The suggested corrigendum below specifies that the same set of WANT macros must be defined at the points in the code where the relevant headers are first included. This results in fewer combinations of interfaces and provides one sets of interfaces that is consistent and complete with respect to a given set of WANT macros.

Suggested Technical Corrigendum

After 7.1.2#4, insert:

[4a] Some standard headers define or declare identifiers contingent on whether certain macros whose names begin with _STDC_WANT_IEC_60559_ and end with _EXT_ are defined (by the user) at the point in the code where the header is first included. Within a preprocessing translation unit, the same set of such macros shall be defined for the first inclusion of all such headers.

Apr 2016 meeting

Committee Discussion

Proposed Technical Corrigendum

Page 5: At the end of 5.3, insert:

After 7.1.2#4, insert:

[4a] Some standard headers define or declare identifiers contingent on whether certain macros whose names begin with _STDC_WANT_IEC_60559_ and end with _EXT_ are defined (by the user) at the point in the code where the header is first included. Within a preprocessing translation unit, the same set of such macros shall be defined for the first inclusion of all such headers.

DR 2 Prev <— Closed —> Next DR 4, or summary at top

DR 4

DR 3 Prev <— Closed —> Next DR 5, or summary at top

Submitter: Jim Thomas et al.
Submission Date: 2016-03-19
Source: WG14
Reference Document: N2029
Subject: Part 3: Error in function name/p>

Summary

Suggested Technical Corrigendum

Apr 2016 meeting

Committee Discussion

Proposed Technical Corrigendum

Page 32: In 12.3, change “scoshdNx” to “coshdNx”.

DR 3 Prev <— Closed —> Next DR 5, or summary at top

DR 5

DR 4 Prev <— Closed —> Next DR 6, or summary at top

Submitter: Jim Thomas
Submission Date: 2016-09-10
Source: WG14
Reference Document: N2077
Subject: Part 1: Is return of same type convertFormat or copy?

Summary

Summary

This is about the issue raised by Joseph Myers in email SC22WG14.14280:

TS 18661-1 says "Whether C assignment (6.5.16) (and conversion as if by assignment) to the same format is an IEC 60559 convertFormat or copy operation is implementation-defined, even if <fenv.h> defines the macro

FE_SNANS_ALWAYS_SIGNAL (F.2.1).".

Does this apply to function return, where the return type of the function is the same as the type of the expression passed to the return statement and no wider evaluation format is in use - that is, may this act as either convertFormat or copy?  C11 F.6 clearly envisages that such a return statement may do a conversion to the same type in the case of wider evaluation formats.  But 6.8.6.4#3 only refers to conversions "If the expression has a type different from the return type of the function in which it appears".

The specification, from F.3#3, quoted above is incomplete in that it doesn’t cover function returns, which are not assignments or conversions as if by assignment. As currently written, C11 + TS18661-1 might be read to exclude the possibility of using convertFormat in this case. A statement should be added to say that the implementation has the option to apply convertFormat to the return value. The change does not break existing implementations.

The effect of convertFormat would be that signaling NaNs would signal and noncanonical representations would be canonicalized. It is extremely unlikely that a program would depend on convertFormat not being used.

Suggested Technical Corrigendum

In Clause 8, to the text for C F.3#3:

[3] Whether C assignment (6.5.16) (and conversion as if by assignment) to the same format is an IEC 60559 convertFormat or copy operation is implementation-defined, even if <fenv.h> defines the macro FE_SNANS_ALWAYS_SIGNAL (F.2.1).

append the sentence:

If the return expression of a return statement is evaluated to the floating-point format of the return type, it is implementation-defined whether a convertFormat operation is applied to the result of the return expression.”

At the end of Clause 8, add:

In F.3#3, attach a footnote to the wording:

Whether C assignment (6.5.16) (and conversion as if by assignment) to the same format is an IEC 60559 convertFormat or copy operation

where the footnote is:

*) Where the source and destination formats are the same, convertFormat operations differ from copy operations in that convertFormat operations raise the “invalid” floating-point exception on signaling NaN inputs and do not propagate non-canonical encodings.

Oct 2016 meeting

Committee Discussion

Proposed Technical Corrigendum

In Clause 8, to the text for C F.3#3:

[3] Whether C assignment (6.5.16) (and conversion as if by assignment) to the same format is an IEC 60559 convertFormat or copy operation is implementation-defined, even if <fenv.h> defines the macro FE_SNANS_ALWAYS_SIGNAL (F.2.1).

append the sentence:

If the return expression of a return statement is evaluated to the floating-point format of the return type, it is implementation-defined whether a convertFormat operation is applied to the result of the return expression.”

At the end of Clause 8, add:

In F.3#3, attach a footnote to the wording:

Whether C assignment (6.5.16) (and conversion as if by assignment) to the same format is an IEC 60559 convertFormat or copy operation

where the footnote is:

*) Where the source and destination formats are the same, convertFormat operations differ from copy operations in that convertFormat operations raise the “invalid” floating-point exception on signaling NaN inputs and do not propagate non-canonical encodings.

DR 4 Prev <— Closed —> Next DR 6, or summary at top

DR 6

DR 5 Prev <— Closed —> Next DR 7, or summary at top

Submitter: Jim Thomas
Submission Date: 2016-09-10
Source: WG14
Reference Document: N2077
Subject: Part 1: fetestexceptflag and exceptions passed to fegetexceptflag

Summary

This is about the issue raised by Joseph Myers in email SC22WG14.14328:

TS 18661-1 says, for fetestexceptflag, "The value of *flagp shall have been set by a previous call to fegetexceptflag.".

This contrasts with the C11 wording for fesetexceptflag, "The value of *flagp shall have been set by a previous call to fegetexceptflag whose second argument represented at least those floating-point exceptions represented by the argument excepts.".  So what happens if more exceptions are specified in the call to fetestexceptflag than were specified in the call to fegetexceptflag?  Then fegetexceptflag may or may not have stored any meaningful representation of the state of the extra exceptions being tested.

I think fetestexceptflag should have the same wording for this issue as fesetexceptflag: "whose second argument represented at least those floating-point exceptions represented by the argument excepts".

fesetexceptflag sets global state, typically a hardware register, whereas fetestexceptflag just reads a variable. It seems more important to avoid spurious data in the former.

Still, there’s no utility in testing spurious flag settings, and placing the same restrictions on fetestexceptflag as on fesetexceptflag might be less error prone.

Suggested Technical Corrigendum

In 15.2, in the new text for C 7.6.2.4a#2, change:

The value of *flagp shall have been set by a previous call to fegetexceptflag.

to:

The value of *flagp shall have been set by a previous call to fegetexceptflag whose second argument represented at least those floating-point exceptions represented by the argument excepts.

Oct 2016 meeting

Committee Discussion

Proposed Technical Corrigendum

In 15.2, in the new text for C 7.6.2.4a#2, change:

The value of *flagp shall have been set by a previous call to fegetexceptflag.

to:

The value of *flagp shall have been set by a previous call to fegetexceptflag whose second argument represented at least those floating-point exceptions represented by the argument excepts.

DR 5 Prev <— Closed —> Next DR 7, or summary at top

DR 7

DR 6 Prev <— Closed —> Next DR 8, or summary at top

Submitter: Jim Thomas
Submission Date: 2016-09-10
Source: WG14
Reference Document: N2077
Subject: Part 1: Editorial changes

Summary

Summary

In CFP email, Fred Tydeman noted:

Searching for "infinite precision" in part 1, most of them have "(as if) to" before it. Except, ffma, ffmal, dfmal which is missing the "(as if)".

Right. In particular, all the functions that round result to narrower type have “(as if)”, except for the fma family.

Suggested Technical Corrigendum

In 14.5, in the new text for C 7.12.13a.5#2, insert “(as if)” before “to infinite precision”.

Oct 2016 meeting

Committee Discussion

Proposed Technical Corrigendum

In 14.5, in the new text for C 7.12.13a.5#2, insert “(as if)” before “to infinite precision”.

DR 6 Prev <— Closed —> Next DR 8, or summary at top

DR 8

DR 7 Prev <— Closed —> Next DR 9, or summary at top

Submitter: Jim Thomas
Submission Date: 2016-09-10
Source: WG14
Reference Document: N2077
Subject: Part 2: Editorial clarification about number digits in the coefficient

Summary

Summary

In 12.5, n is defined to be “the number of digits in the coefficient c”, where the decimal floating-point argument is represented by the triple (s, c, q). The intention is that n is the number of digits in the coefficient of the particular argument, i.e., the number of significant digits, not the maximum number of digits in the coefficient for the type. This might be misread, particularly since 5.2.4.2.2a says

¾    number of digits in the coefficient

DEC32_MANT_DIG                 7

DEC64_MANT_DIG                 16

DEC128_MANT_DIG                34

This part of 5.2.4.2.2a is in the context of characterizing the type, so clearly refers to the type and not any particular representation.

