Re: [RFC PATCH 0/9] New distribution definition for poky-tiny

[Darren Hart <dvhart@linux.intel.com>]

On 12/21/2011 09:33 AM, Chris Larson wrote:
> On Wed, Dec 21, 2011 at 9:10 AM, Darren Hart <dvhart@linux.intel.com> wrote:
>> On 12/21/2011 07:52 AM, Chris Larson wrote:
>>> On Wed, Dec 21, 2011 at 2:02 AM, Darren Hart <dvhart@linux.intel.com> wrote:
>>>> For example, I could not discover a way to use
>>>> oe_filter_out inside poky-tiny.conf to remove "keyboard" for MACHINE_FEATURES.
>>>
>>> This might be best handled via inclusion of a .inc which defines an
>>
>> I assume you mean to include the .inc in task-core-boot?
> 
> If that works, sure, but that wasn't my intent. You can
> include/require a .inc from a .conf.
> 
>>> event handler or anonymous python function that removes it. The
>>
>> I'm not familiar with adding event handlers, can you point me at an
>> example? In general, my concern is that whatever this mechanism is, it
>> shouldn't have to be poky-tiny aware: meaning it needs to glean the
>> knowledge that keymaps should not be added from something else. I
>> suppose this could be some DISTRO_FEATURE, but that is effectively what
>> I've already done in 5/9 "task-core-boot: Allow DISTRO to disable keymaps".
> 
> That seems reasonable. Event handlers are used in a number of places
> in the metadata, including base.bbclass :)
> 
> python do_some_stuff_to_the_config_metadata () {
>     if not isinstance(e, bb.event.ConfigParsed):
>         return
> 
>     # do stuff
> }
> addhandler do_some_stuff_to_the_config_metadata
> 
> 
>>> problem with the := + oe_filter_out method is that it forces an
>>> immediate variable expansion right there, which can cause problems in
>>> some cases.
>>
>> Right. That and using it in a .conf file results in parsing errors. :-)
> 
> := is *technically* perfectly valid in a .conf file, as is
> oe_filter_out, but it's so sensitive to location as to be of limited
> usefulness. All the variables used by the var you're expanding would
> have to have been defined already as of that point in the config.

Hrm, it reported parsing errors on the ${@oe_filter_out not being
defined as well as the "require utils" I added to address that. Maybe
I'm missing something else.

But, I think I'll be taking the VIRTUAL-RUNTIME_keymaps approach
suggested by Martin as that seems consistent with apm and other such things.

-- 
Darren Hart
Intel Open Source Technology Center
Yocto Project - Linux Kernel