Proposed resolution:

This wording is relative to N4981.

[Drafting Note: Two mutually exclusive options are prepared, depicted below by Option A and Option B, respectively.]

Option A: As suggested in issue LWG 2326

1. Modify [rand.req.genl] as indicated:

   -1- Throughout this subclause [rand], the effect of instantiating a template:

   1. […]

   2. (1.4) — that has a template type parameter named RealType is undefined unless the corresponding template argument is cv-unqualified and is one of float, double, or long double.

   3. (1.5) — that has a template type parameter named IntType is undefined unless the corresponding template argument is cv-unqualified and is a standard integer type ([basic.fundamental])one of short, int, long, long long, unsigned short, unsigned int, unsigned long, or unsigned long long.

   4. (1.6) — that has a template type parameter named UIntType is undefined unless the corresponding template argument is cv-unqualified and is a standard unsigned integer type ([basic.fundamental])one of unsigned short, unsigned int, unsigned long, or unsigned long long.

Option B: Make ill-formed.

1. Modify [rand.req.genl] as indicated:

   -1- Throughout this subclause [rand], the effect of instantiating a template:

   1. […]

   2. (1.4) — that has a template type parameter named RealType is undefined unless the corresponding template argument is cv-unqualified and is one of float, double, or long double.

   3. (1.5) — that has a template type parameter named IntType renders the program ill-formedis undefined unless the corresponding template argument is cv-unqualified and is one of short, int, long, long long, unsigned short, unsigned int, unsigned long, or unsigned long long.

   4. (1.6) — that has a template type parameter named UIntType renders the program ill-formedis undefined unless the corresponding template argument is cv-unqualified and is one of unsigned short, unsigned int, unsigned long, or unsigned long long.

[ 2025-10-21; Reflector poll. ]

Set priority to 3 after reflector poll.

"The only benefit of making it UB is allowing it as an extension."

"Does anybody actually want to support a BigInt as an extension? If not, non-integral types should be ill-formed and integral types not in the list should be conditionally-supported."

"I'd like extended arithmetic types to work in general, it's useful for writing generic math libraries to be able to generate random numbers of any arithmetic type that the implementation provides, including short integers and all floating-point types."

"Can have overflows here, differences between whether `char` is signed or unsigned lead to subtle bugs. Most users are better off picking `int` or `unsigned int` (which is what they get from integral promotions anyway) and just using values that are in range for their intended destination type."

[ 2024-05-26; Daniel comments ]

I think that all violations of the bullets [rand.req.genl] (1.4), (1.5), and (1.6) are missed opportunities of Mandates (That is: Make the program ill-formed), because they can be all checked (easily) at compile-time, regardless whether we agree on the question to support int8_t/uint8_t (Violations of (1.1), (1.2), and (1.3) still have to be remain undefined because of additional runtime requirements imposed).

Given that I also think that we should consider to either normatively extend all of (1.4), (1.5), and (1.6) to corresponding extended floating point types and extended (unsigned) integer types, or to the minimum make these extended types conditionally-supported with implementation-defined semantics (A word of power that is used at several places).

As pointed out in LWG issue 2326 (closed as NAD) and on reddit, instantiating e.g. uniform_int_distribution<uint8_t> is undefined behavior because of the requirement [rand.req.genl] bullet (1.5):

-1- Throughout this subclause [rand], the effect of instantiating a template:

1. […]

2. (1.4) — that has a template type parameter named RealType is undefined unless the corresponding template argument is cv-unqualified and is one of float, double, or long double.

3. (1.5) — that has a template type parameter named IntType is undefined unless the corresponding template argument is cv-unqualified and is one of short, int, long, long long, unsigned short, unsigned int, unsigned long, or unsigned long long.

4. (1.6) — that has a template type parameter named UIntType is undefined unless the corresponding template argument is cv-unqualified and is one of unsigned short, unsigned int, unsigned long, or unsigned long long.

This is, in my opinion, a defect; such uses should either be rejected at compile time (made ill-formed), or permitted (as 2326 proposes.)

UB here has undesirable safety implications, because it's possible to write code that produces a random, or a seemingly random, sequence of uint8_t numbers on platform A, but an arbitrarily non-random sequence on platform B (e.g. all zeroes.)

If that sequence is then used in e.g. a cryptographic algorithm, bad things will happen on platform B, and the tests on platform A won't catch the issue.