Suggested Technical Corrigendum

In 12.5, change:

where n is the number of digits in the coefficient c

to:

                  where n is the number of significant digits in the coefficient c

Oct 2016 meeting

Committee Discussion

Proposed Technical Corrigendum

In 12.5, change:

where n is the number of digits in the coefficient c

to:

                  where n is the number of significant digits in the coefficient c

DR 7 Prev <— Closed —> Next DR 9, or summary at top

DR 9

DR 8 Prev <— Closed —> Next DR 10, or summary at top

Submitter: Jim Thomas
Submission Date: 2016-09-10
Source: WG14
Reference Document: N2077
Subject: Part 2,3: Missing specification for usual arithmetic conversions, tgmath

Summary

This is about the issue raised by Joseph Myers in email SC22WG14.14282:

C11 specifies that the usual arithmetic conversions on the pair of types (long double, double) produces a result of type long double.

Suppose long double and double have the same set of values.  TS 18661-3 rewrites the rules for usual arithmetic conversions so that the case "if both operands are floating types and the sets of values of their corresponding real types are equivalent" prefers interchange types to standard types to extended types.  But this leaves the case of (long double, double) unspecified as to which type is chosen, unlike in C11, as those are both standard types.

I think this is a defect in TS 18661-3, and it should say that if both are standard types with the same set of values then long double is preferred to double which is preferred to float, as in C11.

A similar issue could arise if two of the extended types have equivalent sets of values.  I'm not aware of anything to prohibit that, although it seems less likely in practice.  I think the natural fix would be to say that _Float128x is preferred to _Float64x which is preferred to _Float32x.

I think such an issue would also arise for <tgmath.h> (if _Float64x and _Float128x have the same set of values, the choice doesn't seem to be specified).  It also seems possible for the <tgmath.h> rules for purely floating-point arguments to produce a different result from the usual arithmetic conversions (consider the case where _Float32x is wider than long double, and <tgmath.h> chooses long double), and since rules that are the same in most cases but subtly different in obscure cases tend to be confusing, I wonder if it might be better to specify much simpler rules for <tgmath.h>: take the type resulting from the usual arithmetic conversions[*], where integer arguments are replaced by _Decimal64 if there are any decimal arguments and double otherwise.  (That's different from the present rules for e.g. (_Float32x, int), but it's a lot simpler, and seems unlikely in practice to choose a type with a different set of values from the present choice.)

[*] Meaningful for more than two arguments as long as the usual arithmetic conversions are commutative and associative as an operation on pairs of types.

Though substantive, the suggested change to the usual arithmetic conversions is consistent with the intention in TS 18661-3 to specify all the cases (except where neither format is a subset of the other and the formats are not the same). The missing cases were an oversight. The suggested preferences of long double over double over float and _Float128x over _Float64x over _Float32x are the obvious choices.

Joseph Myers notes that the <tgmath.h> specification is incomplete in the same way as the usual arithmetic conversions. He argues for simplifying the specification by referring to the usual arithmetic conversions specification, rather than mostly repeating it, as the current specification does. The suggested Technical Corrigendum below follows this new approach. Though a substantive change to TS 18661-3, the effects on implementations and users are expected to be minimal – worth the simplification.

The suggested Technical Corrigendum below also restores footnote number 62, which is lost in the current TS 18661-3.

Suggested Technical Corrigendum

In clause 8, change the replacement text for 6.3.1.8#1:

If one operand has decimal floating type, the other operand shall not have standard floating type, binary floating type, complex type, or imaginary type.

If both operands have floating types and neither of the sets of values of their corresponding real types is a subset of (or equivalent to) the other, the behavior is undefined.

Otherwise, if both operands are floating types and the sets of values of their corresponding real types are equivalent, then the following rules are applied:

If both operands have the same corresponding real type, no further conversion is needed.

Otherwise, if the corresponding real type of either operand is an interchange floating type, the other operand is converted, without change of type domain, to a type whose corresponding real type is that same interchange floating type.

Otherwise, if the corresponding real type of either operand is a standard floating type, the other operand is converted, without change of type domain, to a type whose corresponding real type is that same standard floating type.

Otherwise, if both operands have floating types, the operand, whose set of values of its corresponding real type is a (proper) subset of the set of values of the corresponding real type of the other operand, is converted, without change of type domain, to a type with the corresponding real type of that other operand.

Otherwise, if one operand has a floating type, the other operand is converted to the corresponding real type of the operand of floating type.

Otherwise, the integer promotions are performed on both operands. Then the following rules are applied to the promoted operands:

. . .

to:

If one operand has decimal floating type, the other operand shall not have standard floating type, binary floating type, complex type, or imaginary type.

If both operands have floating types and neither of the sets of values of their corresponding real types is a subset of (or equivalent to) the other, the behavior is undefined.

If both operands have the same corresponding real type, no further conversion is needed. 

Otherwise, if both operands are floating types and the sets of values of their corresponding real types are equivalent, then the following rules are applied: 

If the corresponding real type of either operand is an interchange floating type, the other operand is converted, without change of type domain, to a type whose corresponding real type is that same interchange floating type.

Otherwise, if the corresponding real type of either operand is long double, the other operand is converted, without change of type domain, to a type whose corresponding real type is long double.

Otherwise, if the corresponding real type of either operand is double, the other operand is converted, without change of type domain, to a type whose corresponding real type is double.

(All cases where float might have the same format as another type are covered above.)

Otherwise, if the corresponding real type of either operand is _Float128x or _Decimal128x, the other operand is converted, without change of type domain, to a type whose corresponding real type is _Float128x or _Decimal128x, respectively.

Otherwise, if the corresponding real type of either operand is _Float64x or _Decimal64x, the other operand is converted, without change of type domain, to a type whose corresponding real type is _Float64x or _Decimal64x, respectively.

Otherwise, if both operands have floating types, the operand, whose set of values of its corresponding real type is a (proper) subset of the set of values of the corresponding real type of the other operand, is converted, without change of type domain62), to a type with the corresponding real type of that other operand.

Otherwise, if one operand has a floating type, the other operand is converted to the corresponding real type of the operand of floating type.

Otherwise, the integer promotions are performed on both operands. Then the following rules are applied to the promoted operands:

. . .

In clause 15, replace:

In 7.25#3c, replace the bullets:

… bullets …

with:

—  If two arguments have floating types and neither of the sets of values of their corresponding real types is a subset of (or equivalent to) the other, the behavior is undefined.

—  If any arguments for generic parameters have type _DecimalM where M ≥ 64 or _DecimalNx where N ≥ 32, the type determined is the widest of the types of these arguments. If _DecimalM and _DecimalNx are both widest types (with equivalent sets of values) of these arguments, the type determined is _DecimalM.

—  Otherwise, if any argument for generic parameters is of integer type and another argument for generic parameters has type _Decimal32, the type determined is _Decimal64.

—  Otherwise, if any argument for generic parameters has type _Decimal32, the type determined is _Decimal32.

—  Otherwise, if the corresponding real type of any argument for generic parameters has type long double, _FloatM where M ≥ 128, or _FloatNx where N ≥ 64, the type determined is the widest of the corresponding real types of these arguments. If _FloatM and either long double or _FloatNx are both widest corresponding real types (with equivalent sets of values) of these arguments, the type determined is _FloatM. Otherwise, if long double and _FloatNx are both widest corresponding real types (with equivalent sets of values) of these arguments, the type determined is long double.

—  Otherwise, if the corresponding real type of any argument for generic parameters has type double, _Float64, or _Float32x, the type determined is the widest of the corresponding real types of these arguments. If _Float64 and either double or _Float32x are both widest corresponding real types (with equivalent sets of values) of these arguments, the type determined is _Float64. Otherwise, if double and _Float32x are both widest corresponding real types (with equivalent sets of values) of these arguments, the type determined is double.

—  Otherwise, if any argument for generic parameters is of integer type, the type determined is double.

—  Otherwise, if the corresponding real type of any argument for generic parameters has type _Float32, the type determined is _Float32.

—  Otherwise, the type determined is float.

