Images no longer automatically include libstdc++6

Images no longer automatically include libstdc++6

[Steve Sakoman]

Using current top of tree to build some of my images I noticed that
quite a few apps seems to be broken at run time (Firefox, Epiphany,
Gnome-mplayer, and others).

All were complaining about libstdc++6 being missing.  Manually
installing the libstdc++6 package fixed the issue.

Any idea what changed that introduced this issue?

Steve

Re: Images no longer automatically include libstdc++6

[J. L.]

On Mon, Sep 6, 2010 at 3:21 PM, Steve Sakoman <sakoman@gmail.com> wrote:
> Using current top of tree to build some of my images I noticed that
> quite a few apps seems to be broken at run time (Firefox, Epiphany,
> Gnome-mplayer, and others).
>
> All were complaining about libstdc++6 being missing.  Manually
> installing the libstdc++6 package fixed the issue.

I have reported this on the gumstix board as well steve and Frans did
a post about this here on OE and the other one libgcc_s.so.1 that
seems to be missing. Thanks for posting your work around I thought I
had tried that on my machine but will try again. here is Frans post
this am [oe] NOTE: Couldn't find shared library provider, also my
boost threads have the errors listed in them as well.

JL


>
> Any idea what changed that introduced this issue?
>
> Steve
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>

Re: Images no longer automatically include libstdc++6

[Khem Raj]

On Mon, Sep 6, 2010 at 3:21 PM, Steve Sakoman <sakoman@gmail.com> wrote:
> Using current top of tree to build some of my images I noticed that
> quite a few apps seems to be broken at run time (Firefox, Epiphany,
> Gnome-mplayer, and others).
>
> All were complaining about libstdc++6 being missing.  Manually
> installing the libstdc++6 package fixed the issue.
>
> Any idea what changed that introduced this issue?

Yes I have been seeing that and spent not so much time on the real cause
but it seems that the depends get lost when we rename the pkgs in debian way
may be thats not the real cause I dont know yet. but that something to look at
even libstdc++-dev ipks miss to have dependencies on libstdc++

>
> Steve
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>

Re: Images no longer automatically include libstdc++6

[J. L.]

On Mon, Sep 6, 2010 at 10:54 PM, Khem Raj <raj.khem@gmail.com> wrote:
> On Mon, Sep 6, 2010 at 3:21 PM, Steve Sakoman <sakoman@gmail.com> wrote:
>> Using current top of tree to build some of my images I noticed that
>> quite a few apps seems to be broken at run time (Firefox, Epiphany,
>> Gnome-mplayer, and others).
>>
>> All were complaining about libstdc++6 being missing.  Manually
>> installing the libstdc++6 package fixed the issue.
>>
>> Any idea what changed that introduced this issue?
>
> Yes I have been seeing that and spent not so much time on the real cause
> but it seems that the depends get lost when we rename the pkgs in debian way
> may be thats not the real cause I dont know yet. but that something to look at
> even libstdc++-dev ipks miss to have dependencies on libstdc++
>

building hal 0.5.14 is one example that gives off the libgcc that I
mentioned with this one. Though there are many that give both errors
or one or the other.

JL
>>
>> Steve
>>
>> _______________________________________________
>> Openembedded-devel mailing list
>> Openembedded-devel@lists.openembedded.org
>> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>>
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>

Re: Images no longer automatically include libstdc++6

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 07-09-10 07:54, Khem Raj wrote:
> On Mon, Sep 6, 2010 at 3:21 PM, Steve Sakoman <sakoman@gmail.com> wrote:
>> Using current top of tree to build some of my images I noticed that
>> quite a few apps seems to be broken at run time (Firefox, Epiphany,
>> Gnome-mplayer, and others).
>>
>> All were complaining about libstdc++6 being missing.  Manually
>> installing the libstdc++6 package fixed the issue.
>>
>> Any idea what changed that introduced this issue?
> 
> Yes I have been seeing that and spent not so much time on the real cause
> but it seems that the depends get lost when we rename the pkgs in debian way
> may be thats not the real cause I dont know yet. but that something to look at
> even libstdc++-dev ipks miss to have dependencies on libstdc++

it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
create a libstdc6++ package, but it does create a libstdc-dev package.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFMhjWEMkyGM64RGpERAkP3AJ9664771JV6Rap+9F6c9xbeh69fZgCggBVt
XerzMrwvIL7HaPknWKd816Y=
=t1WC
-----END PGP SIGNATURE-----

Re: Images no longer automatically include libstdc++6

[Khem Raj]

On Tue, Sep 7, 2010 at 5:52 AM, Koen Kooi <k.kooi@student.utwente.nl> wrote:
>
> it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
> create a libstdc6++ package, but it does create a libstdc-dev package.
>

I see thats plausible. I think its time to look into separating gcc runtime
from gcc on the lines of poky.

-Khem

Re: Images no longer automatically include libstdc++6

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 07-09-10 18:18, Khem Raj wrote:
> On Tue, Sep 7, 2010 at 5:52 AM, Koen Kooi <k.kooi@student.utwente.nl> wrote:
>>
>> it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
>> create a libstdc6++ package, but it does create a libstdc-dev package.
>>
> 
> I see thats plausible. I think its time to look into separating gcc runtime
> from gcc on the lines of poky.

