As I've stated in the bug report, I don't think that's an acceptable solution. This can be fixed without penalizing every user who doesn't use ordered_json.

As I've stated in the bug report, I don't think that's an acceptable solution. This can be fixed without penalizing every user who doesn't use ordered_json.

Does explicit instantiation hurt others?

It hurts everyone's compilation speed. And there is a solution that doesn't.

Also, I'd like to understand this problem better to make sure we're not masking a deeper issue. That's one of the reasons I haven't submitted a PR myself yet.

Coverage decreased (-0.3%) to 99.67% when pulling c67e101 on luistung:issue3732 into 58bd97e on nlohmann:develop.

It hurts everyone's compilation speed. And there is a solution that doesn't.

Also, I'd like to understand this problem better to make sure we're not masking a deeper issue. That's one of the reasons I haven't submitted a PR myself yet.

I got it. But you reminded me another thing. The nlohmann::json has been instantiated by default due to the declaretion of inline nlohmann::json operator "" _json(const char* s, std::size_t n). How can user who does not use nlohmann::json get rid of the instantiation.

True. This was also brought up here recently:

(...) And the template gets instantiated in every translation unit because of the _json UDL, even if you don't use any nlohmann function (although that can somewhat be mitigated with the _fwd header).

There's no way to disable the definition when including json.hpp.

Also, the CI is red.

warning C4459: declaration of 'j' hides global declaration

I've fixed this issue in several PRs that have not been merged. I really should submit a separate PR.

(The noexcept unit test declares a global variable j that conflicts with parameters named j. Easily fixed by renaming the global.)