In the second bullet 7.25#3c, attach a footnote to the wording:

the type determined is the widest

where the footnote is:

*) The term widest here refers to a type whose set of values is a superset of (or equivalent to) the sets of values of the other types.

with:

In 7.25#3c, replace the first sentence and bullets:

[3c] Except for the macros for functions that round result to a narrower type (7.12.13a), use of a type-generic macro invokes a function whose generic parameters have the corresponding real type determined by the corresponding real types of the arguments as follows:

—    First, if any argument for generic parameters has type _Decimal128, the type determined is _Decimal128.

—    Otherwise, if any argument for generic parameters has type _Decimal64, or if any argument for generic parameters is of integer type and another argument for generic parameters has type _Decimal32, the type determined is _Decimal64.

—    Otherwise, if any argument for generic parameters has type _Decimal32, the type determined is _Decimal32.

—    Otherwise, if the corresponding real type of any argument for generic parameters is long double, the type determined is long double.

—    Otherwise, if the corresponding real type of any argument for generic parameters is double or is of integer type, the type determined is double.

—    Otherwise, if any argument for generic parameters is of integer type, the type determined is double.

—    Otherwise, the type determined is float.

with:

[3c] Except for the macros for functions that round result to a narrower type (7.12.13a), use of a type-generic macro invokes a function whose generic parameters have the corresponding real type determined by the types of the arguments for the generic parameters as follows:

—  Arguments of integer type are regarded as having type _Decimal64 if any argument has decimal floating type, and as having type double otherwise.

—  If the function has exactly one generic parameter, the type determined is the corresponding real type of the argument for the generic parameter.

—  If the function has exactly two generic parameters, the type determined is the corresponding real type determined by the usual arithmetic conversions (6.3.1.8) applied to the arguments for the generic parameters.

—  If the function has more than two generic parameters, the type determined is the corresponding real type determined by repeatedly applying the usual arithmetic conversions, first to the first two arguments for generic parameters, then to that result type and the next argument for a generic parameter, and so forth until the usual arithmetic conversions have been applied to the last argument for a generic parameter.

Oct 2016 meeting

Committee Discussion

Apr 2017 meeting

Committee Discussion

The TC in DR 501 includes two changes to TS 18661-3, one for the usual arithmetic conversions, the other for type-generic math. The first change fills in missing conversions for new types in TS 18661-3. The second change simplifies type-generic math by referencing the usual arithmetic conversions, and thereby also fills in missing type-generic math rules for arguments of the new types.

This is a proposal for an alternative change to type-generic math. The original change was proposed for TS 18661-3, where the new types where introduced. However, the change can be made in TS 18661-2, where it is easier to understand and leads to a simplification in TS 18661-3.

Proposed Technical Corrigendum

In TS 18662-3

In clause 8, change the replacement text for 6.3.1.8#1:

If one operand has decimal floating type, the other operand shall not have standard floating type, binary floating type, complex type, or imaginary type.

If both operands have floating types and neither of the sets of values of their corresponding real types is a subset of (or equivalent to) the other, the behavior is undefined.

Otherwise, if both operands are floating types and the sets of values of their corresponding real types are equivalent, then the following rules are applied:

If both operands have the same corresponding real type, no further conversion is needed.

Otherwise, if the corresponding real type of either operand is an interchange floating type, the other operand is converted, without change of type domain, to a type whose corresponding real type is that same interchange floating type.

Otherwise, if the corresponding real type of either operand is a standard floating type, the other operand is converted, without change of type domain, to a type whose corresponding real type is that same standard floating type.

Otherwise, if both operands have floating types, the operand, whose set of values of its corresponding real type is a (proper) subset of the set of values of the corresponding real type of the other operand, is converted, without change of type domain, to a type with the corresponding real type of that other operand.

Otherwise, if one operand has a floating type, the other operand is converted to the corresponding real type of the operand of floating type.

Otherwise, the integer promotions are performed on both operands. Then the following rules are applied to the promoted operands:

. . .

to:

If one operand has decimal floating type, the other operand shall not have standard floating type, binary floating type, complex type, or imaginary type.

If both operands have floating types and neither of the sets of values of their corresponding real types is a subset of (or equivalent to) the other, the behavior is undefined.

If both operands have the same corresponding real type, no further conversion is needed. 

Otherwise, if both operands are floating types and the sets of values of their corresponding real types are equivalent, then the following rules are applied: 

If the corresponding real type of either operand is an interchange floating type, the other operand is converted, without change of type domain, to a type whose corresponding real type is that same interchange floating type.

Otherwise, if the corresponding real type of either operand is long double, the other operand is converted, without change of type domain, to a type whose corresponding real type is long double.

Otherwise, if the corresponding real type of either operand is double, the other operand is converted, without change of type domain, to a type whose corresponding real type is double.

(All cases where float might have the same format as another type are covered above.)

Otherwise, if the corresponding real type of either operand is _Float128x or _Decimal128x, the other operand is converted, without change of type domain, to a type whose corresponding real type is _Float128x or _Decimal128x, respectively.

Otherwise, if the corresponding real type of either operand is _Float64x or _Decimal64x, the other operand is converted, without change of type domain, to a type whose corresponding real type is _Float64x or _Decimal64x, respectively.

Otherwise, if both operands have floating types, the operand, whose set of values of its corresponding real type is a (proper) subset of the set of values of the corresponding real type of the other operand, is converted, without change of type domain62), to a type with the corresponding real type of that other operand.

Otherwise, if one operand has a floating type, the other operand is converted to the corresponding real type of the operand of floating type.

Otherwise, the integer promotions are performed on both operands. Then the following rules are applied to the promoted operands:

. . .

In TS 18661-2

In 12.9, change the introduced [3c] from:

[3c] Except for the macros for functions that round result to a narrower type (7.12.13a), use of a type-generic macro invokes a function whose generic parameters have the corresponding real type determined by the corresponding real types of the arguments as follows:

—    First, if any argument for generic parameters has type _Decimal128, the type determined is _Decimal128.

—    Otherwise, if any argument for generic parameters has type _Decimal64, or if any argument for generic parameters is of integer type and another argument for generic parameters has type _Decimal32, the type determined is _Decimal64.

—    Otherwise, if any argument for generic parameters has type _Decimal32, the type determined is _Decimal32.

—    Otherwise, if the corresponding real type of any argument for generic parameters is long double, the type determined is long double.

—    Otherwise, if the corresponding real type of any argument for generic parameters is double or is of integer type, the type determined is double.

—    Otherwise, if any argument for generic parameters is of integer type, the type determined is double.

—    Otherwise, the type determined is float.

to:

[3c] Except for the macros for functions that round result to a narrower type (7.12.13a), use of a type-generic macro invokes a function whose generic parameters have the corresponding real type determined by the types of the arguments for the generic parameters as follows:

—    Arguments of integer type are regarded as having type _Decimal64 if any argument has decimal floating type, and as having type double otherwise.

—     If the function has exactly one generic parameter, the type determined is the corresponding real type of the argument for the generic parameter.

—    If the function has exactly two generic parameters, the type determined is the corresponding real type determined by the usual arithmetic conversions (6.3.1.8) applied to the arguments for the generic parameters.

—    If the function has more than two generic parameters, the type determined is the corresponding real type determined by repeatedly applying the usual arithmetic conversions, first to the first two arguments for generic parameters, then to that result type and the next argument for a generic parameter, and so forth until the usual arithmetic conversions have been applied to the last argument for a generic parameter.

DR 8 Prev <— Closed —> Next DR 10, or summary at top

DR 10

DR 9 Prev <— Closed —> Next DR 11, or summary at top

Submitter: Jim Thomas
Submission Date: 2016-09-10
Source: WG14
Reference Document: N2077
Subject: Part 1: wrong type for fesetmode parameter

Summary

This is about the issue raised by Joseph Myers in email SC22WG14.14358:

TS 18661-1 gives the declaration of fesetmode as:

int fesetmode(const fenv_t *modep);

The argument should be of type const femode_t *, not const fenv_t *.

--

This was an editorial cut-and-past error. The Description says the argument modep shall point to an objet set by a call to fegetmode, which sets objects of type femode_t. It’s unlikely the function would be implemented with the erroneous type.

