[Releasing] Testing: 08/22/2012 + linux-g++ [fail]

[lars.knoll at nokia.com]

On Aug 23, 2012, at 10:20 AM, ext Thiago Macieira <thiago.macieira at intel.com> wrote:

> On quinta-feira, 23 de agosto de 2012 07.11.07, lars.knoll at nokia.com wrote:
>> On the other hand I am certain distributions could even live with a set of
>> sha1's as a 'release'. The best way for them (ignoring the generated
>> documentation at least) is to start from a snapshot of the git
>> repositories.
> 
> That's what they've been doing for months to test packages. But it does not 
> make a release. The release should have proper, downloadable tarballs. 
> Otherwise, different distros will generate the archives at different times, 
> using different processes, possibly introducing subtle errors.
> 
> If a SHA-1 is enough, then all we need is to git archive those SHA-1 and put 
> them in the same website. That's what I'm asking for.

Fine for me to add, but I need someone to do the actual work and hack the script generating the source packages. Are you signing up?

Currently all Qt packages are generated directly from qt5.git master, so it's simply about extracting the sha1's for each submodule there.
> 
>>> I disagree. It's pointless to release something that can't be properly
>>> built by the people who will take our code to the most number of people.
>> 
>> It's absolutely not pointless. First of all there are the modular source
>> repositories. I think our release announcement should also mention the
>> sha1's for each of them. As said above that's probably a better basis for
>> distributions to start with anyway. Building Qt from a git snapshot is not
>> exactly difficult, and with Qt 5 this should fit the workflow of the
>> distributions a lot better then what we had in Qt 4.x. So we do have an
>> improvement here.
> 
> Indeed. So why can't we have the tarballs in the website?
> 
> Also, please note that you're declaring that building from a set of specific 
> Git snapshots is supported. That means we should test that it works and I 
> haven't seen anyone report that yet. It also implies that, if the build 
> doesn't work, it stops the release too.

Currently, we always create things from a tested qt5.git sha1, so the CI system has tested the modular building quite a bit already.
> 
> Please confirm that this understanding is correct.
> 
>> No. Code wise we're very much there, and any further delay will hurt Qt more
>> then this issue of no split source packages. As said above, I don't see why
>> this should be a show stopper for distributions.
> 
> It would be a showstopper if they couldn't build. But if there's a solution 
> for them to build (see above), then it works.
> 
>> Believe me we're working on modularising the docs. But it's a rather
>> difficult problem to solve properly. Distributions could simply link to the
>> online versions for now.
> 
> I agree with that too. We declare that the building of documentation requires 
> extracting all the source code, which is an inconvenient step. In fact, I'd 
> accept that even in the 5.0 final. Most distros will simply use the online docs 
> or packaged docs anyway.

Ok, agree here. Hopefully this will change however.

Lars