I put back libgcc and libstdc++ for the time being:

http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=241b8865f20b9e3c3beb334535e6cd8452cbf47d

I have a feeling the others needs to put back as well (e.g. fortran),
but this fixes the immediate problem.

regards,

Koen
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFMhoxwMkyGM64RGpERAihiAJkBqBI+yvfQEwEzwgVKQgGWRvRy1QCfS6kF
oOBWADuHZO69kRkC/Rbcsns=
=OyU4
-----END PGP SIGNATURE-----

Re: Images no longer automatically include libstdc++6

[Frans Meulenbroeks]

2010/9/7 Koen Kooi <k.kooi@student.utwente.nl>:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 07-09-10 18:18, Khem Raj wrote:
>> On Tue, Sep 7, 2010 at 5:52 AM, Koen Kooi <k.kooi@student.utwente.nl> wrote:
>>>
>>> it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
>>> create a libstdc6++ package, but it does create a libstdc-dev package.
>>>
>>
>> I see thats plausible. I think its time to look into separating gcc runtime
>> from gcc on the lines of poky.
>
> I put back libgcc and libstdc++ for the time being:
>
> http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=241b8865f20b9e3c3beb334535e6cd8452cbf47d
>
> I have a feeling the others needs to put back as well (e.g. fortran),
> but this fixes the immediate problem.
>
> regards,
>
> Koen


Thanks for the workaround, but what happened to the rule that changes
to the toolchain should be reviewed first ?????

Frans

Re: Images no longer automatically include libstdc++6

[pieterg]

On Tuesday 07 September 2010 21:03:14 Koen Kooi wrote:
> On 07-09-10 18:18, Khem Raj wrote:
> > On Tue, Sep 7, 2010 at 5:52 AM, Koen Kooi <k.kooi@student.utwente.nl> 
wrote:
> >> it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
> >> create a libstdc6++ package, but it does create a libstdc-dev package.
> >
> > I see thats plausible. I think its time to look into separating gcc
> > runtime from gcc on the lines of poky.
>
> I put back libgcc and libstdc++ for the time being:
>
> http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=241b8865f20
>b9e3c3beb334535e6cd8452cbf47d
>
> I have a feeling the others needs to put back as well (e.g. fortran),
> but this fixes the immediate problem.

Not for me, after a clean build today packages are still missing libstdc++ 
in their deps (and as a result the images are still missing libstdc++).
So there must be more to it.

Rgds, Pieter

Re: Images no longer automatically include libstdc++6

[Philip Balister]

On 09/08/2010 10:09 AM, pieterg wrote:
> On Tuesday 07 September 2010 21:03:14 Koen Kooi wrote:
>> On 07-09-10 18:18, Khem Raj wrote:
>>> On Tue, Sep 7, 2010 at 5:52 AM, Koen Kooi<k.kooi@student.utwente.nl>
> wrote:
>>>> it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
>>>> create a libstdc6++ package, but it does create a libstdc-dev package.
>>>
>>> I see thats plausible. I think its time to look into separating gcc
>>> runtime from gcc on the lines of poky.
>>
>> I put back libgcc and libstdc++ for the time being:
>>
>> http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=241b8865f20
>> b9e3c3beb334535e6cd8452cbf47d
>>
>> I have a feeling the others needs to put back as well (e.g. fortran),
>> but this fixes the immediate problem.
>
> Not for me, after a clean build today packages are still missing libstdc++
> in their deps (and as a result the images are still missing libstdc++).
> So there must be more to it.

Does anyone have any idea what commit caused this breakage? It is 
causing a lot of problems for people.

Philip

Re: Images no longer automatically include libstdc++6

[Dallas Foley]

On 10-09-08 07:33 AM, Philip Balister wrote:
> On 09/08/2010 10:09 AM, pieterg wrote:
>> On Tuesday 07 September 2010 21:03:14 Koen Kooi wrote:
>>> On 07-09-10 18:18, Khem Raj wrote:
>>>> On Tue, Sep 7, 2010 at 5:52 AM, Koen Kooi<k.kooi@student.utwente.nl>
>> wrote:
>>>>> it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
>>>>> create a libstdc6++ package, but it does create a libstdc-dev package.
>>>>
>>>> I see thats plausible. I think its time to look into separating gcc
>>>> runtime from gcc on the lines of poky.
>>>
>>> I put back libgcc and libstdc++ for the time being:
>>>
>>> http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=241b8865f20
>>> b9e3c3beb334535e6cd8452cbf47d
>>>
>>> I have a feeling the others needs to put back as well (e.g. fortran),
>>> but this fixes the immediate problem.
>>
>> Not for me, after a clean build today packages are still missing libstdc++
>> in their deps (and as a result the images are still missing libstdc++).
>> So there must be more to it.
>
> Does anyone have any idea what commit caused this breakage? It is causing a lot
> of problems for people.
>
> Philip

With git bisect I get commit d2a283ad002660e46b2f8d40a8ddfabad457d8a4 that 
causes libstdc++ to not get included.