Suggested Technical Corrigendum

In 15.3, in the new text for C 7.6.3.1a#1, change:

          int fesetmode(const fenv_t *modep);

to:

          int fesetmode(const femode_t *modep);

Oct 2016 meeting

Committee Discussion

Proposed Technical Corrigendum

In 15.3, in the new text for C 7.6.3.1a#1, change:

          int fesetmode(const fenv_t *modep);

to:

          int fesetmode(const femode_t *modep);

DR 9 Prev <— Closed —> Next DR 11, or summary at top

DR 11

DR 10 Prev <— Closed —> Next DR 12, or summary at top

Submitter: Jim Thomas
Submission Date: 2016-09-10
Source: WG14
Reference Document: N2077
Subject: Part 2: a-style formatting not IEC 60559 conformant

Summary

The a-style formatting specified in subclause 12.5 of TS 18661-2 is not an IEC 60559 conversion for cases where the formatting precision is less than the length of the coefficient of the input. The specification entails an intermediate rounding to the floating type of the input, which might overflow resulting in a character sequence representation of infinity. IEC 60559 conversions to character sequences do not overflow, unless the language over-restricts the exponent range for character sequence output, which C does not.

Another undesirable aspect of the current specification is that in certain cases it produces results with more precision than given by a width modifier.

Here are some examples, showing the result of the intermediate conversion, with different behaviors for the current spec (“old”) and the spec in the suggested Technical Corrigendum below (“new”):

For _Decimal32 input x with representation (1, 9512345, 90) and specifier ...

%.3Ha

old:                           x                ->             (1, 9510000, 90)                ->             9.510000e96

new:                         x                ->             (1, 951, 94)                            ->             9.51e96

%.2Ha

old:                           x                ->             (1, 9500000, 90)                ->             9.500000e96

new:                         x                ->             (1, 95, 95)                               ->             9.5e96

%.1Ha

old:                           x                ->             Inf                                                ->             inf

new:                         x                ->             (1, 1, 97)                                  ->             1e97

Here’s another example:

For _Decimal32 input x with representation (1, 9512345, 86) and specifier ...

%.2Ha

old:                           x                ->             (1, 950, 90)                            ->             9.50e92

new:                         x                ->             (1, 95, 91)                               ->             9.5e92

The examples use a to-nearest rounding.

As the examples illustrate, the problematic cases for the current “old” spec occur because of the exponent range limitation of the format used for the intermediate conversion.

The suggested Technical Corrigendum below specifies formatting that is IEC 60559 conformant and which honors a width modifier. It does not change the numerical value of the result, except in overflow cases.

Suggested Technical Corrigendum

In 12.5, in the addition to 7.21.6.1#8 and 7.29.2.1#8, under a,A conversion specifiers, change:

If the precision is present (in the conversion specification) and is zero or at least as large as the precision p (5.2.4.2.2) of the decimal floating type, the conversion is as if the precision were missing. If the precision is present (and nonzero) and less than the precision p of the decimal floating type, the conversion first obtains an intermediate result by rounding the input in the type, according to the current rounding direction for decimal floating-point operations, to the number of digits specified by the precision, then converts the intermediate result as if the precision were missing. The length of the coefficient of the intermediate result is the smallest number, at least as large as the formatting precision, for which the quantum exponent is within the quantum exponent range of the type (see 5.2.4.2.2a). The intermediate rounding may overflow.

to:

If the precision P is present (in the conversion specification) and is zero or at least as large as the precision p (5.2.4.2.2) of the decimal floating type, the conversion is as if the precision were missing. If the precision P is present (and nonzero) and less than the precision p of the decimal floating type, the conversion first obtains an intermediate result as follows, where n is the number of significant digits in the coefficient:

If n <= P, set the intermediate result to the input.

If n > P, round the input value, according to the current rounding direction for decimal floating-point operations, to P decimal digits, with unbounded exponent range, representing the result with a P-digit integer coefficient when in the form (s, c, q).

Convert the intermediate result in the manner described above for the case where the precision is missing.

In 12.5, in the addition to 7.21.6.1#8 and 7.29.2.1#8, in EXAMPLE 3, change the results:

9.54321e+93

9.5432e+93

9.543e+93

9.540e+93

9.500e+93

1.0000e+94

inf

to:

9.54321e+93

9.5432e+93

9.543e+93

9.54e+93

9.5e+93

1e+94

1e+97

Oct 2016 meeting

Committee Discussion

Apr 2017 meeting

Committee Discussion

However, the committee is concerned that %a behavior differs from binary floating point and more review is needed. In particular, there were concerns that for the decimal floating point types now the %a format specifier given with a precision is the total number of significant digits, not the number of digits after the decimal point as it has been for other data types.

Proposed Technical Corrigendum

In 12.5, in the addition to 7.21.6.1#8 and 7.29.2.1#8, under a,A conversion specifiers, change:

If the precision is present (in the conversion specification) and is zero or at least as large as the precision p (5.2.4.2.2) of the decimal floating type, the conversion is as if the precision were missing. If the precision is present (and nonzero) and less than the precision p of the decimal floating type, the conversion first obtains an intermediate result by rounding the input in the type, according to the current rounding direction for decimal floating-point operations, to the number of digits specified by the precision, then converts the intermediate result as if the precision were missing. The length of the coefficient of the intermediate result is the smallest number, at least as large as the formatting precision, for which the quantum exponent is within the quantum exponent range of the type (see 5.2.4.2.2a). The intermediate rounding may overflow.

to:

If the precision P is present (in the conversion specification) and is zero or at least as large as the precision p (5.2.4.2.2) of the decimal floating type, the conversion is as if the precision were missing. If the precision P is present (and nonzero) and less than the precision p of the decimal floating type, the conversion first obtains an intermediate result as follows, where n is the number of significant digits in the coefficient:

If n <= P, set the intermediate result to the input.

If n > P, round the input value, according to the current rounding direction for decimal floating-point operations, to P decimal digits, with unbounded exponent range, representing the result with a P-digit integer coefficient when in the form (s, c, q).

Convert the intermediate result in the manner described above for the case where the precision is missing.

In 12.5, in the addition to 7.21.6.1#8 and 7.29.2.1#8, in EXAMPLE 3, change the results:

9.54321e+93

9.5432e+93

9.543e+93

9.540e+93

9.500e+93

1.0000e+94

inf

to:

9.54321e+93

9.5432e+93

9.543e+93

9.54e+93

9.5e+93

1e+94

1e+97

Add, as a new EXAMPLE,

#include <stdio.h>

int main(void) {
  _Decimal32 x = 9512345e90df;
  _Decimal32 x2 = 9512345e86df;
  
  printf("%.3Ha\n", x); // New expected output: 9.51e96
  printf("%.2Ha\n", x); // New expected output: 9.5e96
  printf("%.1Ha\n", x); // New expected output: 1e97
  printf("%.2Ha\n", x2); // New expected output: 9.5e92
  
  return 0;
}
  

DR 10 Prev <— Closed —> Next DR 12, or summary at top

DR 12

DR 11 Prev <— Closed —> Next DR 14, or summary at top

Submitter: Jim Thomas
Submission Date: 2017-03-04
Source: WG14
Reference Document: N2125
Subject: P1: Zero payloads and set payload function

Summary

This is about an issue raised by Joseph Myers in SC22WG14.14450:

The specification for setpayload (and likewise setpayloadsig) says "If pl is not a positive floating-point integer representing a valid payload, *res is set to positive zero."

Does "positive" as applied to "floating-point integer" here mean "with sign bit 0" (the list of definitions in IEEE 754 doesn't include "positive")?  In the preferred encodings for binary interchange formats, 0 is a valid payload for quiet NaNs.  So should +0.0 as an argument to setpayload result in a quiet NaN with payload 0, while -0.0 results in *res being set to +0.0 because -0.0 isn't positive (and for setpayloadsig, both result in *res set to +0.0 because a payload for a signaling NaN has to be nonzero to avoid all mantissa bits being zero)?

A “positive floating-point integer” is a positive integer in the floating-point format, hence it is greater than zero. So, the current specification for setpayload and setpayloadsig is flawed in that it doesn’t allow setting the payload to zero.

A more basic problem is that TS 18661-1 assumes IEC 60559 interprets payloads as integers. This is true for decimal formats. IEC 60559 says:

The payload corresponds to the significand of finite numbers, interpreted as an integer with a maximum value of 10^(3×J)−1, …

The significand c interpreted as an integer is assumed throughout to be non-negative, while the s field in (s, q, c) provides the sign. For decimal, interpreting the bits in the encodings allows the two encoding schemes to have the same payloads and the payloads to fit conceptually with their encoding schemes. 

However, for binary formats, IEC 60559 says:

For binary formats, the payload is encoded in the p−2 least significant bits of the trailing significand field.

Nowhere does it actually interpret the payload for binary formats as an integer.

