Re: std::max: if versus ?:

[Marc Glisse <marc dot glisse at inria dot fr>]

On Thu, 20 Sep 2012, Paolo Carlini wrote:

> On 09/20/2012 01:55 PM, Marc Glisse wrote:
> > Hello,
> >
> > I noticed that std::max is implemented as:
> >
> >       //return  __a < __b ? __b : __a;
> >       if (__a < __b)
> >         return __b;
> >       return __a;
> >
> > Notice how it doesn't use the ?: code but the "if" version instead. Looking 
> > at the history, this seems to come from SGI. Does someone know why it was 
> > done this way?
> I'm going to tell you something very vague and non technical, but in the 
> past, outside std::max, I had quite often to use the above instead of a 
> conditional operator as a second thought, when (maybe on a less common 
> target) a test "unexpectedly" failed with a conversion error. Does that help 
> you figuring out the technical details of the issue?

When a and b have possibly different types, using "if" makes a lot of 
sense (I even advocated replacing ?: with "if" somewhere in the gcc 
codebase in PR50177). Inside std::max, a and b both have exactly the same 
type const _Tp&, so it shouldn't matter.

Note that libcxx uses ?: for its implementation of std::max.

I assume there are historical reasons, some compilers had bugs in their ?: 
support or produced slightly faster code with "if"...

--
Marc Glisse