Re: Images no longer automatically include libstdc++6

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 11-09-10 03:29, Dallas Foley wrote:
> On 10-09-08 07:33 AM, Philip Balister wrote:
>> On 09/08/2010 10:09 AM, pieterg wrote:
>>> On Tuesday 07 September 2010 21:03:14 Koen Kooi wrote:
>>>> On 07-09-10 18:18, Khem Raj wrote:
>>>>> On Tue, Sep 7, 2010 at 5:52 AM, Koen Kooi<k.kooi@student.utwente.nl>
>>> wrote:
>>>>>> it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
>>>>>> create a libstdc6++ package, but it does create a libstdc-dev
>>>>>> package.
>>>>>
>>>>> I see thats plausible. I think its time to look into separating gcc
>>>>> runtime from gcc on the lines of poky.
>>>>
>>>> I put back libgcc and libstdc++ for the time being:
>>>>
>>>> http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=241b8865f20
>>>>
>>>> b9e3c3beb334535e6cd8452cbf47d
>>>>
>>>> I have a feeling the others needs to put back as well (e.g. fortran),
>>>> but this fixes the immediate problem.
>>>
>>> Not for me, after a clean build today packages are still missing
>>> libstdc++
>>> in their deps (and as a result the images are still missing libstdc++).
>>> So there must be more to it.
>>
>> Does anyone have any idea what commit caused this breakage? It is
>> causing a lot
>> of problems for people.
>>
>> Philip
> 
> With git bisect I get commit d2a283ad002660e46b2f8d40a8ddfabad457d8a4
> that causes libstdc++ to not get included.

Yep, that's the one:

./x86_64-linux/shlibs/libssp.ver
./x86_64-linux/shlibs/libstdc++.ver
./x86_64-linux/shlibs/libgfortran.list
./x86_64-linux/shlibs/libssp.list
./x86_64-linux/shlibs/libstdc++.list
./x86_64-linux/shlibs/libgfortran.ver
./x86_64-linux/shlibs/libgcc.list
./x86_64-linux/shlibs/libgcc.ver

For gcc we probably need to add the following to gcc.inc:

SHLIBSDIR = "${STAGING_DIR}/${BASEPKG_HOST_SYS}/shlibs"

That should put the shlibs back where they belong for gcc.

regards,

Koen
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFMiyseMkyGM64RGpERAmS7AKC2VvzfFIzXTYKmgkMm2CF8olPvGACfcM48
XTf/BdxREFQe5KBQnVH/lbw=
=vEE1
-----END PGP SIGNATURE-----

patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Can people ack or nack the following please:


diff --git a/recipes/gcc/gcc-package-cross.inc
b/recipes/gcc/gcc-package-cross.inc
index 0c4ea2d..59e7dde 100644
- --- a/recipes/gcc/gcc-package-cross.inc
+++ b/recipes/gcc/gcc-package-cross.inc
@@ -2,6 +2,8 @@ PACKAGE_STRIP ?= ""
 OLD_PACKAGE_STRIP := "${PACKAGE_STRIP}"
 PACKAGE_STRIP = "no"

+SHLIBSDIR = "${STAGING_DIR_TARGET}/shlibs"
+
 PACKAGES = "libgcc libssp libstdc++ libg2c libgfortran"

 # Called from within gcc-cross, so libdir is set wrong


On 11-09-10 09:09, Koen Kooi wrote:
> On 11-09-10 03:29, Dallas Foley wrote:
>> On 10-09-08 07:33 AM, Philip Balister wrote:
>>> On 09/08/2010 10:09 AM, pieterg wrote:
>>>> On Tuesday 07 September 2010 21:03:14 Koen Kooi wrote:
>>>>> On 07-09-10 18:18, Khem Raj wrote:
>>>>>> On Tue, Sep 7, 2010 at 5:52 AM, Koen Kooi<k.kooi@student.utwente.nl>
>>>> wrote:
>>>>>>> it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
>>>>>>> create a libstdc6++ package, but it does create a libstdc-dev
>>>>>>> package.
>>>>>>
>>>>>> I see thats plausible. I think its time to look into separating gcc
>>>>>> runtime from gcc on the lines of poky.
>>>>>
>>>>> I put back libgcc and libstdc++ for the time being:
>>>>>
>>>>> http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=241b8865f20
>>>>>
>>>>> b9e3c3beb334535e6cd8452cbf47d
>>>>>
>>>>> I have a feeling the others needs to put back as well (e.g. fortran),
>>>>> but this fixes the immediate problem.
>>>>
>>>> Not for me, after a clean build today packages are still missing
>>>> libstdc++
>>>> in their deps (and as a result the images are still missing libstdc++).
>>>> So there must be more to it.
>>>
>>> Does anyone have any idea what commit caused this breakage? It is
>>> causing a lot
>>> of problems for people.
>>>
>>> Philip
> 
>> With git bisect I get commit d2a283ad002660e46b2f8d40a8ddfabad457d8a4
>> that causes libstdc++ to not get included.
> 
> Yep, that's the one:
> 
> ./x86_64-linux/shlibs/libssp.ver
> ./x86_64-linux/shlibs/libstdc++.ver
> ./x86_64-linux/shlibs/libgfortran.list
> ./x86_64-linux/shlibs/libssp.list
> ./x86_64-linux/shlibs/libstdc++.list
> ./x86_64-linux/shlibs/libgfortran.ver
> ./x86_64-linux/shlibs/libgcc.list
> ./x86_64-linux/shlibs/libgcc.ver
> 
> For gcc we probably need to add the following to gcc.inc:
> 
> SHLIBSDIR = "${STAGING_DIR}/${BASEPKG_HOST_SYS}/shlibs"
> 
> That should put the shlibs back where they belong for gcc.
> 
> regards,
> 
> Koen
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFMizPAMkyGM64RGpERAiAzAJ0SQGxLxecig5hDoAIZxXSBNfYQBwCfQaSb
fScJkHjvpFg/2t8Jw+dkGdY=
=iSyb
-----END PGP SIGNATURE-----