However, the payload for binary formats has a natural interpretation as an unsigned integer, so it is reasonable for TS 1866-1 to interpret payloads (for binary and decimal formats) as such.

The suggested Technical Corrigendum below addresses these problems.

Suggested Technical Corrigendum

In 14.10, replace the first sentence:

IEC 60559 defines the payload of a NaN to be a certain part of the NaN’s significand interpreted as an integer.

with:

IEC 60559 defines the payload of a NaN to be a certain part of the NaN’s significand. The payload can be interpreted as an unsigned integer.

In 14.10, in the new C subclause F.10.13, replace:

IEC 60559 defines the payload of a quiet or signaling NaN as an integer value encoded in the significand.

with:

IEC 60559 defines the payload of a quiet or signaling NaN as information encoded in part of the NaN significand. The payload can be interpreted as an unsigned integer.

In 14.10, in the new C subclauses F.10.13.2#2 and F.10.13.3#2, change:

If pl is not a positive floating-point integer representing a valid payload, *res is set to positive zero.

to:

If pl is not a floating-point integer representing a valid payload, *res is set to positive zero.

Committee Discussion

The committee agrees that this is a defect and accepts the Suggested Technical Corrigendum

Apr 2018 meeting

Committee Discussion

Discussion of this proposed change on the reflector prompted a revised change from N2202 which was then again revised in N2213.

The committee accepted these changes.

Subsequently, an additional small change was provided and discussed and accepted.

DR 11 Prev <— Closed —> Next DR 14, or summary at top

DR 13

DR 22 Prev <— Open —> Next DR 16, or summary at top

Submitter: Jim Thomas
Submission Date: 2017-03-04
Source: WG14
Reference Document: N2125
Subject: P3: Type-generic macros for functions that round result to narrower type

Summary

This is about an issue raised by Joseph Myers in SC22WG14.14561:

TS 18661-1 and -2 define type-generic macros for the functions that round

result to a narrower type.  In part 1 these are, for example, fadd and

dadd for addition; in part 2, for example, d32add and d64add.

Part 3 does not seem to make any changes or additions to those macros, and

consequences of that seem nonobvious.  It defines new functions for the

new types: fMaddfN, fMaddfNx, fMxaddfN, fMxaddfNx (where M < N, or M <= N

in the fMaddfNx case), and likewise for decimal types.  But the

type-generic macros remain as defined in 7.25#6a after the changes from

parts 1 and 2 are applied (part 3 does not contain the string "6a").

That is, it's valid to pass the _FloatN and _FloatNx types to the fadd and

dadd macros, and valid to pass the new _DecimalN and _DecimalNx types from

part 3 to the d32add and d64add types.

(a) 7.25#6a says "If the macro prefix is d32 or d64, use of an argument of

standard floating type results in undefined behavior.".  Other places get

amended in part 3 to say "floating type of radix 2" in addition to

"standard floating type".  But it appears it fails to make it undefined to

pass _FloatN or _FloatNx arguments to d32add, d64add etc. type-generic

macros - although clearly it should be undefined.

(b) Passing _Decimal128 to d32add would result in the d32addd128 function

being called, as expected.  But say you pass a _Decimal128x argument.  A

function d32addd128x exists but the specification would seem to result in

d32addd64 being called, which seems unintuitive.  Similar issues apply

with _FloatN and _FloatNx types - calling fadd on them would always call

the fadd function not faddl.  (But in that case there *are* no functions

defined that take _FloatN / _FloatNx arguments and return float or double.  

So the right thing to do is less obvious.)

The following addresses these issues by filling in the missing specification in part 3.

Suggested Technical Corrigendum

In clause 15, after the change to 7.25#6, add:

Change 7.25#6a from:

[6a] The functions that round result to a narrower type have type-generic macros whose names are obtained by omitting any suffix from the function names. Thus, the macros with f or d prefix are:

fadd              fmul                 ffma

dadd              dmul                 dfma

fsub              fdiv                 fsqrt

dsub              ddiv                 dsqrt

and the macros with d32 or d64 prefix are:

d32add            d32mul               d32fma

d64add            d64mul               d64fma

d32sub            d32div               d32sqrt

d64sub            d64div               d64sqrt

All arguments are generic. If any argument is not real, use of the macro results in undefined behavior. If the macro prefix is f or d, use of an argument of decimal floating type results in undefined behavior. If the macro prefix is d32 or d64, use of an argument of standard floating type results in undefined behavior. The function invoked is determined as follows:

—    If any argument has type _Decimal128, or if the macro prefix is d64, the function invoked has the name of the macro, with a d128 suffix.

—    Otherwise, if the macro prefix is d32, the function invoked has the name of the macro, with a d64 suffix.

—    Otherwise, if any argument has type long double, or if the macro prefix is d, the function invoked has the name of the macro, with an l suffix.

—    Otherwise, the function invoked has the name of the macro (with no suffix).

 to:

[6a] The functions that round result to a narrower type have type-generic macros whose names are obtained by omitting any suffix from the function names. Thus, the macros with f or d prefix are:

fadd              fmul                 ffma

dadd              dmul                 dfma

fsub              fdiv                 fsqrt

dsub              ddiv                 dsqrt

and the macros with fM, fMx, dM, or dMx prefix are:

fMadd             fMxmul              dMfma

fMsub             fMxdiv              dMsqrt

fMmul             fMxfma              dMxadd

fMdiv             fMxsqrt             dMxsub

fMfma             dMadd               dMxmul

fMsqrt            dMsub               dMxdiv

fMxadd            dMmul               dMxfma

fMxsub            dMdiv               dMxsqrt

All arguments are generic. If any argument is not real, use of the macro results in undefined behavior. If the macro prefix is f, d, fM, or fMx, use of an argument of decimal floating type results in undefined behavior. If the macro prefix is dM or dMx, use of an argument of standard or binary floating type results in undefined behavior. The function invoked is determined as follows:

—    Arguments that have integer type are regarded as having type _Decimal64 if any argument has decimal floating type, and as having type double otherwise.

—    The unsuffixed name of the function is the name of the macro, and its suffix, if any, corresponds to the parameter type which may be any type with at least the range and precision of the argument types.

In clause 15, at the end of the text appended to the table in 7.25#7, further append:

f32xadd(d, f32x)                any f32xaddfN or f32xaddfNx such that N > 32 and the suffix type, _FloatN or _FloatNx, is at least as wide as double and _Float32x

Committee Discussion

The committee agrees that this is a defect and accepts the Suggested Technical Corrigendum

Apr 2018 meeting

Committee Discussion

After extensive discussion on the mailing list several documents were proposed with new and revised change suggestions. The following revised proposed change is largely drawn from N2213 with further changes reviewed at the meeting.

Proposed Change

In clause 15, after the change to 7.25#6, add:

Change 7.25#6a from:

[6a] The functions that round result to a narrower type have type-generic macros whose names are obtained by omitting any suffix from the function names. Thus, the macros with f or d prefix are:

fadd              fmul                 ffma

dadd              dmul                 dfma

fsub              fdiv                 fsqrt

dsub              ddiv                 dsqrt

and the macros with d32 or d64 prefix are:

d32add            d32mul               d32fma

d64add            d64mul               d64fma

d32sub            d32div               d32sqrt

d64sub            d64div               d64sqrt

All arguments are generic. If any argument is not real, use of the macro results in undefined behavior. If the macro prefix is f or d, use of an argument of decimal floating type results in undefined behavior. If the macro prefix is d32 or d64, use of an argument of standard floating type results in undefined behavior. The function invoked is determined as follows:

—    If any argument has type _Decimal128, or if the macro prefix is d64, the function invoked has the name of the macro, with a d128 suffix.

—    Otherwise, if the macro prefix is d32, the function invoked has the name of the macro, with a d64 suffix.

—    Otherwise, if any argument has type long double, or if the macro prefix is d, the function invoked has the name of the macro, with an l suffix.

