To: J3                                                     J3/26-210
From: Malcolm Cohen
Subject: Interp F23/026 on IEEE_SCALB
Date: 2026-October-20


----------------------------------------------------------------------

NUMBER: F23/026
TITLE: IEEE_SCALB does not conform to ISO/IEC/IEEE 60559
KEYWORDS: IEEE_SCALB, ISO/IEC/IEEE 60559
DEFECT TYPE: Erratum
STATUS: J3 consideration in progress
REFERENCES: 24-007, N2244 (F2023 draft corrigendum 1), N2258

QUESTION:

IEEE_SCALB, as per 24-007 with the changes in N2244 (corrigendum 1)
still fails to conform to ISO/IEC/IEEE 60559 as listed below.

It will be easier to view this in N2258 (draft CD for F2028), as that
has the corrigendum changes applied in-place.

Problem 1: it talks about 2**I, but 60559 states b**I, where b is the
           radix. That is, RADIX(X)**I.

Problem 2: Case (ii), which is X*2**I too large, says that the
           exception IEEE_OVERFLOW occurs. 60559 states that this also
           raises INEXACT (see 60559:7.4).

Problem 3: Case (iii), the underflow case. Again, 60559 states that
           this also raises INEXACT (see 60559:7.5).

Problem 4: Case (iv), X infinite, says that the result is X, with no
           exception raised. This is only true for finite I. When I
           is a NaN, the result is a NaN (and INVALID signals iff I is
           a signalling NaN). When I is -Infinity, the limit as I
           approaches -INF for b**I is zero, leaving us with X*0,
           and INF*0 is a NaN and INVALID signals.

Are these deviations from 60559 deliberate?

ANSWER:

No, these deviations are due to an oversight.
Edits are supplied to bring IEEE_SCALB into conformance with 60559.

EDITS to 24-007.

[472:bottom-8] 17.10 Summary of the procedures, Table 17.2,
             entry for IEEE_SCALB,
             change "2" to "RADIX (X)".
{Editor: Hyperlink to the RADIX intrinsic if reasonably possible.}

[488:2] 17.11.37 IEEE_SCALB, p1 Description,
        Change "2" to "RADIX (X)".

[488:10,11,14] Same subclause, p6 Result Value, Cases (i), (ii), (iii)
               Change "2" to "RADIX (X)", thrice.

[488:11,14] Same paragraph, Cases (ii) and (iii),
            after "shall occur" insert ", together with IEEE_INEXACT",
            twice.

[488:16] Same paragraph, Case (iv),
         after "infinite"
         insert "and I is finite or positive infinity".
         Append sentence
            "If X is infinite and I is negative infinity, the
             IEEE_INVALID exception shall signal; the result is a
             quiet NaN."
{That makes that case read:
    Case (iv): If X is infinite and I is finite or positive infinity,
               the result is the same as X; no exception signals. If X
               is infinite and I is negative infinity, the
               IEEE_INVALID exception shall signal; the result is a
               quiet NaN."}

Note: This leaves the NaN argument case unspecified, but the default
      handling is correct (raise INVALID if any is a signalling NaN,
      return a quiet NaN).

EDITS to N2258.

[564:middle] 18.10 Summary of the procedures, Table 18.2,
             entry for IEEE_SCALB,
             change "2" to "RADIX (X)".
{Editor: Hyperlink to the RADIX intrinsic if reasonably possible.}

[580:32] 18.11.37 IEEE_SCALB, p1 Description,
         Change "2" to "RADIX (X)".

[581:6,7,10] Same subclause, p6 Result Value, Cases (i), (ii), (iii):
             Change "2" to "RADIX (X)", thrice.

[581:7,10] Same paragraph, Cases (ii) and (iii),
        after "shall occur" insert ", together with IEEE_INEXACT",
        twice.

[581:13] Same paragraph, Case (iv),
         after "infinite" insert "and I is finite or positive infinity".
         Append sentence
            "If X is infinite and I is negative infinity, the
             IEEE_INVALID exception shall signal; the result is a
             quiet NaN."
{That makes that case read:
    Case (iv): If X is infinite and I is finite or positive infinity,
               the result is the same as X; no exception signals. If X
               is infinite and I is negative infinity, the
               IEEE_INVALID exception shall signal; the result is a
               quiet NaN."}

Note: This leaves the NaN argument case unspecified, but the default
      handling is correct (raise INVALID if any is a signalling NaN,
      return a quiet NaN).

SUBMITTED BY: Malcolm Cohen

HISTORY: 26-nnn   m241  Submitted

----------------------------------------------------------------------