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Frans Meulenbroeks]

2010/9/11 Koen Kooi <k.kooi@student.utwente.nl>:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Can people ack or nack the following please:
>
>
> diff --git a/recipes/gcc/gcc-package-cross.inc
> b/recipes/gcc/gcc-package-cross.inc
> index 0c4ea2d..59e7dde 100644
> - --- a/recipes/gcc/gcc-package-cross.inc
> +++ b/recipes/gcc/gcc-package-cross.inc
> @@ -2,6 +2,8 @@ PACKAGE_STRIP ?= ""
>  OLD_PACKAGE_STRIP := "${PACKAGE_STRIP}"
>  PACKAGE_STRIP = "no"
>
> +SHLIBSDIR = "${STAGING_DIR_TARGET}/shlibs"
> +
>  PACKAGES = "libgcc libssp libstdc++ libg2c libgfortran"
>
>  # Called from within gcc-cross, so libdir is set wrong
>
>
Haven't had time to test this yet, but I feel at least a PR bump is
missing (or perhaps a set of PR bumps)

Frans

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 12-09-10 10:34, Frans Meulenbroeks wrote:
> 2010/9/11 Koen Kooi <k.kooi@student.utwente.nl>:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> Can people ack or nack the following please:
>>
>>
>> diff --git a/recipes/gcc/gcc-package-cross.inc
>> b/recipes/gcc/gcc-package-cross.inc
>> index 0c4ea2d..59e7dde 100644
>> - --- a/recipes/gcc/gcc-package-cross.inc
>> +++ b/recipes/gcc/gcc-package-cross.inc
>> @@ -2,6 +2,8 @@ PACKAGE_STRIP ?= ""
>>  OLD_PACKAGE_STRIP := "${PACKAGE_STRIP}"
>>  PACKAGE_STRIP = "no"
>>
>> +SHLIBSDIR = "${STAGING_DIR_TARGET}/shlibs"
>> +
>>  PACKAGES = "libgcc libssp libstdc++ libg2c libgfortran"
>>
>>  # Called from within gcc-cross, so libdir is set wrong
>>
>>
> Haven't had time to test this yet, but I feel at least a PR bump is
> missing (or perhaps a set of PR bumps)

Since it would require a PR bump for everything that is built with gcc,
I'm going to say:

That's what DISTRO_PR is for, which is up to distro maintainers to bump.
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFMjJlIMkyGM64RGpERAoBRAJ93j8dDURHApz8ieOElzKQJCQfOEgCfX/kz
x3qbuoq+k391UOukjTiEy1o=
=2ueT
-----END PGP SIGNATURE-----

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Frans Meulenbroeks]

2010/9/12 Koen Kooi <k.kooi@student.utwente.nl>:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 12-09-10 10:34, Frans Meulenbroeks wrote:

>> Haven't had time to test this yet, but I feel at least a PR bump is
>> missing (or perhaps a set of PR bumps)
>
> Since it would require a PR bump for everything that is built with gcc,
> I'm going to say:
>
> That's what DISTRO_PR is for, which is up to distro maintainers to bump.

Still I feel PR or INC_PR bumps of the affected gcc recipes is
appropriate, if only to fix it for distro's that do not immediately
bump DISTRO_PR.
We have quite some distro's and not all are maintained equally well,
so I can imagine it'll take some time for them to bump
Actually some distro's seem orphaned, so for those it might never
happen and bumping the GCC PR's at least will give them the fix too.

Thinking of it let's see how many distro's actually use DISTRO_PR:
../../conf/distro/angstrom-2008.1.conf:DISTRO_PR = ".6"
../../conf/distro/kaeilos-2010.conf:DISTRO_PR = ".1"
../../conf/distro/kaeilos.conf:DISTRO_PR = ".5"
../../conf/distro/angstrom-2010.x.conf:DISTRO_PR = ".7"
../../conf/distro/shr.conf:DISTRO_PR = ".5"

So that is 5 out of the 32 recipes.
Even one of your own (angstrom-2008.1-legacy.conf) does ot use DISTRO_PR

Very nice to keep the users of those 27 distro's in the cold!
Not.

Frans.

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 12-09-10 13:27, Frans Meulenbroeks wrote:
> 2010/9/12 Koen Kooi <k.kooi@student.utwente.nl>:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> On 12-09-10 10:34, Frans Meulenbroeks wrote:
> 
>>> Haven't had time to test this yet, but I feel at least a PR bump is
>>> missing (or perhaps a set of PR bumps)
>>
>> Since it would require a PR bump for everything that is built with gcc,
>> I'm going to say:
>>
>> That's what DISTRO_PR is for, which is up to distro maintainers to bump.
> 
> Still I feel PR or INC_PR bumps of the affected gcc recipes is
> appropriate, if only to fix it for distro's that do not immediately
> bump DISTRO_PR.
> We have quite some distro's and not all are maintained equally well,
> so I can imagine it'll take some time for them to bump
> Actually some distro's seem orphaned, so for those it might never
> happen and bumping the GCC PR's at least will give them the fix too.

