[Buildroot] [PATCH] package/libglib2: add missing dependency on host-python[3]

[Andrey Smirnov]

On Sat, Jul 8, 2017 at 4:03 PM, Arnout Vandecappelle <arnout at mind.be> wrote:
>
>
> On 06-07-17 02:04, Andrey Smirnov wrote:
>> On Tue, Jul 4, 2017 at 5:36 PM, Arnout Vandecappelle <arnout at mind.be> wrote:
>>>
>>>
>>> On 05-07-17 02:30, Andrey Smirnov wrote:
>>>> Building gio/gdbus-2.0/codegen/ subdirectory requires host to have a
>>>> python interpreter. Add 'host-python'/'host-python3' as a dependency
>>>> to reflect that fact and prevent build failures.
>>>
>>>  Do you have an actual build failure for this? We already require a python to be
>>> present on the host, so there shouldn't be a need to build one here. And we
>>> really do want to avoid that libglib2 pulls in host-python (to limit the build
>>> time).
>>
>> It's a corner case involving partial rebuilds, so maybe not worth
>> trying to fix, but it can be reproduced in configuration that pulls
>> in/builds some flavor of "host-python" _before_ "host-libglib2" by
>> doing:
>>
>> find ${BUILD_DIR} -name ".stamp_host_installed" | xargs rm -f
>> rm -rf ${HOST_DIR}
>> make host-libglib2-install
>>
>> what happens is host-libglib2's ./configure picks up python from PATH,
>> which if built after host-python, ends up being python from
>> ${HOST_DIR}, but since host-libglib2 doesn't reference host-python as
>> a dependency re-executing "host-libglib2-install" ends up in failure.
>
>  We typically solve this by adding a dependency only if BR2_PACKAGE_PYTHON{,3}
> is selected. However...
>

Ah, makes sense. Good to know.

>
>> Another option to handle this would be to explicitly specify which
>> python to use with --with-python when configuring libglib2.
>
>  This is a lot more convenient, since the dependency really is not needed. Since
> we require both python and which to be in PATH, you can do
>
> --with-python=`which python`
>
>  Can you roll a patch for that? Don't forget to add a brief comment why this is
> needed, and a more extensive explanation in the commit log.
>

Yeah, sounds good. I'll send a separate patch, so please disregard this one.

Thanks,
Andrey Smirnov