define a container for reading bytes from a file

[oliverlee]

Define a byte_container class template for use in tests. It provides
std::byte-oriented access (as opposed to char) to the bytes in a
file or string.

Change-Id: I364fcb0efaae9ac79bb78198912b885b3d9f54c5

[garymm]

I guess this is more type safe because of static_cast rather than reinterpret_cast?
But it is a lot of code, and I think has an extra copy (vs the reinterpet_cast).
If that's right I'm inclined to stick with reinterpret_cast, or just switch everything from byte to uint8_t. I know byte is conceptually what we want, but it seems like it interacts with the file I/O library badly and I don't really see any downside to using uint8_t.

Thoughts?

[oliverlee]

I guess this is more type safe because of static_cast rather than reinterpret_cast? But it is a lot of code, and I think has an extra copy (vs the reinterpet_cast). If that's right I'm inclined to stick with reinterpret_cast, or just switch everything from byte to uint8_t. I know byte is conceptually what we want, but it seems like it interacts with the file I/O library badly and I don't really see any downside to using uint8_t.

Thoughts?

I don't think static_cast vs reinterpret_cast is a big deal but more that it's useful to have a type that provides byte traversal and owns the data. The internal storage can be changed from std::byte to char and then reinterpret_cast moved to the begin and end functions.

There's an extra copy for the in_place constructor overload, as we need to copy all the chars into the container, but I don't think that's a huge problem since this is currently test only.

std::byte is not an arithmetic type and cannot be implicitly be promoted to int. Outside of reading input files and constant strings we use in test cases, I think this should make of the interface that uses char so poor ergonomics with iostreams should be handled going forward.