[question]: macro to disable floating point support

[rianquinn]

We are interested in using JSON in our Bareflank Hypervisor for both comms and configuration files, and would love to use this implementation. Kernel / Hypervisor code however should not use double / float as these data types can generate exceptions that cannot be handled well inside the kernel.

It appears that you already support signed / unsigned integers. Would you be interested in a patch that disables (likely via a macro of some sort) double / float completely to support people interested in using this library in non-userspace code?

[nlohmann]

Hi @rianquinn, this sounds interesting!

Did you see this paragraph in the README which could be relevant?

As the exact type of a number is not defined in the JSON specification, this library tries to choose the best fitting C++ number type automatically. As a result, the type double may be used to store numbers which may yield floating-point exceptions in certain rare situations if floating-point exceptions have been unmasked in the calling code. These exceptions are not caused by the library and need to be fixed in the calling code, such as by re-masking the exceptions prior to calling library functions.

How would you deactivate floating-point support? Did you try setting number_float_t to a different type? If possible, I would like to avoid getting to the #ifdef-hell...

[rianquinn]

The perfect solution would prevent float / double types / instructions from being compiled into the code. The reality is, so long as the code is not executed, it's fine. I did not try setting number_float_t to a different type, but my guess is it would collide with your existing unsigned / signed ints. The question is, When a JSON string is read, and it sees a "42" and not a "42.0" does it use ints and not floats from start to finish? So long as floats and doubles are never used, it should be fine.

[nlohmann]

You're right - setting number_float_t to an integer type yields collisions. So far, I do not see a simple solution to exclude any floating-point numbers from the class.

[nlohmann]

Any idea how to proceed on this?

[rianquinn]

I think the best thing to do is probably close the ticket. A couple of days ago I grabbed the source and gutted anything with "float" or "double". It only took about 5 mins, and the resulting code worked great. Based on those results, this implementation is perfect for what we need it for, but the only way I know to implement this more generically is to use #defines which I would agree would clutter what you have for a very small use case. For now, my plan is to maintain a fork with the modifications in our repo. I'm watching this repo, so anytime you roll a release, I'll make the mods, and push to our repo, and let you know if anything you guys change breaks what we are doing. But yeah, I wrote the https://github.com/QtMark/qmjson code, and I really like what you have done with this C++ only implementation so we will use it for sure.

[nlohmann]

This is fine with me. Thanks for the quick response, and maybe we find way to cope with your usecase in a general way!