cc @carlos-granados Hi Carlos, you might like this one... I'm making the immutable traverser part of the Rector core, so we have it without vendor patching 👍 Thanks for your innovation and work to make it happen 🚀

cc @staabm This is the immutable node traverser idea in new ImmutableNodeTraverser class. We saw 20-30 % performance improvements across various projects. I think PHPStan would benefit from this approach too. Tip for your next performance bottleneck hunting :)

Let's ship this to test in real projects 🚀

@TomasVotruba A very interesting take. Only problem I can see is if somehow Nikic's PhpParser greatly modified their own parser, but I don't really see that happening

@carlos-granados Thanks for feedback and welcome back 👍 (still travelling?)

I was checking, and past ~3 years only one constant was added. I think we can handle the update + make the node traverser even faster for our purposes with similar caching you've introduced.

huhu,

This is the immutable node traverser idea in new ImmutableNodeTraverser class. We saw 20-30 % performance improvements across various projects. I think PHPStan would benefit from this approach too. Tip for your next performance bottleneck hunting :)

@TomasVotruba I looked thru the PR, but was not able to pin-point what in this PR makes it faster compared to before this PR.

or does this PR implement the optimization which was already done in #6232 just in a different way?

IIRC phpstan is already doing something similar: see #6232 (comment)

@staabm Yes, it's same as #6232 , just without vendor patching. I pinged you now because I know vendor patching is very risky business (broken by PHPUnit recently), so would not recommend it.

This PR implements the same logic that is easy to reproduce without touching vendor 👍

As for PHPStan, lazy registry saves some performance. Not sure how it's implemented on traverser level.