<format>: dynamic width or precision of non-integral type should be rejected

[mordante]

Describe the bug
Per [format.string.std]/7

… If the corresponding formatting argument is not of integral type 
… an exception of type format_­error is thrown.

When using a non-integral type as arg-id in std::format the code violates [format.fmt.string]/3 and is therefore ill-formed. The STL incorrectly accepts the format string, however, and throws an exception at run-time https://godbolt.org/z/b868c3M8z.

#include <format>
#include <iostream>

int main() {
    try {
        std::cout << std::format("{:*^{}}\n", 'a', 1.1);
    } catch(...) {
        std::cout << "die\n";
    }
    try {
        std::cout << std::format("{:.{}}\n", "abc", 1.1);
    } catch(...) {
        std::cout << "die\n";
    }
}

[cpplearner]

Currently _Basic_format_string uses _Format_checker which calls _Compile_time_parse_format_specs which simply calls formatter::parse, which does not know the type of other formatting arguments.

It might be possible to e.g. make _Format_checker handle known types specially, but should it do something that can't be mimicked by user-defined formatters?

(Aside: MSVC STL requires /utf-8 for compile-time format string checking. With the default execution character set, the checking is disabled.)

[frederick-vs-ja]

The sentence is changed in C++26 by WG21-P2757R3, which seemingly claims that a very similar case (std::format("{:>{}}", "hello", "10")) is well-formed in C++23 (or C++20 with DRs).

If the The option is valid only if the corresponding formatting argument is not of standard signed or unsigned integer type, or . If its value is negative, an exception of type format_error is thrown.

[CaseyCarter]

The sentence is changed in C++26 by WG21-P2757R3, which seemingly claims that a very similar case (std::format("{:>{}}", "hello", "10")) is well-formed in C++23 (or C++20 with DRs).

The paper is confused. That case (and the case in this thread) are clearly ill-formed per [format.fmt.string]/3. The issue the paper actually addresses is that [format.fmt.string]/3 isn't implementable for non-Standard format arguments with the current machinery in the Standard.

[frederick-vs-ja]

The sentence is changed in C++26 by WG21-P2757R3, which seemingly claims that a very similar case (std::format("{:>{}}", "hello", "10")) is well-formed in C++23 (or C++20 with DRs).

The paper is confused. That case (and the case in this thread) are clearly ill-formed per [format.fmt.string]/3. The issue the paper actually addresses is that [format.fmt.string]/3 isn't implementable for non-Standard format arguments with the current machinery in the Standard.

Thanks. It seems that we should treat the wording change in that paragraph editorial...

[mordante]

@StephanTLavavej To answer your question in #4152. I believe it's not incorrect since C++23, but since C++20.
http://eel.is/c++draft/format#string.std-10

If { arg-idopt } is used in a width or precision option, the value of the corresponding formatting argument
is used as the value of the option.
The option is valid only if the corresponding formatting argument is of standard signed or unsigned integer
type. If its value is negative, an exception of type format_error is thrown.

Since the value is not a "standard signed or unsigned integer type". It's not valid.

The wording in C++20 was slightly different
https://timsong-cpp.github.io/cppwp/n4861/format#string.std-7, but close enough. It changed by https://cplusplus.github.io/LWG/issue3720