—    Otherwise, the function invoked has the name of the macro (with no suffix).

 to:

[6a] The functions that round result to a narrower type have type-generic macros whose names are obtained by omitting any suffix from the function names. Thus, the macros with f or d prefix are:

fadd              fmul                 ffma

dadd              dmul                 dfma

fsub              fdiv                 fsqrt

dsub              ddiv                 dsqrt

and the macros with fM, fMx, dM, or dMx prefix are:

fMadd             fMxmul              dMfma

fMsub             fMxdiv              dMsqrt

fMmul             fMxfma              dMxadd

fMdiv             fMxsqrt             dMxsub

fMfma             dMadd               dMxmul

fMsqrt            dMsub               dMxdiv

fMxadd            dMmul               dMxfma

fMxsub            dMdiv               dMxsqrt

All arguments are generic. If any argument is not real, use of the macro results in undefined behavior. If the macro prefix is f or d, use of an argument of interchange or extended floating type results in undefined behavior. If the macro prefix is fM, or fMx, use of an argument of standard or decimal floating type results in undefined behavior. If the macro prefix is dM or dMx, use of an argument of standard or binary floating type results in undefined behavior.  The function invoked is determined as follows:

—  Arguments that have integer type are regarded as having type double if the macro prefix is f or d, as having type _Float64 if the macro prefix is fM or fMx, and as having type _Decimal64 if the macro prefix is dM or dMx.

—  If the function has exactly one generic parameter, the type determined is the type of the argument.

—  If the function has exactly two generic parameters, the type determined is the type determined by the usual arithmetic conversions (6.3.1.8) applied to the arguments.

—  If the function has three generic parameters, the type determined is the type determined by applying the usual arithmetic conversions twice, first to the first two arguments, then to that result type and the third argument.

—  If no function with the given prefix has the parameter type determined above, the parameter type is determined from the prefix as follows:

In clause 15, at the end of the text appended to the table in 7.25#7, further append:

fsub(d, ld)          fsubl

f32add(f64x, f64)     f32addf64x

d32xsqrt(n)          d32xsqrtd64

f32mul(f128, f32x)    f32mulf128 if _Float128 is at least as wide as _Float32x, or f32mulf32x if _Float32x is wider than _Float128

f32fma(f32x, n, f32x)  f32fmaf64 if _Float64 is at least as wide as _Float32x, or f32fmaf32x if _Float32x is wider than _Float64

ddiv(ld, f128)                 undefined

f32fma(f64, d, f64)      undefined

fmul(dc, d)                        undefined

f32add(f32, f32)             f32addf64(f32, f32)

f32xsqrt(f32)                   f32xsqrtf64x(f32) if _Float64x is

supported, else f32xsqrtf64

f64div(f32x, f32x)        f64divf128(f32x, f32x)

DR 22 Prev <— Open —> Next DR 16, or summary at top

DR 14

DR 12 Prev <— Closed —> Next DR 15, or summary at top

Submitter: Jim Thomas
Submission Date: 2017-03-04
Source: WG14
Reference Document: N2125
Subject: P2: Effect of %a vs %A conversion specifiers

Summary

The specification in TS 18661-2 for a,A conversion specifiers for decimal describes the behavior in terms of  f and e formatting. The intention was that the A conversion specifier would have the effects of F and E formatting. The following Technical Corrigendum corrects this oversight, using wording similar to that in C11 for the g,G conversion specifiers.

Suggested Technical Corrigendum

In 12.5, in the text added to 7.21.6.1#8 and 7.29.2.1#8, under a,A conversion specifiers, replace the bullets:

—    if −(n+5) ≤ q ≤ 0, use style f formatting with formatting precision equal to −q,

—    otherwise, use style e formatting with …

with:

—    if −(n+5) ≤ q ≤ 0, use style f (or style F in the case of an A conversion specifier) with formatting precision equal to −q,

—    otherwise, use style e (or style E in the case of an A conversion specifier) with …

Committee Discussion

The committee agrees that this is a defect and accepts the Suggested Technical Corrigendum

Proposed Technical Corrigendum

In 12.5, in the text added to 7.21.6.1#8 and 7.29.2.1#8, under a,A conversion specifiers, replace the bullets:

—    if −(n+5) ≤ q ≤ 0, use style f formatting with formatting precision equal to −q,

—    otherwise, use style e formatting with …

with:

—    if −(n+5) ≤ q ≤ 0, use style f (or style F in the case of an A conversion specifier) with formatting precision equal to −q,

—    otherwise, use style e (or style E in the case of an A conversion specifier) with …

DR 12 Prev <— Closed —> Next DR 15, or summary at top

DR 15

DR 14 Prev <— Closed —> Next DR 1, or summary at top

Submitter: Jim Thomas
Submission Date: 2017-10-25
Source: WG14
Reference Document: N2171
Subject: P3: Characteristic macros for non-arithmetic formats

Summary

This DR addresses an issue related to DR 501 and IEC 60559 non-arithmetic interchange formats.

TS 18661-3 attempted to change the definition of DECIMAL_DIG so that it would apply to all supported IEC 60559 formats. DR 501 shows that this change is problematic. After several unsuccessful attempts at a suitable TC, WG 14 and the CFP group have agreed to obsolesce DECIMAL_DIG in favor of the type_DECIMAL_DIG macros.

The type_DECIMAL_DIG macros are helpful in using conversions between binary floating-point formats and decimal character sequences. TS 18661-3 specifies such conversions for non-arithmetic binary interchange formats, but neglects to include the macros for such formats.

Likewise, the type_DIG macros too are helpful in using conversions between binary floating-point formats and decimal character sequences, though TS 18661-3 does not specify them for non-arithmetic formats.

The suggested TC below extends the set of FLTN_DECIMAL_DIG and FLTN_DIG macros to include ones for IEC 60559 non-arithmetic binary interchange formats.

Suggested Technical Corrigendum

In TS 18661-3 5.3, in the text for 5.2.4.2.2#6c, after the list for supported types _FloatN, insert:

for IEC 60559 non-arithmetic binary interchange formats of width N:

FLTN_DECIMAL_DIG     FLTN_DIG            

In TS 18661-3 clause 7, in the text for 5.2.4.2.2a#1, change:

The prefix FLTN_ indicates a binary interchange floating type of width N.

to:

The prefix FLTN_ indicates a binary interchange floating type or a non-arithmetic binary interchange format of width N.

Change the last sentence of  5.2.4.2.2a#1 from:

Conversely, for each such type that the implementation does not provide, <float.h> shall not define the associated macros in the following lists.

to:

Conversely, for each such type that the implementation does not provide, <float.h> shall not define the associated macros in the following list, except, the implementation shall define the macros FLTN_DECIMAL_DIG and FLTN_DIG if it supports IEC 60559 non-arithmetic binary interchange formats of width N by providing the encoding-to-encoding conversion functions in <math.h> and the string-to-encoding and string-from-encoding functions in <stdlib.h>.

Committee Discussion

Proposed Technical Clarification

In TS 18661-3 5.3, in the text for 5.2.4.2.2#6c, after the list for supported types _FloatN, insert:

for IEC 60559 non-arithmetic binary interchange formats of width N:

FLTN_DECIMAL_DIG     FLTN_DIG            

In TS 18661-3 clause 7, in the text for 5.2.4.2.2a#1, change:

The prefix FLTN_ indicates a binary interchange floating type of width N.

to:

The prefix FLTN_ indicates a binary interchange floating type or a non-arithmetic binary interchange format of width N.

Change the last sentence of  5.2.4.2.2a#1 from:

Conversely, for each such type that the implementation does not provide, <float.h> shall not define the associated macros in the following lists.

to:

Conversely, for each such type that the implementation does not provide, <float.h> shall not define the associated macros in the following list, except, the implementation shall define the macros FLTN_DECIMAL_DIG and FLTN_DIG if it supports IEC 60559 non-arithmetic binary interchange formats of width N by providing the encoding-to-encoding conversion functions in <math.h> and the string-to-encoding and string-from-encoding functions in <stdlib.h>.

DR 14 Prev <— Closed —> Next DR 1, or summary at top

DR 16

DR 13 Prev <— Open —> Next DR 20, or summary at top

Submitter: Jim Thomas
Submission Date: 2017-10-25
Source: WG14
Reference Document: N2178
Subject: P1: tgmath cbrt macro

Summary

This DR addresses a problem noted by Joseph Myers in emails SC22WG14.14743 and 14744:

… the example [in TS 18661-1 clause 16, for 7.25#6b – see below] implies that "#undef cbrtl" before calling the cbrt type-generic macro would mean it's not affected by constant rounding modes, but the actual normative text says "is affected by constant rounding modes" with no such caveat.  

and

… neither example definition [in C11 or TS 18661-1] is valid because they might call a block-scope cbrtf / cbrtl; they need to avoid such a block-scope identifier, or a macro defined by the user, while still depending on whether expansion of the 
standard header cbrtf / cbrtl macros has been suppressed at that point.

