
 meta/classes/update-rc.d.bbclass |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
Qi.Chen@windriver.com - Sept. 12, 2013, 11:18 a.m.
On 09/12/2013 06:25 PM, Martin Jansa wrote:
> On Thu, Sep 12, 2013 at 05:43:55PM +0800, Qi.Chen@windriver.com wrote:
>> From: Chen Qi <Qi.Chen@windriver.com>
>>
>> If a package inherits update-rc.d.bbclass, it actually needs it to
>> update the symlinks under /etc/rc?.d/ directories. This is a runtime
>> dependency. For example, if we try to upgrade the dropbear on target,
>> its postinst process will need the update-rc.d utility available,
>> otherwise, the upgrade process will fail.
>>
>> Besides, if we only recommend update-rc.d, the do_rootfs process will
>> fail once NO_RRECOMMENDATIONS is enabled. Because in OE, we try to do
>> the postinsts at rootfs time.
> IIRC it was changed to allow systemd images to be built completely
> without update-rc.d (with BAD_RECOMMENDATIONS += "update-rc.d") and it's
> IMHO still valid use-case as update-rc.d is inherited even when systemd
> is used.
>

Now that the 'rm_sysvinit_initdir' function has been added to the 
systemd.bbclass, I think we can safely use RDEPENDS here. Because
1) If the /etc/init is removed, update-rc.d command in postinst has no 
real effect.
2) If the /etc/init is not removed, that means symlinks under 
/etc/rc?.d/ should be set up.

So I think there's no negative effect if update-rc.d is installed in 
case of a systemd system.

Also, I don't think adding update-rc.d to BAD_RECOMMENDATIONS is a good 
idea.
For example, we have a systemd based image running and we are installing 
a new package to that image. That package has an init script but no 
corresponding service file. If we only recommend update-rc.d and it's in 
BAD_RECOMMENDATIONS, the install process will fail. This is not what 
user might expect.

So I'd like to suggest that we change it back to RDEPENDS.

Best Regards,
Chen Qi

>> This patch fixes this problem by changing the RRECOMMENDS to RDEPENDS
>> to let the packages actually rdepend on update-rc.d.
>>
>> [YOCTO #5169]
>>
>> Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
>> ---
>>   meta/classes/update-rc.d.bbclass |    2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/meta/classes/update-rc.d.bbclass b/meta/classes/update-rc.d.bbclass
>> index 5588569..3dee13a 100644
>> --- a/meta/classes/update-rc.d.bbclass
>> +++ b/meta/classes/update-rc.d.bbclass
>> @@ -6,7 +6,7 @@ UPDATERCD_virtclass-cross = ""
>>   UPDATERCD_class-native = ""
>>   UPDATERCD_class-nativesdk = ""
>>   
>> -RRECOMMENDS_${UPDATERCPN}_append = " ${UPDATERCD}"
>> +RDEPENDS_${UPDATERCPN}_append = " ${UPDATERCD}"
>>   
>>   INITSCRIPT_PARAMS ?= "defaults"
>>   
>> -- 
>> 1.7.9.5
>>
>> _______________________________________________
>> Openembedded-core mailing list
>> Openembedded-core@lists.openembedded.org
>> http://lists.openembedded.org/mailman/listinfo/openembedded-core
Martin Jansa - Sept. 12, 2013, 12:18 p.m.
On Thu, Sep 12, 2013 at 07:18:12PM +0800, ChenQi wrote:
> On 09/12/2013 06:25 PM, Martin Jansa wrote:
> > On Thu, Sep 12, 2013 at 05:43:55PM +0800, Qi.Chen@windriver.com wrote:
> >> From: Chen Qi <Qi.Chen@windriver.com>
> >>
> >> If a package inherits update-rc.d.bbclass, it actually needs it to
> >> update the symlinks under /etc/rc?.d/ directories. This is a runtime
> >> dependency. For example, if we try to upgrade the dropbear on target,
> >> its postinst process will need the update-rc.d utility available,
> >> otherwise, the upgrade process will fail.
> >>
> >> Besides, if we only recommend update-rc.d, the do_rootfs process will
> >> fail once NO_RRECOMMENDATIONS is enabled. Because in OE, we try to do
> >> the postinsts at rootfs time.
> > IIRC it was changed to allow systemd images to be built completely
> > without update-rc.d (with BAD_RECOMMENDATIONS += "update-rc.d") and it's
> > IMHO still valid use-case as update-rc.d is inherited even when systemd
> > is used.
> >
> 
> Now that the 'rm_sysvinit_initdir' function has been added to the 
> systemd.bbclass, I think we can safely use RDEPENDS here. Because
> 1) If the /etc/init is removed, update-rc.d command in postinst has no 
> real effect.
> 2) If the /etc/init is not removed, that means symlinks under 
> /etc/rc?.d/ should be set up.
> 
> So I think there's no negative effect if update-rc.d is installed in 
> case of a systemd system.
> 
> Also, I don't think adding update-rc.d to BAD_RECOMMENDATIONS is a good 
> idea.
> For example, we have a systemd based image running and we are installing 
> a new package to that image. That package has an init script but no 
> corresponding service file. If we only recommend update-rc.d and it's in 
> BAD_RECOMMENDATIONS, the install process will fail. This is not what 
> user might expect.
> 
> So I'd like to suggest that we change it back to RDEPENDS.

