Thank you very much!

In some of my other repositories, it was suggested to switch to nu-ansi-term. What are the differences?

hyperfine does not show this because it pipes all output to /dev/null

That's what hyperfines --output <WHERE> option is for:

        --output <WHERE>
            Control where the output of the benchmark is redirected. <WHERE> can be:
            
            null: Redirect output to /dev/null (the default). Note that some programs like
            'grep' detect when standard output is /dev/null and apply certain
            optimizations. To avoid that, consider using '--output=pipe'.
            
            pipe: Feed the output through a pipe before discarding it.
            
            inherit: Don't redirect the output at all (same as '--show-output').
            
            <FILE>: Write the output to the given file.

You can use --output=inherit to measure hexyls colorization in combination with terminal rendering speed. By the way, on the hyperfine issue tracker, we're even thinking about a --output=pty option (or similar) which would attach the output to some pseudo terminal. Exactly for purposes like this one.

Thank you very much!

In some of my other repositories, it was suggested to switch to nu-ansi-term. What are the differences?

To be honest, I just picked the first option that was suggested on the advisory. From a cursory glance, nu-ansi-term seems to be a drop-in replacement with similar convenience methods, whereas anstyle is a lower-level API for ANSI color codes that maintains compatibility between different crates.

Currently, none of these libraries expose direct access to the color codes, instead returning an impl Display. In my first write, I rendered each code at the site, which created a 3x performance penalty. To avoid the allocation, this requires an awkward workaround where I have to render the codes as strings first globally and then refer to them later. I was planning on raising an issue to add these to the public API, since I'd rather not hardcode these values.

That's what hyperfines --output <WHERE> option is for:

Oh, I didn't know about this! This is very useful and I should have been using it when benchmarking performance to begin with...

Here's proof for the claims I made about random and text data.

Cool - thank you for the benchmark results. Is this not ready for review yet?

I wanted to test a different way of writing the Color prefixes that doesn't use once_cell.

This PR is ready to merge. It will probably cause conflicts with #170 so I think it would be best to merge that first, then I can make changes to this one so it does not conflict.

Thank you, sounds good.

#170 has been merged now. Thank you for your help @sharifhsn.

Thank you very much. This looks great. I know this will be next-to-impossible to properly test automatically (because the ANSI escape sequence output changed), but I trust you did some manual checks to make sure that the output still looks the same, color-wise?

Yes, I checked on both Linux and Windows (I don't have access to macOS).