FIX: store_be(size_t(42)) would not compile on Xcode

[reneme]

Xcode defines uint64_t as "unsigned long long" and size_t as "unsigned long int". Both are 64bit types, but the compiler failed to call reverse_bytes() for size_t regardless. It claimed the call is ambiguous. This was initially found by @FAlbertDev in #4119.

The workaround makes reverse_bytes() an (awful) template that just assumes to get a std::unsigned_integral T and disambiguates on sizeof(T) using if constexpr. I'm certainly open to better ideas. But then again: C++23 will obsolete this anyway.

[FAlbertDev]

Thanks! XCode hates this one weird trick.

[randombit]

I'm confused why you would ever want to do this. The number of bytes written will depend on the size of size_t, which means it's not really useful for serialization. What am I missing?

[coveralls]

coverage: 91.779% (-0.003%) from 91.782%
when pulling 2f83991 on Rohde-Schwarz:fix/store_be
into 4c3e9dc on randombit:master.

[reneme]

This is not supposed to change any behavior. It's just working around Xcode's compiler not being able to dispatch the overload of reverse_bytes.

See here for a build failure on macOS: https://github.com/randombit/botan/actions/runs/9515856655/job/26230969296?pr=4119

[randombit]

This is not supposed to change any behavior.

Sure but my point was that we had not hit this so far because saying store_be(some_size_t) isn't really something that makes sense because the number of bytes emitted depends on the size of size_t, and so isn't really usable for anything. Serialize it, hash it, whatever, the difference in byte count will mean you have different output.

I looked at the specific code in ounsworth.cpp that triggered this and I guess store_be with a size_t is at least usable in that case, because you're post-processing it to truncate unused bytes. This feels like a real edge case to me (wtf is up with these overly complicated variable length encodings) but meh.

[reneme]

Sure but my point was that we had not hit this so far because saying store_be(some_size_t) isn't really something that makes sense because the number of bytes emitted depends on the size of size_t, and so isn't really usable for anything.

True! I didn't even think of that. But that answers: "How did that ever work?" 🤣