The text in question is:

[6b] A type-generic macro corresponding to a function indicated in the table in 7.6.1a is affected by constant rounding modes (7.6.2). Note that the type-generic macro definition in the example in 6.5.1.1 does not conform to this specification. A conforming macro could be implemented as follows:

#define cbrt(X)     _Generic((X),                  \

               long double: cbrtl(X),         \

               default: _Roundwise_cbrt(X),   \

               float: cbrtf(X)                \

               )

where _Roundwise_cbrt() is equivalent to cbrt() invoked without macro-replacement suppression.

The cause of the problems is the use of cbrtl and cbrtf in the macro definition. The suggested TC below replaces these uses with _Roundwise_ prefixed identifiers similar to _Roundwise_cbrt.

Suggested Technical Corrigendum

In TS 18661-1, clause 16, replace:

#define cbrt(X)     _Generic((X),                  \

               long double: cbrtl(X),         \

               default: _Roundwise_cbrt(X),   \

               float: cbrtf(X)                \

               )

where _Roundwise_cbrt() is equivalent to cbrt() invoked without macro-replacement suppression.

with

#define cbrt(X)     _Generic((X),                       \

               long double: _Roundwise_cbrtl(X),   \

               default: _Roundwise_cbrt(X),        \

               float: _Roundwise_cbrtf(X)               \

               )

where _Roundwise_cbrtl(), _Roundwise_cbrt(), and _Roundwise_cbrtf() are equivalent to cbrtl(), cbrt(), and cbrtf(), respectively, invoked without macro-replacement suppression.

Committee Discussion

There was considerable discussion on this issue. The first point is that the _Generic example cited that is proposed to be fixed was not intended to impose requirements, yet has elicited two fixes so far, this being a third. The second point is that the fix offered would likely elicit numerous compiler errors as stated and no longer serves its original intention of illustrating _Generic best practice usage. Lastly, lacking a clear simple example, is there a problem here that needs clarification becomes uncertain.

Apr 2018 meeting

Committee Discussion

A new document N2212 was presented with a much simpler proposed change. It was accepted by the committee.

Proposed Change

In TS 18661-1, clause 16, replace:

       #define cbrt(X) _Generic((X), \
                       long double: cbrtl(X), \
                       default: _Roundwise_cbrt(X), \
                       float: cbrtf(X) \
                       )

where _Roundwise_cbrt() is equivalent to cbrt() invoked without macro- replacement suppression.

with

       #define cbrt(X) _Generic((X), \
                       long double: _Roundwise_cbrtl, \
                       default: _Roundwise_cbrt, \
                       float: _Roundwise_cbrtf \
                       )(X)

where _Roundwise_cbrtl(), _Roundwise_cbrt(), and _Roundwise_cbrtf() are equivalent to cbrtl(), cbrt(), and cbrtf(), respectively, invoked without macro- replacement suppression.

DR 13 Prev <— Open —> Next DR 20, or summary at top

DR 17

DR 19 Prev <— Review —> Next DR 18, or summary at top

Submitter: C Floating Point Group
Submission Date: 2018-02-11
Source: WG14
Reference Document: N2203
Subject: P3: incommensurate arguments for comparison macros

Summary

This DR addresses a problem noted by Joseph Myers in email SC22WG14.14885:

The usual arithmetic conversions in TS 18661-3 include "If both operands have floating types and neither of the sets of values of their corresponding real types is a subset of (or equivalent to) the other, the behavior is undefined.".

Thus, for example, if neither of long double and _Float128 has a set of values that is a subset of the other, given

long double a;
_Float128 b;

it's undefined to have the expression "a < b".

Now what about the expression "isless (a, b)"?  By analogy with the direct comparison, it would seem natural for it to be undefined.  But while 18661-2 explicitly disallows using those macros with one decimal and one non-decimal argument, I see nothing to disallow the case where neither set of values is a subset of the other, and the definition of these macros doesn't actually include the usual arithmetic conversions.

It was an oversight to not disallow argument types neither of which is a subset (or equivalent to) the other.

Suggested Technical Corrigendum

In TS 18661-3, at the end of clause 12 (just before 12.1), insert:

To 7.12.14#1, append:

If neither of the sets of values of the argument formats is a subset of (or equivalent to) the other, the behavior is undefined. 

Apr 2018 meeting

Committee Discussion

The committee accepted the Suggested Technical Corrigendum as the Proposed Change (using our new terminology).

Proposed Change

In TS 18661-3, at the end of clause 12 (just before 12.1), insert:

To 7.12.14#1, append:

If neither of the sets of values of the argument formats is a subset of (or equivalent to) the other, the behavior is undefined. 

DR 19 Prev <— Review —> Next DR 18, or summary at top

DR 18

DR 17 Prev <— Review —> Next DR 19, or summary at top

Submitter: C Floating Point Group
Submission Date: 2018-02-22
Source: WG14
Reference Document: N2204
Subject: P4: missing specification of preferred quantum exponents

Summary

TS 18661-4 neglected to specify the preferred quantum exponent for its new functions. This was an oversight. The following suggested TC adds the missing specification.

Suggested Technical Corrigendum

In TS 18661-4, at the end of clause 7, append:

In the Preferred Quantum Exponents table in 5.2.4.2.2a#7, insert before the final row:

In TS 18661-4, at the end of clause 8, append:

In the Preferred Exponents Table in 5.2.4.2.2a#7, append:

Apr 2018 meeting

Committee Discussion

The committee accepted the Suggested Technical Corrigendum as the Proposed Change (using our new terminology).

Proposed Change

In TS 18661-4, at the end of clause 7, append:

In the Preferred Quantum Exponents table in 5.2.4.2.2a#7, insert before the final row:

In TS 18661-4, at the end of clause 8, append:

In the Preferred Exponents Table in 5.2.4.2.2a#7, append:

DR 17 Prev <— Review —> Next DR 19, or summary at top

DR 19

DR 18 Prev <— Review —> Next DR 17, or summary at top

Submitter: C Floating Point Group
Submission Date: 2018-03-16
Source: WG14
Reference Document: N2210
Subject: P1: updating underflow definition

Summary

C11 7.12.1 says

The result underflows if the magnitude of the mathematical result is so small that the mathematical result cannot be represented, without extraordinary roundoff error, in an object of the specified type.232)

232) The term underflow here is intended to encompass both ‘‘gradual underflow’’ as in IEC 60559 and also ‘‘flush-to-zero’’ underflow

The C definition of underflow doesn’t accomplish its intention as expressed in footnote 232. It’s too restrictive to encompass IEC 60559 gradual underflow, because IEC 60559 underflow can occur without extraordinary roundoff error.

In IEC 60559:2011, the underflow flag is raised (in C terminology, set) if and only if the result is tiny and inexact, even if the roundoff error is the same as it would have been if the full (normal) precision of the type were available, i.e., even if there is no extraordinary roundoff error.

IEC 60559:1989 offered the implementation the option to raise the underflow flag for tiny results with extraordinary roundoff error, though it also offered the option to raise the underflow flag for tiny inexact results, which is what most implementations did. IEC 60559:2011 dropped the option to raise the underflow flag based on extraordinary roundoff error.

The following suggested TC aims to loosens the C definition of underflow to encompass IEC 60559:2011 underflow behavior, which is based on tiny inexact results.

Suggested Technical Corrigendum

In TS 18661-1, before 14.1, insert:

14.0 C underflow

The following change to C11 loosens the C definition of underflow to encompass IEC 60559 gradual underflow, as is its stated intention (see C11 footnote 232).

Changes to C11:

Change the first sentence in 7.12.1#6 from:

[6] The result underflows if the magnitude of the mathematical result is so small that the mathematical result cannot be represented, without extraordinary roundoff error, in an object of the specified type.232) …

to:

[6] The result underflows if the magnitude of the mathematical result is nonzero and less than the minimum normal number in the type.232) …

Apr 2018 meeting

Committee Discussion

The committee accepted the Suggested Technical Corrigendum as the Proposed Change (using our new terminology).

Proposed Change

In TS 18661-1, before 14.1, insert:

14.0 C underflow

The following change to C11 loosens the C definition of underflow to encompass IEC 60559 gradual underflow, as is its stated intention (see C11 footnote 232).