It won't, since you need to rebuild everything that uses gcc. Bumping PR
for gcc-cross would just give people a false sense of security. I'm not
against bumping PR for gcc-cross, I'm just saying that it doesn't come
close to fixing the problem at hand, which is packages with missing
dependencies (libgcc for normal ones, libstdc++ for c++ ones,
libgfortran for fortran ones, etc).

> Thinking of it let's see how many distro's actually use DISTRO_PR:
> ../../conf/distro/angstrom-2008.1.conf:DISTRO_PR = ".6"
> ../../conf/distro/kaeilos-2010.conf:DISTRO_PR = ".1"
> ../../conf/distro/kaeilos.conf:DISTRO_PR = ".5"
> ../../conf/distro/angstrom-2010.x.conf:DISTRO_PR = ".7"
> ../../conf/distro/shr.conf:DISTRO_PR = ".5"
> 
> So that is 5 out of the 32 recipes.
> Even one of your own (angstrom-2008.1-legacy.conf) does ot use DISTRO_PR

Do you know how requires work in OE?

conf/distro$ grep require angstrom-2008.1-legacy.conf
require conf/distro/angstrom-2008.1.conf

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFMjL3+MkyGM64RGpERAsRvAJ43TZNi07Md1wcPxdz5T6KaYYl8AwCgvNLY
X+Hd1AJVQULZ0BDAiiU6X5A=
=hIhF
-----END PGP SIGNATURE-----

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Frans Meulenbroeks]

2010/9/12 Koen Kooi <k.kooi@student.utwente.nl>:

> On 12-09-10 13:27, Frans Meulenbroeks wrote:

> It won't, since you need to rebuild everything that uses gcc. Bumping PR
> for gcc-cross would just give people a false sense of security. I'm not
> against bumping PR for gcc-cross, I'm just saying that it doesn't come
> close to fixing the problem at hand, which is packages with missing
> dependencies (libgcc for normal ones, libstdc++ for c++ ones,
> libgfortran for fortran ones, etc).

I know it does not fix anything, but it fixes most of it (because
although there is a missing dep, if you install libstdc++/libgcc
manually or install a package made after the fix is applied things
still work.
>
>> Thinking of it let's see how many distro's actually use DISTRO_PR:
>> ../../conf/distro/angstrom-2008.1.conf:DISTRO_PR = ".6"
>> ../../conf/distro/kaeilos-2010.conf:DISTRO_PR = ".1"
>> ../../conf/distro/kaeilos.conf:DISTRO_PR = ".5"
>> ../../conf/distro/angstrom-2010.x.conf:DISTRO_PR = ".7"
>> ../../conf/distro/shr.conf:DISTRO_PR = ".5"
>>
>> So that is 5 out of the 32 recipes.
>> Even one of your own (angstrom-2008.1-legacy.conf) does ot use DISTRO_PR
>
> Do you know how requires work in OE?

Yes.
The grep above does not reveal these. There might be similar cases for
some of the other 32-5-1 = 26 distro's.
I just did not check.

Anyway the above just illustrates again that it is is probably quite
useful to clearly identify maintainers for all distro's and move the
unmaintained one to a separate folder or so, so people using those at
least are aware that they are using an unmaintained distro and as such
are kind-a "on their own"

Frans
>
> conf/distro$ grep require angstrom-2008.1-legacy.conf
> require conf/distro/angstrom-2008.1.conf
>
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.5 (Darwin)
>
> iD8DBQFMjL3+MkyGM64RGpERAsRvAJ43TZNi07Md1wcPxdz5T6KaYYl8AwCgvNLY
> X+Hd1AJVQULZ0BDAiiU6X5A=
> =hIhF
> -----END PGP SIGNATURE-----
>
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Philip Balister]

On 09/12/2010 07:27 AM, Frans Meulenbroeks wrote:
> 2010/9/12 Koen Kooi<k.kooi@student.utwente.nl>:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> On 12-09-10 10:34, Frans Meulenbroeks wrote:
>
>>> Haven't had time to test this yet, but I feel at least a PR bump is
>>> missing (or perhaps a set of PR bumps)
>>
>> Since it would require a PR bump for everything that is built with gcc,
>> I'm going to say:
>>
>> That's what DISTRO_PR is for, which is up to distro maintainers to bump.
>
> Still I feel PR or INC_PR bumps of the affected gcc recipes is
> appropriate, if only to fix it for distro's that do not immediately
> bump DISTRO_PR.
> We have quite some distro's and not all are maintained equally well,
> so I can imagine it'll take some time for them to bump
> Actually some distro's seem orphaned, so for those it might never
> happen and bumping the GCC PR's at least will give them the fix too.

If the DISTRO is not maintained, maybe it should be cleaned up and left 
around to confuse people who think it should work well.

