**Submitter:**Fred Tydeman

**Submission Date:** 2018-12-12

**Document:** WG14 N2321

**Reference Documents:** N2302

Summary

For most floating-point formats, **nextafterl**(1.L,2.L) is
the normalized number whose value is 1.L+**LDBL_EPSION**.
However, when **long double** is a pair of **double**s, it
is not clear what the standard requires (could be
1.L+**LDBL_TRUE_MIN**), what implementations do, and what
users want. At least one implementation punted on this by not
providing **nextafterl**; while they do provide
**nextafter** and **nextafterf**

It seems that making **nextafterl**(1.L,2.L) be
1.L+**LDBL_EPSION** for all implementations would help with
portability.

Technical Corrigendum:

Add this paragraph:

A finite result shall be either a zero, a subnormal floating-point number, or a normalized floating-point number.

to:

§7.12.11.3 The **nextafter** functions,
§7.12.11.4 The **nexttoward** functions, §7.12.? The
**nextup** functions, and §7.12.? The **nextdown**
functions in the **Description** section.