Switch to std::format

[edolstra]

Is your feature request related to a problem? Please describe.

Unlike boost::format, std::format is supposed to check format strings at compile time, so it should prevent problems like #10293 once and for all.

Describe the solution you'd like
A clear and concise description of what you want to happen.

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.

Priorities

Add 👍 to issues you find important.

[SkamDart]

I looked into this a bit and std::format does not seem to be very well supported across compilers yet. See Godbolt and currently supported native x86_64-linux gcc/clang toolchains.

Would it be acceptable to add {fmt} as a dependency until std::format is stabilized in the toolchains nix supports?

nix use-format → cat format_test.cc
#include <cassert>
#include <format>

int main()
{
    std::string message = std::format("The answer is {}.", 42);
    assert(message == "The answer is 42.");
}
nix use-format → clang++ -o format_test -std=c++2a format_test.cc
format_test.cc:2:10: fatal error: 'format' file not found
#include <format>
         ^~~~~~~~
1 error generated.
nix use-format → clang++ --version
clang version 16.0.6
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /nix/store/xph7agmqsna5ia7h699acfmk9gp4iks3-clang-16.0.6/bin
nix use-format → g++ --version
gcc (GCC) 12.3.0
Copyright (C) 2022 Free Software Foundation, Inc.
This is free software; see the source for copying conditions.  There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.

nix use-format → g++ -o format_test -std=c++2a format_test.cc
format_test.cc:2:10: fatal error: format: No such file or directory
    2 | #include <format>
      |          ^~~~~~~~
compilation terminated.

[edolstra]

Yes that might be an option. Or we wait for GCC 13 and Clang 14.

[rahulharpal1603]

Yes that might be an option. Or we wait for GCC 13 and Clang 14.

Hi, I am new to open source and have looked into this issue for a while now. If I am not wrong, our target file is this: https://github.com/NixOS/nix/blob/master/src/libutil/fmt.hh

Since std::format is now supported in gcc compiler version 13.1.0, we can replace boost::format with std::format. But, this will require many changes to the code base, mainly due to differences in the syntax of boost::format and std::format.

In the fmt.hh file, the class HintFmt is defined as a wrapper class for boost::format in the fmt.hh file. There are around 450 calls to HintFmt() in the entire codebase. In each call to HintFmt(), the first argument is a std::string e.g :-

In the file https://github.com/NixOS/nix/blob/master/src/libexpr/parser-state.hh at line 68, the following code is present:

inline void ParserState::dupAttr(Symbol attr, const PosIdx pos, const PosIdx prevPos)
{
    throw ParseError({
        .msg = HintFmt("attribute '%1%' already defined at %2%", symbols[attr], positions[prevPos]),
        .pos = positions[pos]
    });
}

if we are using std::format instead of boost::format, we must change the string "attribute '%1%' already defined at %2%" to "attribute '{0}' already defined at {1}".

We can replace these strings now, consider this call to HintFmt():
link: https://github.com/NixOS/nix/blob/master/src/libexpr/primops/fetchClosure.cc at line 161

    if (!fromPath)
        throw Error({
            .msg = HintFmt("attribute '%s' is missing in call to 'fetchClosure'", "fromPath"),
            .pos = state.positions[pos]
        });

Like printf(), boost::format supports format specifiers like %s , but std::format does not. So, we have to change attribute '%s' is missing in call to 'fetchClosure'" to attribute '{}' is missing in call to 'fetchClosure'" for using std::format.

We must also rewrite the HintFmt constructors and the operator overloads to support std::format.

Is this a correct approach? Please look into this.

[xokdvium]

Just wanted to mention that I've been working on a clang-based tool to automate such a migration.