Add support for custom information per benchmark run

[jonas-schulze]

• Set via bench.context("name", "value")
• Render/access via {{context(name)}}
• Rendering variables without a value produce empty output (for that variable) throws an exception
• Values persist to subsequent calls to Bench::run; my use case assumes that all variables are needed for all runs
• Reset via bench.clearContext()

Unfortunately, I haven't been able to build and therefore check the docs I wrote.

Also, I saw that some methods are overloaded for char const* and std::string const&. Should Bench::context do the same? What about Result::context?

Do you think context is a feasible name? If not, how should these methods be called?

[jonas-schulze]

Ubuntu hirsute has reached EOL, so I guess the CI failure is ok?

Regardning AppVeyor, I have no experience with Windoes, so don't know how to address the error message. I added a member to Config, which may be the reason, but I wonder why it built just fine on the (recent) Ubuntus.

'ankerl::nanobench::Config::Config(ankerl::nanobench::Config &&) noexcept': is not a special member function which can be defaulted

[jonas-schulze]

about const char* overloads, please add one, and the implementation can just call the std::string version then.

Given that there is a constructor std::string(char const*), this conversion should happen implicitly when calling context(std::string const&,...), shouldn't it? Passing a char const* to it works just fine. I am not opposed to add the overloads, I just want to understand their benefits.

Some more observations:

• Some overloads are not DRY, e.g. Bench::name
• Some methods calls are the other way around, i.e. std::string const& to char const*, e.g. Bench::unit and Bench::render (search for c_str)

Is that something that should be addressed?

[jonas-schulze]

I addressed all open issues. This is ready for review again. 🙂

[martinus]

Thanks a lot!