commit 2c403979c03898c679c5a1e1092aec784dbeb77c
Author: Ross Burton <ross.burton@intel.com>
Date:   Tue Mar 5 11:57:00 2013 +0000

update-rcd: drop depends to recommends, check for update-rcd in scripts

This class is still checking if update-rc.d is installed before using
it, so nothing should fail terribly if image developer decides that he
want's image without any sysvinit related stuff.

If you want image which supports both you can add runtime dependency on
update-rc.d somewhere else - it doesn't need to be pulled by every
recipe which inherits update-rc.d bbclass.

> Best Regards,
> Chen Qi
> 
> >> This patch fixes this problem by changing the RRECOMMENDS to RDEPENDS
> >> to let the packages actually rdepend on update-rc.d.
> >>
> >> [YOCTO #5169]
> >>
> >> Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
> >> ---
> >>   meta/classes/update-rc.d.bbclass |    2 +-
> >>   1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/meta/classes/update-rc.d.bbclass b/meta/classes/update-rc.d.bbclass
> >> index 5588569..3dee13a 100644
> >> --- a/meta/classes/update-rc.d.bbclass
> >> +++ b/meta/classes/update-rc.d.bbclass
> >> @@ -6,7 +6,7 @@ UPDATERCD_virtclass-cross = ""
> >>   UPDATERCD_class-native = ""
> >>   UPDATERCD_class-nativesdk = ""
> >>   
> >> -RRECOMMENDS_${UPDATERCPN}_append = " ${UPDATERCD}"
> >> +RDEPENDS_${UPDATERCPN}_append = " ${UPDATERCD}"
> >>   
> >>   INITSCRIPT_PARAMS ?= "defaults"
> >>   
> >> -- 
> >> 1.7.9.5
> >>
> >> _______________________________________________
> >> Openembedded-core mailing list
> >> Openembedded-core@lists.openembedded.org
> >> http://lists.openembedded.org/mailman/listinfo/openembedded-core
>
Qi.Chen@windriver.com - Sept. 13, 2013, 1:53 a.m.
On 09/12/2013 08:18 PM, Martin Jansa wrote:
> On Thu, Sep 12, 2013 at 07:18:12PM +0800, ChenQi wrote:
>> On 09/12/2013 06:25 PM, Martin Jansa wrote:
>>> On Thu, Sep 12, 2013 at 05:43:55PM +0800, Qi.Chen@windriver.com wrote:
>>>> From: Chen Qi <Qi.Chen@windriver.com>
>>>>
>>>> If a package inherits update-rc.d.bbclass, it actually needs it to
>>>> update the symlinks under /etc/rc?.d/ directories. This is a runtime
>>>> dependency. For example, if we try to upgrade the dropbear on target,
>>>> its postinst process will need the update-rc.d utility available,
>>>> otherwise, the upgrade process will fail.
>>>>
>>>> Besides, if we only recommend update-rc.d, the do_rootfs process will
>>>> fail once NO_RRECOMMENDATIONS is enabled. Because in OE, we try to do
>>>> the postinsts at rootfs time.
>>> IIRC it was changed to allow systemd images to be built completely
>>> without update-rc.d (with BAD_RECOMMENDATIONS += "update-rc.d") and it's
>>> IMHO still valid use-case as update-rc.d is inherited even when systemd
>>> is used.
>>>
>> Now that the 'rm_sysvinit_initdir' function has been added to the
>> systemd.bbclass, I think we can safely use RDEPENDS here. Because
>> 1) If the /etc/init is removed, update-rc.d command in postinst has no
>> real effect.
>> 2) If the /etc/init is not removed, that means symlinks under
>> /etc/rc?.d/ should be set up.
>>
>> So I think there's no negative effect if update-rc.d is installed in
>> case of a systemd system.
>>
>> Also, I don't think adding update-rc.d to BAD_RECOMMENDATIONS is a good
>> idea.
>> For example, we have a systemd based image running and we are installing
>> a new package to that image. That package has an init script but no
>> corresponding service file. If we only recommend update-rc.d and it's in
>> BAD_RECOMMENDATIONS, the install process will fail. This is not what
>> user might expect.
>>
>> So I'd like to suggest that we change it back to RDEPENDS.
> commit 2c403979c03898c679c5a1e1092aec784dbeb77c
> Author: Ross Burton <ross.burton@intel.com>
> Date:   Tue Mar 5 11:57:00 2013 +0000
>
> update-rcd: drop depends to recommends, check for update-rcd in scripts
>
> This class is still checking if update-rc.d is installed before using
> it, so nothing should fail terribly if image developer decides that he
> want's image without any sysvinit related stuff.
>
> If you want image which supports both you can add runtime dependency on
> update-rc.d somewhere else - it doesn't need to be pulled by every
> recipe which inherits update-rc.d bbclass.

