Performance facility

Hey,

I just released a tiny library where I really focused on perf and optimizations: https://github.com/ngryman/tree-crawl.

I was thinking about bringing some stuff I’ve learned on the way here. I think we could add some benchmark, profiling and tracing to the repo.

Multiple goals:

• Keep track of performance evolution.
• Isolate chunks of code that are potentially slow and optimize them.
• Make the code optimizer friendly (monomorphic, avoid deopts and bailouts).

@jbucaran I would gladly focus on this part if you are interested 😃

PS: My first optimization would probably focus on the stack used in h. Basically implementing a really basic stack and avoiding pop as much as possible gave a nice +15% boost in my library.

[ngryman]

@jbucaran Is the string/number concatenation feature of h mandatory? Correct me if I’m wrong, but except if you are using h directly (which is unlikely), you don’t need such a facility. The perf/memory impact of this feature is huge compared to the usage IMHO.

EDIT: I emphasis on concatenation, but the real potential bottleneck is the variable number of parameters for children. Couldn’t we just have an array of children?

[jorgebucaran]

@ngryman Closing in favor of your other issue #183.