[Development] Let's please drop MSVC 2013 for 5.10

[Marc Mutz]

On 2017-06-23 21:45, Adam Treat wrote:
> On 06/23/2017 03:23 PM, Marc Mutz wrote:
>> [forgot to CC list]
>> 
>> On 2017-06-23 19:50, Thiago Macieira wrote:
>>> On Friday, 23 June 2017 09:17:55 PDT Marc Mutz wrote:
>>>> The above argument makes no sense to me. What value does quoting
>>>> download numbers for 5.9, an LTS, have, to argue about dropping the
>>>> compiler from 5.10. Ever since we provide LTSs (yes, once), we drop
>>>> compilers in the version _after_ the LTS, which is kind of the 
>>>> natural
>>>> point to drop stuff.
>>> 
>>> The point is that in June 2017, 30% of the Windows downloads were for 
>>> MSVC
>>> 2013. it doesn't matter that this is an LTS release or not: way too 
>>> many
>>> people are using that compiler. We need to "wean" them off 2013, so 
>>> I'm
>>> starting a note in the 5.9.1 changelog that it will be gone in 5.11 
>>> (one year
>>> advance notice):
>> 
>> We didn't "wean people off" 2012, either. And how can we decide to 
>> drop 2013 in 5.11 if we don't know what the download numbers will be, 
>> then, yet? Who's crystal ball makes such decisions possible?
> 
> Now you favor using retrospective downloads as sole criterion? You
> only want to drop support when downloads go below a certain threshold?
> 
> Or is this your attempt at a reductio ad absurdum argument that using
> downloads to inform is only legit when used retrospectively as sole
> criterion?

Neither. In particular, it's not about downloads as a sole criterion. I 
was wondering how we can know that by 5.11 the download ratio for 2013 
is sufficiently low so as to commit to the drop date already now whilst 
at the same time using the current high download numbers to prevent 
dropping 2013 from 5.10 right now. What's the contingency plan in case 
the download numbers six months from now are still 30%? Either we still 
drop, because we said so, then the download argument becomes totally 
bogus, or we don't drop, in which case the announcement Thiago intends 
to add to the 5.9.1 changelog will be totally bogus.

The deeper point is that I think we should drop stuff in LTS+1, not 
after LTS+1. If we don't drop 2013 for 5.10, we need to carry it until 
the next Qt 5 LTS, imo. I don't care much either way, as my baby 
(QStringView) works fine with MSVC 2013. But getting there _was_ hard, 
and the pain will only grow for every new feature we add (though GCC can 
also be a PITA, as QThread::create() currently shows).

Thanks,
Marc