Thanks, Martin.

I think I made a mistake.
The real cause is that in image.bbclass, the rootfs_uninstall_unneeded 
function tries to remove the update-rc.d package but it's not installed 
because NO_RECOMMENDATIONS is enabled.
I'm gonna remove update-rc.d from the uninstall list in that function.
Patch will be sent out soon.

Thanks a lot,
Chen Qi


>> Best Regards,
>> Chen Qi
>>
>>>> This patch fixes this problem by changing the RRECOMMENDS to RDEPENDS
>>>> to let the packages actually rdepend on update-rc.d.
>>>>
>>>> [YOCTO #5169]
>>>>
>>>> Signed-off-by: Chen Qi <Qi.Chen@windriver.com>
>>>> ---
>>>>    meta/classes/update-rc.d.bbclass |    2 +-
>>>>    1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/meta/classes/update-rc.d.bbclass b/meta/classes/update-rc.d.bbclass
>>>> index 5588569..3dee13a 100644
>>>> --- a/meta/classes/update-rc.d.bbclass
>>>> +++ b/meta/classes/update-rc.d.bbclass
>>>> @@ -6,7 +6,7 @@ UPDATERCD_virtclass-cross = ""
>>>>    UPDATERCD_class-native = ""
>>>>    UPDATERCD_class-nativesdk = ""
>>>>    
>>>> -RRECOMMENDS_${UPDATERCPN}_append = " ${UPDATERCD}"
>>>> +RDEPENDS_${UPDATERCPN}_append = " ${UPDATERCD}"
>>>>    
>>>>    INITSCRIPT_PARAMS ?= "defaults"
>>>>    
>>>> -- 
>>>> 1.7.9.5
>>>>
>>>> _______________________________________________
>>>> Openembedded-core mailing list
>>>> Openembedded-core@lists.openembedded.org
>>>> http://lists.openembedded.org/mailman/listinfo/openembedded-core

Patch

diff --git a/meta/classes/update-rc.d.bbclass b/meta/classes/update-rc.d.bbclass
index 5588569..3dee13a 100644
--- a/meta/classes/update-rc.d.bbclass
+++ b/meta/classes/update-rc.d.bbclass
@@ -6,7 +6,7 @@  UPDATERCD_virtclass-cross = ""
 UPDATERCD_class-native = ""
 UPDATERCD_class-nativesdk = ""
 
-RRECOMMENDS_${UPDATERCPN}_append = " ${UPDATERCD}"
+RDEPENDS_${UPDATERCPN}_append = " ${UPDATERCD}"
 
 INITSCRIPT_PARAMS ?= "defaults"