Philip


>
> Thinking of it let's see how many distro's actually use DISTRO_PR:
> ../../conf/distro/angstrom-2008.1.conf:DISTRO_PR = ".6"
> ../../conf/distro/kaeilos-2010.conf:DISTRO_PR = ".1"
> ../../conf/distro/kaeilos.conf:DISTRO_PR = ".5"
> ../../conf/distro/angstrom-2010.x.conf:DISTRO_PR = ".7"
> ../../conf/distro/shr.conf:DISTRO_PR = ".5"
>
> So that is 5 out of the 32 recipes.
> Even one of your own (angstrom-2008.1-legacy.conf) does ot use DISTRO_PR
>
> Very nice to keep the users of those 27 distro's in the cold!
> Not.
>
> Frans.
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Frans Meulenbroeks]

2010/9/12 Philip Balister <philip@balister.org>:

> If the DISTRO is not maintained, maybe it should be cleaned up and left
> around to confuse people who think it should work well.
>

Philip,

I hope you wanted to say "...cleaned up and NOT left around..."
Otherwise it does not really make sense to me.

BTW: I already tried to identify distro maintainers and unmaintained
distros. See [1]
I still feel it is a good plan to either remove unmaintaned distros,
or move them (and the files that are specific to those distros) to an
unmaintained directory.
However it was even impossible to create consensus to remove files in
distro/include that were not used by any distro (and most likely not
maintained either) See [2]

[1] http://lists.linuxtogo.org/pipermail/openembedded-devel/2010-August/022698.html
[2] http://thread.gmane.org/gmane.comp.handhelds.openembedded/35427/focus=35946

Frans.

PS: and if anyone would need it, it would still be availalble in git
as well as in the stable/2009 branch. I can also imagine that we
create a tag before actually cleaning up.

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Philip Balister]

On 09/12/2010 08:49 AM, Frans Meulenbroeks wrote:
> 2010/9/12 Philip Balister<philip@balister.org>:
>
>> If the DISTRO is not maintained, maybe it should be cleaned up and left
>> around to confuse people who think it should work well.
>>
>
> Philip,
>
> I hope you wanted to say "...cleaned up and NOT left around..."
> Otherwise it does not really make sense to me.

I did. I am not functioning on a full load of coffee.

Seriously, some people may be attached to those distro's. However, if 
they are not using OISTRO_PR to force rebuilds may be due to their use 
case being different from say Angstrom. DISTRO_PR is useful for people 
maintained feeds for end users. Some of the DISTRO's without it, may 
just build images from clean builds as needed, so they don't really need 
it, and won't need the PR bump being discussed.

Philip

>
> BTW: I already tried to identify distro maintainers and unmaintained
> distros. See [1]
> I still feel it is a good plan to either remove unmaintaned distros,
> or move them (and the files that are specific to those distros) to an
> unmaintained directory.
> However it was even impossible to create consensus to remove files in
> distro/include that were not used by any distro (and most likely not
> maintained either) See [2]
>
> [1] http://lists.linuxtogo.org/pipermail/openembedded-devel/2010-August/022698.html
> [2] http://thread.gmane.org/gmane.comp.handhelds.openembedded/35427/focus=35946
>
> Frans.
>
> PS: and if anyone would need it, it would still be availalble in git
> as well as in the stable/2009 branch. I can also imagine that we
> create a tag before actually cleaning up.
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Frans Meulenbroeks]

2010/9/12 Philip Balister <philip@balister.org>:
> On 09/12/2010 08:49 AM, Frans Meulenbroeks wrote:
>>
>> 2010/9/12 Philip Balister<philip@balister.org>:
>>
>>> If the DISTRO is not maintained, maybe it should be cleaned up and left
>>> around to confuse people who think it should work well.
>>>
>>
>> Philip,
>>
>> I hope you wanted to say "...cleaned up and NOT left around..."
>> Otherwise it does not really make sense to me.
>
> I did. I am not functioning on a full load of coffee.
>
> Seriously, some people may be attached to those distro's. However, if they
> are not using OISTRO_PR to force rebuilds may be due to their use case being
> different from say Angstrom. DISTRO_PR is useful for people maintained feeds
> for end users. Some of the DISTRO's without it, may just build images from
> clean builds as needed, so they don't really need it, and won't need the PR
> bump being discussed.

I can understand that people are attached to some distros. I still
have a weak spot for slugos as that brought me to oe several years ago
(and the NSLU2 was a lot of fun!)

On the PR bump: even if you have a distro without DISTRO_PR, one may
locally still do a pull and do a partial rebuild (at least I do not do
a full rebuild every day)., so a PR bump would help.
Of course this fixes a fairly big problem, so it might be speeded up,
but in general I would say that a change which breaks a lot of things
(and requiring a full rebuild) should be announced clearly and timely
so people can be aware of it.
Thinking of it, we might also create/have/want something like OE_PR
which can be bumped if there are changes that require a complete
rebuild. This could be bumped for important, incompatible changes in
the toolchain or in key recipes.

Frans.

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Apart from bikeshedding on PR bumps, does this patch work for people
besides me and is the solution acceptable for gcc-cross?

