broken ncurses support for python

broken ncurses support for python

[Sergey Lapin]

commit dd05e06b89906002f68d616a6326c962e725bc54
Author: Christopher Larson <chris_larson@mentor.com>
Date:   Tue Jan 10 13:51:18 2012 -0600

    ncurses-native: install to libdir, not base_libdir

breaks python ncurses support.

Any ideas on proper fixing? For myself I revert this locally and
everything works perfectly.
(I think .so files and .a archives should be in usr/lib while .so
can go to /lib which is not the case with current ncurses.)
so mv in do_install should be sufficient, not --libdir=${base_libdir}.

All the best,
S.

Re: broken ncurses support for python

[Sergey Lapin]

On Tue, Mar 6, 2012 at 3:08 PM, Sergey Lapin <slapinid@gmail.com> wrote:
> commit dd05e06b89906002f68d616a6326c962e725bc54
> Author: Christopher Larson <chris_larson@mentor.com>
> Date:   Tue Jan 10 13:51:18 2012 -0600
>
>    ncurses-native: install to libdir, not base_libdir

I was wrong with this one, actual culprit is this one:

commit 796c3d038fb7892a5e5206fb10217623de18853f
Author: Scott Garman <scott.a.garman@intel.com>
Date:   Wed Jan 4 22:30:29 2012 -0800


>
> breaks python ncurses support.
>
> Any ideas on proper fixing? For myself I revert this locally and
> everything works perfectly.
> (I think .so files and .a archives should be in usr/lib while .so
> can go to /lib which is not the case with current ncurses.)
> so mv in do_install should be sufficient, not --libdir=${base_libdir}.
>
> All the best,
> S.

Re: broken ncurses support for python

[Scott Garman]

On 03/06/2012 05:06 AM, Sergey Lapin wrote:
> On Tue, Mar 6, 2012 at 3:08 PM, Sergey Lapin<slapinid@gmail.com>  wrote:
>> commit dd05e06b89906002f68d616a6326c962e725bc54
>> Author: Christopher Larson<chris_larson@mentor.com>
>> Date:   Tue Jan 10 13:51:18 2012 -0600
>>
>>     ncurses-native: install to libdir, not base_libdir
>
> I was wrong with this one, actual culprit is this one:
>
> commit 796c3d038fb7892a5e5206fb10217623de18853f
> Author: Scott Garman<scott.a.garman@intel.com>
> Date:   Wed Jan 4 22:30:29 2012 -0800
>
>
>>
>> breaks python ncurses support.
>>
>> Any ideas on proper fixing? For myself I revert this locally and
>> everything works perfectly.
>> (I think .so files and .a archives should be in usr/lib while .so
>> can go to /lib which is not the case with current ncurses.)
>> so mv in do_install should be sufficient, not --libdir=${base_libdir}.

Could you be more specific about how python ncurses support is broken? I 
assume you're getting errors of unresolved symbols. If that's the case, 
you may need to rebuild the application so it links to ncurses from 
base_libdir.

Scott

-- 
Scott Garman
Embedded Linux Engineer - Yocto Project
Intel Open Source Technology Center

Re: broken ncurses support for python

[Sergey Lapin]

On Tue, Mar 6, 2012 at 11:20 PM, Scott Garman <scott.a.garman@intel.com> wrote:
> On 03/06/2012 05:06 AM, Sergey Lapin wrote:
>>
>> On Tue, Mar 6, 2012 at 3:08 PM, Sergey Lapin<slapinid@gmail.com>  wrote:
>>>
>>> commit dd05e06b89906002f68d616a6326c962e725bc54
>>> Author: Christopher Larson<chris_larson@mentor.com>
>>> Date:   Tue Jan 10 13:51:18 2012 -0600
>>>
>>>    ncurses-native: install to libdir, not base_libdir
>>
>>
>> I was wrong with this one, actual culprit is this one:
>>
>> commit 796c3d038fb7892a5e5206fb10217623de18853f
>> Author: Scott Garman<scott.a.garman@intel.com>
>> Date:   Wed Jan 4 22:30:29 2012 -0800
>>
>>
>>>
>>> breaks python ncurses support.
>>>
>>> Any ideas on proper fixing? For myself I revert this locally and
>>> everything works perfectly.
>>> (I think .so files and .a archives should be in usr/lib while .so
>>> can go to /lib which is not the case with current ncurses.)
>>> so mv in do_install should be sufficient, not --libdir=${base_libdir}.
>
>
> Could you be more specific about how python ncurses support is broken? I
> assume you're getting errors of unresolved symbols. If that's the case, you
> may need to rebuild the application so it links to ncurses from base_libdir.
No, python fails to find the library.
And it is bad practice to put develeopment files in /lib anyway.
I think, putting .so.x.x intu lib was sufficient instead of setting
whole libdir to /lib.
Or development files might be moved to /usr/lib back (and .so symlink re-created
to address this).

