Jason Marshall

So if I'm reading the code correctly, the current to_html code is in the NIF, so this would be quite a lot of work, built on parse_document?

Example: I don't want markdown links to open in the same browser tab, so I used some other comments and your examples to build this: https://gist.github.com/jdmarshall/563c8612ce3c78ff612ace30bd0c5cf7 That is for just...

Because the summary is calculated after teardown, I don't even have the option to force this behavior in my own tests.

That's what I'm using as the workaround. I'm definitely still having some sort of issue with ensuring V8 is completely warm, but then that's a very old benchmarking problem.

Do you suppose it would break anyone if you calculated the performance numbers prior to teardown? I have no way of knowing if anyone but you is depending on the...

That's where all the data goes, yes. But teardown is the last event I can see between tasks, and it is called immediately before the P## calculations are done. I...

I've gone down a rabbit hole. Send help. https://github.com/tdunning/t-digest Unfortunately the whole tinybench library is about the size of this algorithm (although it's in Java so maybe divide that by...

This one doesn't seem too bad. The actual logic part of it is about 2 computer screens, and it's based on a refinement of t-digest. It's also the only one...

Workaround: It looks like you could clear the extra statistics in the 'cycle' event (though to me 'teardown' sounds like the spot where one would expect to manipulate system state).

Nodejs now has builtin histogram support in its performance tools and I am slowly migrating over to those.