LWG was concerned about the following wording in [linalg.general].
Within all the functions in [linalg], any calls to abs, conj, imag, and real are unqualified.
A suggestion was to reuse the wording form used in [contents] p3 at the end, like the following.
Throughout the specification in [linalg], the meanings of the unqualified names abs, conj, imag, and real are established as-if by performing argument-dependent lookup (6.5.4)."
LWG's impression of the intention is that abs, conj, imag, and real are selected via overload resolution on a candidate set that includes:
the correspondingly named functions defined in <cmath> ([cmath.syn]), and
the lookup set produced by argument-dependent lookup (6.5.4).
I clarified that the abs / conj / imag / real machinery in P1673 exists for the following reasons.
For custom types, use ADL to find the functions.
For custom types, assume that if conj, imag, or real can't be found via ADL, then the type represents a non-complex number.
For Standard types, "fix" behavior of existing <cmath> etc. functions (e.g., so that abs of an unsigned integer is the integer, and conj of a real number doesn't have complex type).
P2642R2
LWG was concerned that the wording says "apply all wording from P2642R2," but P2642R2 is still in LEWG review. This wording replaced some mdspan layouts that were originally in the proposal. I think P2642 is completely separable from P1673, so we might not even need this.
Fix declaration form of exposition-only concepts
In [linalg.syn], replace patterns like this:
template<class T>
concept in-vector; = // exposition only
with
template<class T>
concept in-vector = _see below_; // exposition only
Fix [linalg.general]
[linalg.general] (1.2) and (1.3), change
The pattern xT should be read as “the transpose of x.”
to the following (as "should" sounds like implementations could ignore it):
The pattern xT has the meaning “the transpose of x."
Varna 2023 LWG comments from chat
Regarding abs, conj, imag, and real
LWG was concerned about the following wording in [linalg.general].
A suggestion was to reuse the wording form used in [contents] p3 at the end, like the following.
LWG's impression of the intention is that abs, conj, imag, and real are selected via overload resolution on a candidate set that includes:
<cmath>([cmath.syn]), andI clarified that the abs / conj / imag / real machinery in P1673 exists for the following reasons.
<cmath>etc. functions (e.g., so that abs of an unsigned integer is the integer, and conj of a real number doesn't have complex type).P2642R2
LWG was concerned that the wording says "apply all wording from P2642R2," but P2642R2 is still in LEWG review. This wording replaced some mdspan layouts that were originally in the proposal. I think P2642 is completely separable from P1673, so we might not even need this.
Fix declaration form of exposition-only concepts
In [linalg.syn], replace patterns like this:
with
Fix [linalg.general]
[linalg.general] (1.2) and (1.3), change
to the following (as "should" sounds like implementations could ignore it):