[OE-core] Future of GCC

[Khem Raj]

On Tue, Jul 12, 2016 at 3:27 PM, Richard Purdie
<richard.purdie at linuxfoundation.org> wrote:
> On Tue, 2016-07-12 at 14:34 -0700, akuster808 wrote:
>>
>> On 07/12/2016 02:24 PM, Burton, Ross wrote:
>> > On 12 July 2016 at 22:14, akuster808 <akuster808 at gmail.com> wrote:
>> >
>> > > > Personally I was thinking that gcc 5.x and 6.x should stay in
>> > > > oe-core for
>> > > > this cycle, and then drop 5.x after the release.
>> > >
>> > >
>> > > Wouldn't that be dropped iff GCC 7.0 is release? or are you
>> > > saying we
>> > > should only have one GCC version?
>> > >
>> >
>> > Well, depends on the adoption and migration problems.  I don't
>> > think we
>> > should carry three versions,
>>
>> I agree. 3 is too many.
>>
>>  and ideally one, but two is acceptable to ease
>> > migration.
>>
>> One makes Stable maintenance less costly in time.
>

When we upgrade gcc and it becomes default then most of the validation
and testing for that release happens with that version of gcc. Yes we have
been carrying the older version of gcc however it goes our largely untested
in the new release. We even dont know if the new recipes we upgrade in
the given release have any issues or build with this version of gcc. Everyone
should just default to the default gcc that comes with the release.  So if you
have been on 2.0 release and your apps building with gcc5, upgrading to yocto
2.2 where gcc6 becomes default may not guarantee you well tested metadata
for gcc5. So its in your interest to start using gcc6 with 2.2, and if
one still needs
to keep using gcc5, its not a big deal for them to carry the compiler recipes
in their own layer.


> I'm personally a fan of one if we can do it. We've taken a bit of an
> "easier" path recently but it might be time to change that. Right now
> I'm not aware of any of our core usecases which need 5.x, all work with
> 6.x. I am aware of some BSPs on older kernels which would however have
> issues.
>
> I did nearly send a 5.x removal patch but wasn't sure it would be
> accepted by people quite yet...

You should send it. This will help us concentrate on improving the quality
of default compiler.

>
> Cheers,
>
> Richard
>
> --
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core at lists.openembedded.org
> http://lists.openembedded.org/mailman/listinfo/openembedded-core