C++ Logo

SG12

Advanced search

Subject: Re: [ub] Lvalue-to-rvalue conversion and reinterpret_cast-ed two's complement integers
From: Jens Maurer (Jens.Maurer_at_[hidden])
Date: 2019-06-27 01:16:59


On 26/06/2019 23.22, Language Lawyer wrote:
> Excuse my annoyingness, I know I shouldn't have sent this...
>
> Lets look at the code below:
>
> int i = -1;
> unsigned u = reinterpret_cast<unsigned&>(i);
>
> To my reading of the standard, it has undefined behavior.
> `reinterpret_cast<unsigned&>(i)` is an lvalue of `unsigned int` type denoting the `i` object of type `int` which holds the value `-1`.
> Lvalue-to-rvalue conversion applied to this lvalue in the initialization produces a prvalue of type `unsigned int`.
> The result of this prvalue is determined according to [conv.lval]/3, of which the bullet â„–3.4 apply:
>
> (3.4) — Otherwise, the value contained in the object indicated by the glvalue is the prvalue result.
>
> So, we have an `unsigned int` prvalue with `-1` as its result.
> According to [basic.fundamental]/2, the range representable by `unsigned int` is from 0 to 2^N - 1 (inclusive) for some N ≥ 16.
> [expr.pre]/4 suggests that we have UB here:
>
> [expr.pre]/4 If during the evaluation of an expression, the result is not mathematically defined or not in the range of representable values for its type, the behavior is undefined.
>
> Given that on most implementations (before and, especially, after P1236R1 was merged into the WP) one would get `u` initialized to `UINT_MAX` as the "result" of the UB, I propose to legalize this behavior.

Why doesn't bit_cast<> satisfy your needs?

Jens

> *Proposed wording*
>
> Modify [conv.lval]/3 as follows (changes are relative to N4820):
>
> The result of the conversion is determined according to the following rules:
> — If T is cv std::nullptr_­t, the result is a null pointer constant ([conv.ptr]). [ Note: Since the conversion does not access the object to which the glvalue refers, there is no side effect even if T is volatile-qualified ([intro.execution]), and the glvalue can refer to an inactive member of a union ([class.union]). — end note]
> — Otherwise, if T has a class type, the conversion copy-initializes the result object from the glvalue.
> — Otherwise, if the object to which the glvalue refers contains an invalid pointer value ([basic.stc.dynamic.deallocation], [basic.stc.dynamic.safety]), the behavior is implementation-defined.
> <ins> — Otherwise, if T is a signed integer type and the object to which the glvalue refers has corresponding unsigned integer type or vice versa, the result is the unique value from the range of representable values for the type T that is congruent to the stored value of the object modulo 2^N, where N is the width of the type T ([basic.fundamental]).</ins>
> — Otherwise, the value contained in the object indicated by the glvalue is the prvalue result.
> _______________________________________________
> ub mailing list
> ub_at_[hidden]
> http://www.open-std.org/mailman/listinfo/ub
>


SG12 list run by herb.sutter at gmail.com