On 11-09-10 09:46, Koen Kooi wrote:
> Can people ack or nack the following please:
> 
> 
> diff --git a/recipes/gcc/gcc-package-cross.inc
> b/recipes/gcc/gcc-package-cross.inc
> index 0c4ea2d..59e7dde 100644
> --- a/recipes/gcc/gcc-package-cross.inc
> +++ b/recipes/gcc/gcc-package-cross.inc
> @@ -2,6 +2,8 @@ PACKAGE_STRIP ?= ""
>  OLD_PACKAGE_STRIP := "${PACKAGE_STRIP}"
>  PACKAGE_STRIP = "no"
> 
> +SHLIBSDIR = "${STAGING_DIR_TARGET}/shlibs"
> +
>  PACKAGES = "libgcc libssp libstdc++ libg2c libgfortran"
> 
>  # Called from within gcc-cross, so libdir is set wrong
> 
> 
> On 11-09-10 09:09, Koen Kooi wrote:
>> On 11-09-10 03:29, Dallas Foley wrote:
>>> On 10-09-08 07:33 AM, Philip Balister wrote:
>>>> On 09/08/2010 10:09 AM, pieterg wrote:
>>>>> On Tuesday 07 September 2010 21:03:14 Koen Kooi wrote:
>>>>>> On 07-09-10 18:18, Khem Raj wrote:
>>>>>>> On Tue, Sep 7, 2010 at 5:52 AM, Koen Kooi<k.kooi@student.utwente.nl>
>>>>> wrote:
>>>>>>>> it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
>>>>>>>> create a libstdc6++ package, but it does create a libstdc-dev
>>>>>>>> package.
>>>>>>>
>>>>>>> I see thats plausible. I think its time to look into separating gcc
>>>>>>> runtime from gcc on the lines of poky.
>>>>>>
>>>>>> I put back libgcc and libstdc++ for the time being:
>>>>>>
>>>>>> http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=241b8865f20
>>>>>>
>>>>>> b9e3c3beb334535e6cd8452cbf47d
>>>>>>
>>>>>> I have a feeling the others needs to put back as well (e.g. fortran),
>>>>>> but this fixes the immediate problem.
>>>>>
>>>>> Not for me, after a clean build today packages are still missing
>>>>> libstdc++
>>>>> in their deps (and as a result the images are still missing libstdc++).
>>>>> So there must be more to it.
>>>>
>>>> Does anyone have any idea what commit caused this breakage? It is
>>>> causing a lot
>>>> of problems for people.
>>>>
>>>> Philip
> 
>>> With git bisect I get commit d2a283ad002660e46b2f8d40a8ddfabad457d8a4
>>> that causes libstdc++ to not get included.
> 
>> Yep, that's the one:
> 
>> ./x86_64-linux/shlibs/libssp.ver
>> ./x86_64-linux/shlibs/libstdc++.ver
>> ./x86_64-linux/shlibs/libgfortran.list
>> ./x86_64-linux/shlibs/libssp.list
>> ./x86_64-linux/shlibs/libstdc++.list
>> ./x86_64-linux/shlibs/libgfortran.ver
>> ./x86_64-linux/shlibs/libgcc.list
>> ./x86_64-linux/shlibs/libgcc.ver
> 
>> For gcc we probably need to add the following to gcc.inc:
> 
>> SHLIBSDIR = "${STAGING_DIR}/${BASEPKG_HOST_SYS}/shlibs"
> 
>> That should put the shlibs back where they belong for gcc.
> 
>> regards,
> 
>> Koen
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFMjSFkMkyGM64RGpERAkPvAJwPBO4vhW/iXLTQVdr6AjMoacCJnACeJhyy
N5VkQ1ouu2dCj1tet4p4z6Y=
=9JPc
-----END PGP SIGNATURE-----

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[pieterg]

On Saturday 11 September 2010 09:46:08 Koen Kooi wrote:
> Can people ack or nack the following please:
>
>
> diff --git a/recipes/gcc/gcc-package-cross.inc
> b/recipes/gcc/gcc-package-cross.inc
> index 0c4ea2d..59e7dde 100644
> --- a/recipes/gcc/gcc-package-cross.inc
> +++ b/recipes/gcc/gcc-package-cross.inc
> @@ -2,6 +2,8 @@ PACKAGE_STRIP ?= ""
>  OLD_PACKAGE_STRIP := "${PACKAGE_STRIP}"
>  PACKAGE_STRIP = "no"
>
> +SHLIBSDIR = "${STAGING_DIR_TARGET}/shlibs"
> +
>  PACKAGES = "libgcc libssp libstdc++ libg2c libgfortran"
>
>  # Called from within gcc-cross, so libdir is set wrong

Yes, that fixes it for me.

Rgds, Pieter

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[dfoley]

On 10-09-12 02:08 PM, pieterg wrote:
> On Saturday 11 September 2010 09:46:08 Koen Kooi wrote:
>> Can people ack or nack the following please:
>>
>>
>> diff --git a/recipes/gcc/gcc-package-cross.inc
>> b/recipes/gcc/gcc-package-cross.inc
>> index 0c4ea2d..59e7dde 100644
>> --- a/recipes/gcc/gcc-package-cross.inc
>> +++ b/recipes/gcc/gcc-package-cross.inc
>> @@ -2,6 +2,8 @@ PACKAGE_STRIP ?= ""
>>   OLD_PACKAGE_STRIP := "${PACKAGE_STRIP}"
>>   PACKAGE_STRIP = "no"
>>
>> +SHLIBSDIR = "${STAGING_DIR_TARGET}/shlibs"
>> +
>>   PACKAGES = "libgcc libssp libstdc++ libg2c libgfortran"
>>
>>   # Called from within gcc-cross, so libdir is set wrong
>
> Yes, that fixes it for me.
>
> Rgds, Pieter

ack

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Khem Raj]

