[oe] Yocto Project and OE - Where now?

[Koen Kooi]

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

On 18-01-11 21:48, Tom Rini wrote:
> On 01/18/2011 01:12 PM, Koen Kooi wrote:
>> -----BEGIN PGP SIGNED MESSAGE-----
>> Hash: SHA1
>>
>> On 18-01-11 17:54, Tom Rini wrote:
>>> On 01/18/2011 01:05 AM, Otavio Salvador wrote:
>>>> On Tue, Jan 18, 2011 at 05:21, Graeme Gregory<dp at xora.org.uk>   wrote:
>>>>> On 17/01/2011 19:01, Frans Meulenbroeks wrote:
>>>>>> - where possible stick to one recipe per package. This reduces the
>>>>>> maintenance work and reduces the QA nightmare of lots of different
>>>>>> permutations.
>>>>>> I feel one recipe per package should be the common case for
>>>>>> applications, and preferably also for libs (although I am well aware
>>>>>> that especially in the latter case multiple versions cannot always be
>>>>>> avoided).
>>>>>
>>>>> OE is not a distro so this is a non starter already, please don't bog
>>>>> down this discussion by re-opening this again. Angstrom 2008, Angstrom
>>>>> 2010, kaelios and slugos are all released distributions with different
>>>>> versions of apps just as a starter and they arent even near the total
>>>>> number of distros in OE.
>>>>
>>>> I disagree. I think having too many versions of a package just makes
>>>> difficult to get things done:
>>>>
>>>>    - it increases the amount of maintainence work;
>>>>    - has a bigger time to get bugs spoted;
>>>>
>>>> Users of old distros ought to use a specific repository and branch.
>>>> Master ought to be kept clean for 'next distro release'.
>>>
>>> I agree, at least going forward.  We must make it easier for
>>> distributions to say "here is my 'stable' release" and "here is my
>>> development release".
>>>
>>> First, I'm not picking on Angstrom here, really, I swear.  It's just a
>>> good example.
>>>
>>> But we also don't want to be unreasonable or unbending here.  We'll have
>>> to have multiple udevs (due to having different kernel versions as some
>>> HW isn't on the latest and greatest).  And if DistroA says they really
>>> want to stick to busybox 1.17.4 for a while, we should let that happen
>>> too.  But I don't think we want to have to carry on the recipes that
>>> angstrom-2008.1 wants and angstrom-2010.x wants and angstrom-2011.x
>>> wants and angstrom-2012.x want into 2013, in master.
>>
>> And noone says you should. At some point 2010.x works well enough to
>> force 2008.1 into hiding and start 201Y.x. The current situation where
>> the "unstable" 2010.x ended up in a product is largely due to the gcc
>> people breaking the NEON intrinsics interface API in between 4.3 and 4.5.
> 
> ick, I didn't know about that...

If anyone asks why uhd doesn't build for angstrom 2008, but it does
build for 2010, you now know why...

>>> For example, at some point we want to switch to libtool 2.4 only.  And
>>> that would certainly be a headache for angstrom-2008.1 (but we're glad,
>>> really! for angstrom-2010.x using 2.4 and testing and fixing things). So
>>> wouldn't it be a good thing to be able to say that if you want
>>> angstrom-2008.1 you do ... this ... and get the layers that give a good
>>> stable 2008.1, based on whatever policy Angstrom wants for doing
>>> updates?
>>
>> In the past the angstrom people created a stable branch and supported
>> that for a given release. The same can be done in the layering script,
>> where it would just lock down to certain revisions of various layers.
> 
> So, I think we agree.  Distros should be saying "if you want our stable
> release you should be over here..." and if you want our development WIP,
> you should be over here.

Exactly. Although it would make sense to have them coexist in the same
tree for a while while sorting out infrastructure. There are only so
many work hours in a day :)

>> But in the end if boils down to "Does OE wants to make life hard for
>> DISTROs or easy". Frans is firmly in the "make it hard" camp, I hope
>> others have a saner point of view.
>>
>> If you're forcing 90% of your users to put e.g. udev_162.bb in their
>> layer you're doing it wrong. But you're also doing it wrong if you have
>> 20 udev recipes :)
> 
> I think we also agree here.  But what's the rule of thumb(s) we want to
> have, to provide enough choice without too much headache?  As I said
> elsewhere, .inc files should probably be used a whole lot more, to help
> with the problem of recipe bugfixing and N recipes for an app with the
> problem.  We should probably also say that in addition to the "keep the
> last GPLv2+ version around" rule of thumb we should also have a "keep
> the latest stable release" around too.  But what else?  To use busybox
> as an example, do we really need to keep 1.18.0 and 1.18.1 around when
> we have 1.18.2?  How about if we make the delta between the 3 be just
> the SRC_URI + checksums?

Something like:

* keep gplv2 around if upstream switched to gplv3 at some point
* allow old stable, stable and dev (e.g. glib 2.24, 2.26 and 2.27+git)
* allow active contributers some leeway to test multiple subreleases
(e.g. busybox 1.18.[0-99]
* The maintainer can have as many versions as he wants to.
* toolchain is exempt from all that since having 20 binutils versions is
sometimes needed when building for 20 different archs[1].

But I seriously think we are overengineering this. We should have people
actually working on oe-core and meta-oe reach a consensus when
encountering problems.

regards,

Koen

[1] Although you can cheat by having different SRCREVs for each arch in
binutils_git.bb
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFNNg6aMkyGM64RGpERAjdvAJoDSyv62VkZhr1Gs5pdWIYd0BZOdwCfYAjG
ManP61a+tHnNNuaC3a1kMWU=
=Wlmx
-----END PGP SIGNATURE-----