Changes to C11:

Change the first sentence in 7.12.1#6 from:

[6] The result underflows if the magnitude of the mathematical result is so small that the mathematical result cannot be represented, without extraordinary roundoff error, in an object of the specified type.232) …

to:

[6] The result underflows if the magnitude of the mathematical result is nonzero and less than the minimum normal number in the type.232) …

DR 18 Prev <— Review —> Next DR 17, or summary at top

DR 20

DR 16 Prev <— Open —> Next DR 21, or summary at top

Submitter: C Floating Point Group
Submission Date: 2018-03-16
Source: WG14
Reference Document: N2211
Subject: P1: changes for obsolescing DECIMAL_DIG

Summary

N2211 described changes in C11 and TS 18661 to remove references to DECIMAL_DIG, which CR501 is expected to obsolesce. The changes that apply to C11 are collected in N2253 as an update to the suggested TC in CR501. The changes that apply to TS 18661-1 compose the CR in this document. The remaining change is for TS 18661-3, which will be covered by a CR in a subsequent document.

Suggested Technical Corrigendum

In 7.1, omit:

Change footnote 361) from:

361)  If the minimum-width IEC60559 extended format (64 bits of precision) is supported, DECIMAL_DIG shall be at least 21. If IEC 60559 double (53 bits of precision) is the widest IEC 60559 format supported, then DECIMAL_DIG shall be at least 17. (By contrast, LDBL_DIG and DBL_DIG are 18 and 15, respectively, for these formats.)

to:

361)  If the minimum-width IEC 60559 binary64-extended format (64 bits of precision) is supported, DECIMAL_DIG shall be at least 21. If IEC 60559 binary64 (53 bits of precision) is the widest IEC 60559 format supported, then DECIMAL_DIG shall be at least 17. (By contrast, LDBL_DIG and DBL_DIG are 18 and 15, respectively, for these formats.)

In 10.1, change:

After F.5#2, insert:

[2a] The <float.h> header defines the macro

CR_DECIMAL_DIG

if and only if __STDC_WANT_IEC_60559_BFP_EXT__ is defined as a macro at the point in the source file where <float.h> is first included. If defined, CR_DECIMAL_DIG expands to an integral constant expression suitable for use in #if preprocessing directives whose value is a number such that conversions between all supported types with IEC 60559 binary formats and character sequences with at most CR_DECIMAL_DIG significant decimal digits are correctly rounded. The value of CR_DECIMAL_DIG shall be at least DECIMAL_DIG + 3. If the implementation correctly rounds for all numbers of significant decimal digits, then CR_DECIMAL_DIG shall have the value of the macro UINTMAX_MAX.

[2b] Conversions of types with IEC 60559 binary formats to character sequences with more than CR_DECIMAL_DIG significant decimal digits shall correctly round to CR_DECIMAL_DIG significant digits and pad zeros on the right.

[2c] Conversions from character sequences with more than CR_DECIMAL_DIG significant decimal digits to types with IEC 60559 binary formats shall correctly round to an intermediate character sequence with CR_DECIMAL_DIG significant decimal digits, according to the applicable rounding direction, and correctly round the intermediate result (having CR_DECIMAL_DIG significant decimal digits) to the destination type. The “inexact” floating-point exception is raised (once) if either conversion is inexact. (The second conversion may raise the “overflow” or “underflow” floating-point exception.)

In F.5#2c, attach a footnote to the wording:

The “inexact” floating-point exception is raised (once) if either conversion is inexact.

where the footnote is:

*) The intermediate conversion is exact only if all input digits after the first CR_DECIMAL_DIG digits are 0.

to:

Replace the content of F.5 with:

[1] The <float.h> header defines the macro

CR_DECIMAL_DIG

if and only if __STDC_WANT_IEC_60559_BFP_EXT__ is defined as a macro at the point in the source file where <float.h> is first included. If defined, CR_DECIMAL_DIG expands to an integral constant expression suitable for use in #if preprocessing directives whose value is a number such that conversions between all supported IEC 60559 binary formats and character sequences with at most CR_DECIMAL_DIG significant decimal digits are correctly rounded. The value of CR_DECIMAL_DIG shall be at least M + 3, where M is the maximum value of the T_DECIMAL_DIG macros for IEC 60559 binary formats. If the implementation correctly rounds for all numbers of significant decimal digits, then CR_DECIMAL_DIG shall have the value of the macro UINTMAX_MAX.

[2] Conversions of types with IEC 60559 binary formats to character sequences with more than CR_DECIMAL_DIG significant decimal digits shall correctly round to CR_DECIMAL_DIG significant digits and pad zeros on the right.

[3] Conversions from character sequences with more than CR_DECIMAL_DIG significant decimal digits to types with IEC 60559 binary formats shall correctly round to an intermediate character sequence with CR_DECIMAL_DIG significant decimal digits, according to the applicable rounding direction, and correctly round the intermediate result (having CR_DECIMAL_DIG significant decimal digits) to the destination type. The “inexact” floating-point exception is raised (once) if either conversion is inexact. (The second conversion may raise the “overflow” or “underflow” floating-point exception.)

[4] The specification in this subclause assures conversion between IEC 60559 binary format and decimal character sequence follows all pertinent recommended practice. It also assures conversion from IEC 60559 format to decimal character sequence with at least T_DECIMAL_DIG digits and back, using to-nearest rounding, is the identity function, where T is the macro prefix for the format.

[5] Functions such as strtod that convert character sequences to floating types honor the rounding direction. Hence, if the rounding direction might be upward or downward, the implementation cannot convert a minus-signed sequence by negating the converted unsigned sequence.

In F.5#3, attach a footnote to the wording:

The “inexact” floating-point exception is raised (once) if either conversion is inexact.

where the footnote is:

*) The intermediate conversion is exact only if all input digits after the first CR_DECIMAL_DIG digits are 0.

Apr 2018 meeting

Committee Discussion

The paper was only briefly discussed.

This resolution is tied to the Floating Point CR 22 as well as to the C2x DR 501.

DR 16 Prev <— Open —> Next DR 21, or summary at top

DR 21

DR 20 Prev <— Open —> Next DR 22, or summary at top

Submitter: C Floating Point Group
Submission Date: 2018-03-24
Source: WG14
Reference Document: N2224
Subject: P1: printf of one-digit character string

Summary

It is possible that a floating-point value, when converted to a one-digit character string, results in odd numbers no matter which way rounding is done. For roundTiesToEven, IEC 60559 specifies that the larger magnitude value be used.

Suggested Technical Corrigendum:

Add the following to TS 18661-1 clause 10.2 Conversions to character sequences:

Add to F.5 Binary-decimal conversion:

NOTE: IEC 60559 specifies that conversion to one-digit character strings using roundTiesToEven when both choices are odd, shall produce the value with the larger magnitude. This can happen with 9.5 whose nearest neighbors are 9.e0 and 1.e1, both of which are odd.

Apr 2018 meeting

Committee Discussion

The committee, once it understood the three floating values supplied, agreed with the proposed change.

Subsequently, the author supplied revised values making such an understanding easier. The committee as a whole has yet to see the tweaked values presented below as the Proposed Change.

Proposed Change

Add the following to TS 18661-1 clause 10.2 Conversions to character sequences:

Add to F.5 Binary-decimal conversion:

NOTE: IEC 60559 specifies that conversion to one-digit character strings using roundTiesToEven when both choices are odd, shall produce the value with the larger magnitude. This can happen with 9.5e2 whose nearest neighbors are 9.e2 and 1.e3, both of which are odd.

DR 20 Prev <— Open —> Next DR 22, or summary at top

DR 22

DR 21 Prev <— Open —> Next DR 13, or summary at top

Submitter: C Floating Point Group
Submission Date: 2018-03-16
Source: WG14
Reference Document: N2211
Subject: P2: changes for obsolescing DECIMAL_DIG

Summary

N2211 described changes in C11 and TS 18661 to remove references to DECIMAL_DIG, which CR501 is expected to obsolesce. The changes that apply to C11 are collected in N2253 as an update to the suggested TC in CR501. The changes that apply to TS 18661-1 are in the CR in N2254. The remaining change is for TS 18661-3, which is covered by the CR in this document.

Suggested Technical Corrigendum

At the end of clause 7, omit:

With the following change, DECIMAL_DIG characterizes conversions of supported IEC 60559 encodings, which may be wider than supported floating types.

Change to C11 + TS18661-1 + TS18661-2:

In 5.2.4.2.2#11, change the bullet defining DECIMAL_DIG from:

—  number of decimal digits, n, such that any floating-point number in the widest supported floating type with …

to:

—  number of decimal digits, n, such that any floating-point number in the widest of the supported floating types and the supported IEC 60559 encodings with …

Apr 2018 meeting

Committee Discussion

The paper was only briefly discussed.

This resolution is tied to the Floating Point CR 20 as well as to the C2x DR 501.

DR 21 Prev <— Open —> Next DR 13, or summary at top