C++ Logo

SG12

Advanced search

Subject: Re: [ub] Justification for < not being a total order on pointers?
From: Ville Voutilainen (ville.voutilainen_at_[hidden])
Date: 2013-08-26 14:04:00


On 26 August 2013 22:01, Daniel Krügler <daniel.kruegler_at_[hidden]> wrote:

> 2013/8/26 Ville Voutilainen <ville.voutilainen_at_[hidden]>:
> > On 26 August 2013 19:00, Jeffrey Yasskin <jyasskin_at_[hidden]> wrote:
> >>
> >> The library's running into problems because < is unspecified on
> >> pointers that don't point to the same object. (C++14CD[expr.rel]p4)
> >> Naively, this restriction makes a bit of sense, since we can't give a
> >> particular answer for any two pointers, but it's causing problems for
> >> some generic library components. Specifically:
> >>
> >> std::map<T*> works because std::less has a special case for pointers.
> >> ([comparisons]p14)
> >>
> >> std::map<std::tuple<T*>> gives undefined behavior because
> >> std::less<tuple> uses tuple's operator<, which uses T*'s operator<.
> >>
> >> std::map<std::optional<T*>> goes back to defined behavior because we
> >> special-cased optional's operator< to use std::less:
> >> [optional.relops]p4
> >
> > I have strong reasons to believe we'll see an NB comment proposing
> std::less
> > to be specialized for tuple<T*> and containers.
>
> I don't understand why there should be different specializations
> instead of a single specialization:
>
> less<tuple<T...>>
>
> which defines the operator() overload in terms of std::less<T> for
> each T in order. Same thing for pair and optional and containers.
>
>
>
Sounds like a reasonable approach to me.



SG12 list run by herb.sutter at gmail.com