On Sat, Sep 11, 2010 at 12:46 AM, Koen Kooi <k.kooi@student.utwente.nl> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Can people ack or nack the following please:
>
>
> diff --git a/recipes/gcc/gcc-package-cross.inc
> b/recipes/gcc/gcc-package-cross.inc
> index 0c4ea2d..59e7dde 100644
> - --- a/recipes/gcc/gcc-package-cross.inc
> +++ b/recipes/gcc/gcc-package-cross.inc
> @@ -2,6 +2,8 @@ PACKAGE_STRIP ?= ""
>  OLD_PACKAGE_STRIP := "${PACKAGE_STRIP}"
>  PACKAGE_STRIP = "no"
>
> +SHLIBSDIR = "${STAGING_DIR_TARGET}/shlibs"
> +

the patch itself seems good to me

Acked-by: Khem Raj <raj.khem@gmail.com>


>  PACKAGES = "libgcc libssp libstdc++ libg2c libgfortran"
>
>  # Called from within gcc-cross, so libdir is set wrong
>
>
> On 11-09-10 09:09, Koen Kooi wrote:
>> On 11-09-10 03:29, Dallas Foley wrote:
>>> On 10-09-08 07:33 AM, Philip Balister wrote:
>>>> On 09/08/2010 10:09 AM, pieterg wrote:
>>>>> On Tuesday 07 September 2010 21:03:14 Koen Kooi wrote:
>>>>>> On 07-09-10 18:18, Khem Raj wrote:
>>>>>>> On Tue, Sep 7, 2010 at 5:52 AM, Koen Kooi<k.kooi@student.utwente.nl>
>>>>> wrote:
>>>>>>>> it happens when you build 'gcc' after 'gcc-cross', gcc will doesn't
>>>>>>>> create a libstdc6++ package, but it does create a libstdc-dev
>>>>>>>> package.
>>>>>>>
>>>>>>> I see thats plausible. I think its time to look into separating gcc
>>>>>>> runtime from gcc on the lines of poky.
>>>>>>
>>>>>> I put back libgcc and libstdc++ for the time being:
>>>>>>
>>>>>> http://cgit.openembedded.org/cgit.cgi/openembedded/commit/?id=241b8865f20
>>>>>>
>>>>>> b9e3c3beb334535e6cd8452cbf47d
>>>>>>
>>>>>> I have a feeling the others needs to put back as well (e.g. fortran),
>>>>>> but this fixes the immediate problem.
>>>>>
>>>>> Not for me, after a clean build today packages are still missing
>>>>> libstdc++
>>>>> in their deps (and as a result the images are still missing libstdc++).
>>>>> So there must be more to it.
>>>>
>>>> Does anyone have any idea what commit caused this breakage? It is
>>>> causing a lot
>>>> of problems for people.
>>>>
>>>> Philip
>>
>>> With git bisect I get commit d2a283ad002660e46b2f8d40a8ddfabad457d8a4
>>> that causes libstdc++ to not get included.
>>
>> Yep, that's the one:
>>
>> ./x86_64-linux/shlibs/libssp.ver
>> ./x86_64-linux/shlibs/libstdc++.ver
>> ./x86_64-linux/shlibs/libgfortran.list
>> ./x86_64-linux/shlibs/libssp.list
>> ./x86_64-linux/shlibs/libstdc++.list
>> ./x86_64-linux/shlibs/libgfortran.ver
>> ./x86_64-linux/shlibs/libgcc.list
>> ./x86_64-linux/shlibs/libgcc.ver
>>
>> For gcc we probably need to add the following to gcc.inc:
>>
>> SHLIBSDIR = "${STAGING_DIR}/${BASEPKG_HOST_SYS}/shlibs"
>>
>> That should put the shlibs back where they belong for gcc.
>>
>> regards,
>>
>> Koen
> -----BEGIN PGP SIGNATURE-----
> Version: GnuPG v1.4.5 (Darwin)
>
> iD8DBQFMizPAMkyGM64RGpERAiAzAJ0SQGxLxecig5hDoAIZxXSBNfYQBwCfQaSb
> fScJkHjvpFg/2t8Jw+dkGdY=
> =iSyb
> -----END PGP SIGNATURE-----
>
>
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel
>



-- 
-Khem

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Frans Meulenbroeks]

2010/9/12 Koen Kooi <k.kooi@student.utwente.nl>:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Apart from bikeshedding on PR bumps, does this patch work for people
> besides me and is the solution acceptable for gcc-cross?
>
I don't know, I was planning to set up a test, but your usual kind and
friendly replies caused me to direct my efforts and cpu cycles on
different issues.

Anyway, as long as there is no PR bump, for me this is a NAK.
If it is fixed it should be fixed in the best way. A PR bump is part of that.

Frans

Re: patch to fix libstdc++ problem, was Re: Images no longer automatically include libstdc++6

[Graham Gower]

Frans, Koen.

You guys need to stop this shit.