Compilation errors (VS 2022 32-bit /W3 /SDL)

[Kemp-J]

When building a project that uses the jwt-cpp header in Visual Studio 2022 with warning level 3 and SDL checks enabled there are several errors raised related to suspicious implicit casts. These can be easily solved with a few tweaks. Specifically:

• The value returned by jwt::alphabet::index is currently a uint32_t, which is smaller than the ptrdiff_t that the call to std::distance returns. The return value can be switched to size_t to solve this.
• The sextet_* and triple variables in jwt::base::details::decode call the above and so need to be size_t as well.

In modern C++ both of the above points can be solved with auto in preference to specifying the types.

• The argument passed to system_clock::from_time_t in jwt::basic_claim::as_date requires a static_cast to a time_t.

I was going to raise an issue about a common base class for exceptions, but it looks like someone got there first :)

[prince-chrismc]

Hmm, never heard of /sdl very cool see msvc getting way better.

I would love to see a PR with these fixes 🙏

[Kemp-J]

I'll see what I can do. I use git elsewhere but I'm quite new to using it with github 😬

[prince-chrismc]

It's the same tbh, just create a fork (there a button top left) then git clone your copy, and the rest is the same.
After you push you changes to a branch, this repo will have a "open pull request banner" if not the pull request tab has a "new" button :)

It sounds like a lot but the CLI will even output hints along the way with links!

[Kemp-J]

A followup issue - using nlohmann json with jwt-cpp has additional errors for 32-bit builds. The source of this is that is_substr_start_end_index_signature and is_substr_start_index_signature check for the existence of substr member functions on string_type, but they do so on the assumption that they will accept integer_type for the parameters. The integer_type for nlohmann json is int64_t, the substr function defined for std::string has size_t parameters, and thus the error as the compiler sees you effectively trying to pass int64_t values to a size_t parameter (unsigned int on 32-bit builds). Given that, as far as I can tell, you don't actually need substr to accept integer_type specifically, is it worth swapping the check to look for substr taking size_t?

[prince-chrismc]

Windows 32-bit?! I've not touched that in 8 years :) I have no idea.

[Kemp-J]

Yeah, I have a few things I have to build as 32-bit for... historical reasons.

I guess there's mainly a question of intent here. If there is a reason substr needs to take integer_type, because it's called with values directly from a token, then it'll need some creativity. I don't think your calls to substr are used like that though (pretty much just hardcoded values and ones calculated from other sources) so moving the check away from integer_type wouldn't cause any problems. If you have insight on anything I missed then I'd be happy to hear it, otherwise I'll roll that into my changes. Might as well have 32-bit working if it only requires a small tweak here and there 🙂