Re: broken ncurses support for python

[Scott Garman]

On 03/06/2012 02:33 PM, Sergey Lapin wrote:
>>>> breaks python ncurses support.
>>>>
>>>> Any ideas on proper fixing? For myself I revert this locally and
>>>> everything works perfectly.
>>>> (I think .so files and .a archives should be in usr/lib while .so
>>>> can go to /lib which is not the case with current ncurses.)
>>>> so mv in do_install should be sufficient, not --libdir=${base_libdir}.
>>
>>
>> Could you be more specific about how python ncurses support is broken? I
>> assume you're getting errors of unresolved symbols. If that's the case, you
>> may need to rebuild the application so it links to ncurses from base_libdir.
> No, python fails to find the library.
> And it is bad practice to put develeopment files in /lib anyway.
> I think, putting .so.x.x intu lib was sufficient instead of setting
> whole libdir to /lib.
> Or development files might be moved to /usr/lib back (and .so symlink re-created
> to address this).

Thanks, I understand what you're saying now. I hope to look into this 
tomorrow and develop a fix; I've been away at a conference all week so far.

Scott

-- 
Scott Garman
Embedded Linux Engineer - Yocto Project
Intel Open Source Technology Center

Re: broken ncurses support for python

[Scott Garman]

On 03/08/2012 10:48 AM, Scott Garman wrote:
> On 03/06/2012 02:33 PM, Sergey Lapin wrote:
>>>>> breaks python ncurses support.
>>>>>
>>>>> Any ideas on proper fixing? For myself I revert this locally and
>>>>> everything works perfectly.
>>>>> (I think .so files and .a archives should be in usr/lib while .so
>>>>> can go to /lib which is not the case with current ncurses.)
>>>>> so mv in do_install should be sufficient, not --libdir=${base_libdir}.
>>>
>>>
>>> Could you be more specific about how python ncurses support is broken? I
>>> assume you're getting errors of unresolved symbols. If that's the
>>> case, you
>>> may need to rebuild the application so it links to ncurses from
>>> base_libdir.
>> No, python fails to find the library.
>> And it is bad practice to put develeopment files in /lib anyway.
>> I think, putting .so.x.x intu lib was sufficient instead of setting
>> whole libdir to /lib.
>> Or development files might be moved to /usr/lib back (and .so symlink
>> re-created
>> to address this).
>
> Thanks, I understand what you're saying now. I hope to look into this
> tomorrow and develop a fix; I've been away at a conference all week so far.

Thanks again Sergey for troubleshooting this and suggesting a solution. 
I filed bug #2070 to track this and have created the following fix for it:

http://git.yoctoproject.org/cgit/cgit.cgi/poky-contrib/commit/?h=sgarman/python-ncurses-fix&id=23c73da5f13b6c586353edb8143fa21058bf3a55

I'm going to queue up some build tests to run over the weekend and plan 
to submit this first thing on Monday should that go well.

Regards,

Scott

-- 
Scott Garman
Embedded Linux Engineer - Yocto Project
Intel Open Source Technology Center

Re: broken ncurses support for python

[Andreas Oberritter]

Hello Scott,

On 10.03.2012 05:57, Scott Garman wrote:
> I filed bug #2070 to track this and have created the following fix for it:
> 
> http://git.yoctoproject.org/cgit/cgit.cgi/poky-contrib/commit/?h=sgarman/python-ncurses-fix&id=23c73da5f13b6c586353edb8143fa21058bf3a55

this will break if libdir equals base_libdir. A conditional is required
around mkdir and mv.

Regards,
Andreas

Re: broken ncurses support for python

[Scott Garman]

On 03/10/2012 03:06 AM, Andreas Oberritter wrote:
> Hello Scott,
>
> On 10.03.2012 05:57, Scott Garman wrote:
>> I filed bug #2070 to track this and have created the following fix for it:
>>
>> http://git.yoctoproject.org/cgit/cgit.cgi/poky-contrib/commit/?h=sgarman/python-ncurses-fix&id=23c73da5f13b6c586353edb8143fa21058bf3a55
>
> this will break if libdir equals base_libdir. A conditional is required
> around mkdir and mv.

Hi Andreas,

Good catch - my build testing uncovered this and I've made that and some 
other changes as well.

Thanks,

Scott

-- 
Scott Garman
Embedded Linux Engineer - Yocto Project
Intel Open Source Technology Center