WIP: replace SPDLOG_* macros with constexpr funcs + source_location

[franklange]

I wanted to investigate what it would take to replace the existing log macros (eg. SPDLOG_TRACE) with constexpr functions that make use of the new std::source_location [1] that will be available in GCC11 [2], as originally discussed here:
#1823

The idea of std::source_location is that you can provide it as a default-value parameter for your log function and it will still log all the info from the call-site.
However, combining default-value params with variadic template functions is almost impossible or at least quite involved. I opted for using user-defined deduction guides [3] as advocated on SO [4].

This is more an experiment but please let me know what you think :)
Would this be a direction worth pursuing?

---

[1] https://en.cppreference.com/w/cpp/utility/source_location
[2] https://en.cppreference.com/w/cpp/compiler_support
[3] https://en.cppreference.com/w/cpp/language/class_template_argument_deduction#User-defined_deduction_guides
[4] https://stackoverflow.com/questions/57547273/how-to-use-source-location-in-a-variadic-template-function

[franklange]

This compiled and worked for me on Linux + GCC 10.2.0.

[franklange]

Apparently needed for deduction guides.

[gabime]

Interesting approach. Thanks, for sharing. I wonder if the struct trick is the only way to achieve this. Ideally, this would be added directly to the existing functions.

[franklange]

Interesting approach. Thanks, for sharing. I wonder if the struct trick is the only way to achieve this. Ideally, this would be added directly to the existing functions.

Heyhey, thanks for the quick reply!
Unfortunately I couldn't find any other way of combining a variadic template with a default-value parameter. So I guess this might be the way to go for all the fmt-style functions.

Regarding bigger picture: so you envision using source_location for all the log functions? eg:

// spdlog.h

template<typename T>
inline void trace(const T &msg, std::source_location loc = std::source_location::current())
{
    default_logger_raw()->trace(loc, msg); // needs mods in logger.h to take + forward source_location
}

I wasn't quite sure if it's desirable to "force" logging the source location in every log function. If I understand the API correctly then currently having these additional macros allows me to chose wether I'm interested in src_location or not.

So ya, let me know what you think. I can also add this to the existing functions but then I'd have to adjust logger.h as well.

[gabime]

So ya, let me know what you think. I can also add this to the existing functions but then I'd have to adjust logger.h as well.

If the only benefit is that one can call spdlog_trace instead of SPDLOG_TRACE, then I don't think it worth the extra complexity (not to mention it only supported in c++20).

Also, the macros are zero code if the level is disabled at compile time, so they can't be completely replaced with the proposed.

[franklange]

Also, the macros are zero code if the level is disabled at compile time, so they can't be completely replaced with the proposed.

Shouldn't the if constexpr inside the functions here make sure that they're empty and optimized away if the corresponding level is disabled?

I don't mind throwing this away. I started this purely based on your "thumbs up" on the original idea posted here: #1823

What did you have in mind there?

[gabime]

Shouldn't the if constexpr inside the functions here make sure that they're empty and optimized away if the corresponding level is disabled?

Its not exactly the same. For example the args to the function would still be evaluated (spdlog_debug(“{}”, some_expensive_func());

What did you have in mind there?

That should be considered, but like I said, it seems to not worth it in the current state of affairs.