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

[André Pönitz]

On Fri, Jun 16, 2017 at 07:09:20PM -0700, Thiago Macieira wrote:
> On Friday, 16 June 2017 16:35:54 PDT André Pönitz wrote:
> > On Fri, Jun 16, 2017 at 12:52:24PM -0700, Thiago Macieira wrote:
> > > Here's a tricky question: how do you create a QStringLiteral with
> > > non-ASCII characters that will work on all our supported
> > > compilers?
> > 
> > I don't have to, unless I have profiler supported evidence that it
> > matters.
> 
> QStringLiteral is infinitely faster at runtime than QString::fromUtf8.
> It also occupies roughly 33% less memory than QString::fromUtf8 of a
> literal, as the latter needs the narrow character equivalent
> (read-only memory) plus the UTF-16 converted text (heap, which has
> overhead).

We are talking about a handful strings with non-7-bit chars,
mostly in tests.

Cycles spend for these cases will never sum up to the order of
magnitude to one failed CI integration.

> This is for a simple, static message. 
> 
> > In the common case where it does not matter, the string can be
> > created at run time.
> 
> At a cost and overhead.

Dwarfed by the costs of the proposed alternative.
 
> >     QSTRINGLITERAL('M', 0xf6, 0xf6, 0xf6, 'p', '!')
> > 
> > instead of
> > 
> >     QStringLiteral("Möööp!")
> 
> Yeah, no thanks.

We are talking about a handful strings with non-7-bit chars,
mostly in tests.

Worse uglifications have been proposed to Qt proper for even
less gain.
 
> The whole point is that we can do better. All we need to do is drop an
> old, buggy compiler.

That's not 'better', that's a price to pay. 

I don't have any stake in MSVC nor in Windows in general, in fact
given that it's weekend, I do not care at all.

Decision making should weigh pros and cons. The observed pattern is
however "We see something sub-optimal, therefore we choose the other
approach, not matter matter what". This is unreasonable.

Andre'