This page is a snapshot from the LWG issues list, see the Library Active Issues List for more information and the meaning of Open status.

2833. Library needs to specify what it means when it declares a function constexpr

Section: [variant.ctor] Status: Open Submitter: Richard Smith Opened: 2016-11-28 Last modified: 2018-08-25

Priority: 2

View other active issues in [variant.ctor].

View all other issues in [variant.ctor].

View all issues with Open status.


The library has lots of functions declared constexpr, but it's not clear what that means. The constexpr keyword implies that there needs to be some invocation of the function, for some set of template arguments and function arguments, that is valid in a constant expression (otherwise the program would be ill-formed, with no diagnostic required), along with a few side conditions. I suspect the library intends to require something a lot stronger than that from implementations (something along the lines of "all calls that could reasonably be constant subexpressions are in fact constant subexpressions, unless otherwise stated").

[variant.ctor]/1 contains this, which should also be fixed:

"This function shall be constexpr if and only if the value-initialization of the alternative type T0 would satisfy the requirements for a constexpr function."

This is the wrong constraint: instead of constraining whether the function is constexpr, we should constrain whether a call to it is a constant subexpression.


This is has some considerable overlap with LWG 2289 but is phrased in a more general way.

[2016-12-16, Issues Telecon]

Priority 2; this is also the general case of 2829.

[2017-02-20, Alisdair comments and suggests concrete wording]

Below is is draft wording I was working on at Issaquah to try to address both issues.

[2017-11 Albuquerque Wednesday issue processing]

Status to Open; really needs a paper.

STL says "What about plus<T>?" plus<int> needs to be usable in a constexpr context, but plus<string> can't be.

[2017-11 Albuquerque Saturday issues processing]

Geoffrey to write a paper resolving this.

[2018-06 Rapperswil Thursday issues processing]

Geoffrey has been unable to write this paper due to time constraints. He wrote up his progress here. Daniel has offered to help someone to write this paper; he's willing to be a co-author.

[2018-08-23 Batavia Issues processing]

Michael Wong to investigate.

Proposed resolution:

This wording is relative to N4640.

  1. Modify [constexpr.functions] as indicated: constexpr functions and constructors [constexpr.functions]

    -1- This International Standard explicitly requires that certain standard library functions are constexpr (9.2.5 [dcl.constexpr]). If the specification for a templated entity requires that it shall be a constexpr templated entity, then that templated entity shall be usable in a constant expression.. An implementation shall notmay declare anyadditional standard library function signature as constexpr except for those where it is explicitly required. Within any header that provides any non-defining declarations of constexpr functions or constructors an implementation shall provide corresponding definitions.