Re: ABI handling for linux-2.6
Sven Luther <sven.luther@wanadoo.fr> writes:
> On Wed, Jun 14, 2006 at 06:29:21PM +0200, Martin Michlmayr wrote:
>> * Sven Luther <sven.luther@wanadoo.fr> [2006-06-14 18:25]:
>> > The issue are ABI changes, but also introduction of new flavours.
>>
>> Yeah, just like any package... I don't see why we should get special
>> treatment.
>
> Because abi breakage and the subsequent NEW handling cause undue delay in
> security updates for example.
>
> But more to the point, what is the reason of NEW handling ?
>
> It is to test :
>
> 1) that licencing or crypto or whatever are correctly handled. This is
> hardly going to be a problem for the kernel, and currently only affects new
> source packages, and it can be circunvented by uploading an anodine NEW
> package, and then changing the content on the sly, like Mr schiling has done
> for star or smake for example.
>
> 2) that there is no undue multiplication of new binary packages. Well, this
> asks the question of responsability, and of who is in the end responsible to
> make the decision. If we want to add a new flavour, is it our right as the
> kernel team to do so, or do the ftp-masters need to take the decision, and
> if this is the case, are they knowledgeable enough to take the right
> decision ?
>
> 3) as protection of library-abi changes, in order to not to hose large
> amounts of packages and cause a mess.
4) To send a mail to the US gov. due to crypto export regulations.
> My claim is that none of these reasons apply to the kernel package, since we
> are hardly introducing new source code, and the licence is not going to change
> from a day to the next, and we as the kernel team have been even more involved
> in licence issues than the ftp-master, as the non-free firmware problem has
> showed. Furthermore, the abi-changes is not akin of the library ones, and we
> are working in order to control the whole involved subset of packages in one
> way or another. The only claim would be 2), where the ftp-master team could
> decide to control the introduction of new flavours, but this has never been
> done, and i do believe that in the current situation the kernel team has the
> last word on this.
An abi change in the kernel is exactly the same as an abi change in
say mozilla. When the abi changes all modules/plugins have to be
recompiled. Same thing.
> On the other hand, the only reason to keep NEW for linux-2.6, is because aj
> decided so and refused to even discuss it, and because it has always been done
> like this.
Be aware that "it has always been done that way" comes along with "the
infrastructure does it that way" and "who will write the patch for
this". Three very strong arguments that will nock you out. :)
> Friendly,
>
> Sven Luther
MfG
Goswin
Reply to: