Date: Sat, 29 Apr 2023 11:27:30 -0500
On Wed, Apr 26, 2023 at 8:21 AM Arthur O'Dwyer via Std-Proposals <
std-proposals_at_[hidden]> wrote:
> FWIW, I'm strongly opposed to this. (And I did a double-take at Barry's
> post: is it April Fool's Day already?)
>
The hell is wrong with you?
> C++ isn't Perl; we don't use Unix shell globs like that. In C++, curly
> braces have a couple of different meanings (code block,
> initializer-sequence), but not "shell glob."
>
> (1) Ville has shown C++'s existing syntax for this:
> namespace LVT = longish::verbosish::tedious;
> using LVT::X;
> using LVT::Y;
> using LVT::Z;
>
> (2) It is a long-standing guideline that you shouldn't introduce multiple
> declarations on the same line of code.
> int x, y, z; // worse
>
> int x;
> int y;
> int z; // better
>
The reason this guideline exists for variables doesn't really apply to
using declarations: C declarators make multiple variables in one line
confusing, and once you start adding initializers to your multiple
variables in a line, it can be easy to miss an initialization and it's
particularly hard for users to understand.
On the other hand, this:
using std::format, std::format_to, std::formatter;
is just a shorter version of:
using std::format;
using std::format_to;
using std::formatter;
Repeating "using" doesn't really add much to the reader. Good practice to
split up using declarations from different namespaces, but the point OP is
making is that repeating "std::" doesn't really buy much either. This:
using std::{format, format_to, formatter};
is easier to read still.
>
> (3) The "zero, one, infinity" rule. In practice, `using`-declarations
> rarely come in twos and threes. There's either one using-declaration (e.g.
> to enable ADL
> <https://quuxplusone.github.io/blog/2022/04/30/poison-pill-redux/>), or
> hundreds (e.g. in the implementation of the <cmath> header
> <https://github.com/llvm/llvm-project/blob/main/libcxx/include/cmath#L335>).
> Neither scenario is a good one for this proposed (mis)feature. Anytime
> this feature saves a significant amount of typing, the result will be
> unfriendly to the human reader — and the code isn't likely to have been
> generated by a human typist in the first place.
>
The premise of this is not true - using declarations come in small groups
(i.e. more than one, fewer than "hundreds") pretty frequently. Introducing
an arbitrary terse namespace alias which remains in scope forever, which is
your proposed solution to this problem, is much more unfriendly to the
human reader than OP's proposal.
>
> (4) Another meaning of "C++ isn't Perl": We don't *need* to assign a
> meaning to every single sequence of ASCII characters. It's okay for some
> sequences of characters to be syntax errors; or to be reserved for future
> standardization, until we come up with a feature that is so useful that it
> deserves a dedicated syntax. The mere fact that "::{" is currently
> unassigned, is not an argument in favor of giving it *this* meaning.
>
As Jason already pointed out but bears repeating, this was obviously not
the argument being made
>
> John: Did you know about
> namespace XYZ = longish::verbosish::tedious;
> before this week? Now that you know about it, does it adequately solve
> your problem?
>
> –Arthur
>
I can't see ever using a namespace alias just for using declarations. This
is actively detracting from readability, in a way that doesn't actually
make your code meaningfully shorter. We're talking about changing:
using longish::verbosish::tedious::X;
using longish::verbosish::tedious::Y;
using longish::verbosish::tedious::Z;
to:
namespace LVT = longish::verbosish::tedious;
using LVT::X;
using LVT::Y;
using LVT::Z;
But the former is clearer, you already know what
longish::verbosish::tedious is, but LVT is meaningless. Sure, now you can
write:
namespace LVT = longish::verbosish::tedious;
using LVT::X, LVT::Y, LVT::Z;
So that you can end up using fewer lines of code, but this isn't... really
better. Maybe somebody prefers this, but I can't see ever writing it.
On the other hand, this:
using longish::verbosish::tedious::{X, Y, Z};
Is clearer precisely because it's shorter, and doesn't require adding extra
[effectively meaningless] identifiers to your program.
Barry
std-proposals_at_[hidden]> wrote:
> FWIW, I'm strongly opposed to this. (And I did a double-take at Barry's
> post: is it April Fool's Day already?)
>
The hell is wrong with you?
> C++ isn't Perl; we don't use Unix shell globs like that. In C++, curly
> braces have a couple of different meanings (code block,
> initializer-sequence), but not "shell glob."
>
> (1) Ville has shown C++'s existing syntax for this:
> namespace LVT = longish::verbosish::tedious;
> using LVT::X;
> using LVT::Y;
> using LVT::Z;
>
> (2) It is a long-standing guideline that you shouldn't introduce multiple
> declarations on the same line of code.
> int x, y, z; // worse
>
> int x;
> int y;
> int z; // better
>
The reason this guideline exists for variables doesn't really apply to
using declarations: C declarators make multiple variables in one line
confusing, and once you start adding initializers to your multiple
variables in a line, it can be easy to miss an initialization and it's
particularly hard for users to understand.
On the other hand, this:
using std::format, std::format_to, std::formatter;
is just a shorter version of:
using std::format;
using std::format_to;
using std::formatter;
Repeating "using" doesn't really add much to the reader. Good practice to
split up using declarations from different namespaces, but the point OP is
making is that repeating "std::" doesn't really buy much either. This:
using std::{format, format_to, formatter};
is easier to read still.
>
> (3) The "zero, one, infinity" rule. In practice, `using`-declarations
> rarely come in twos and threes. There's either one using-declaration (e.g.
> to enable ADL
> <https://quuxplusone.github.io/blog/2022/04/30/poison-pill-redux/>), or
> hundreds (e.g. in the implementation of the <cmath> header
> <https://github.com/llvm/llvm-project/blob/main/libcxx/include/cmath#L335>).
> Neither scenario is a good one for this proposed (mis)feature. Anytime
> this feature saves a significant amount of typing, the result will be
> unfriendly to the human reader — and the code isn't likely to have been
> generated by a human typist in the first place.
>
The premise of this is not true - using declarations come in small groups
(i.e. more than one, fewer than "hundreds") pretty frequently. Introducing
an arbitrary terse namespace alias which remains in scope forever, which is
your proposed solution to this problem, is much more unfriendly to the
human reader than OP's proposal.
>
> (4) Another meaning of "C++ isn't Perl": We don't *need* to assign a
> meaning to every single sequence of ASCII characters. It's okay for some
> sequences of characters to be syntax errors; or to be reserved for future
> standardization, until we come up with a feature that is so useful that it
> deserves a dedicated syntax. The mere fact that "::{" is currently
> unassigned, is not an argument in favor of giving it *this* meaning.
>
As Jason already pointed out but bears repeating, this was obviously not
the argument being made
>
> John: Did you know about
> namespace XYZ = longish::verbosish::tedious;
> before this week? Now that you know about it, does it adequately solve
> your problem?
>
> –Arthur
>
I can't see ever using a namespace alias just for using declarations. This
is actively detracting from readability, in a way that doesn't actually
make your code meaningfully shorter. We're talking about changing:
using longish::verbosish::tedious::X;
using longish::verbosish::tedious::Y;
using longish::verbosish::tedious::Z;
to:
namespace LVT = longish::verbosish::tedious;
using LVT::X;
using LVT::Y;
using LVT::Z;
But the former is clearer, you already know what
longish::verbosish::tedious is, but LVT is meaningless. Sure, now you can
write:
namespace LVT = longish::verbosish::tedious;
using LVT::X, LVT::Y, LVT::Z;
So that you can end up using fewer lines of code, but this isn't... really
better. Maybe somebody prefers this, but I can't see ever writing it.
On the other hand, this:
using longish::verbosish::tedious::{X, Y, Z};
Is clearer precisely because it's shorter, and doesn't require adding extra
[effectively meaningless] identifiers to your program.
Barry
Received on 2023-04-29 16:27:43