feat(semantic): add set of AstType that exist in AST

[camchenry]

• part of linter: dynamically skip running rules that do not have the relevant node types #12223

Adds a bitset that we compute as we push nodes. Since we have to call add_node for every AST node anyway, adding a small bit of code here to insert a single bit into the set should be very cheap. This enables us to do optimizations later on in the linter to quickly check if a file has any of a specific node kind. I've added some debug assertions to check that the bounds checks are respected so that we can do an insert_unchecked call. Not sure if this is good enough or not though.

[camchenry]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• perf(linter): add should_run based on node types (batch 3) #12231
• perf(linter): add should_run based on node types (batch 2) #12229
• perf(linter): add should_run based on node types (batch 1) #12228
• feat(semantic): add methods to check if AST contains any/all node types #12227
• feat(semantic): add set of AstType that exist in AST #12226 👈 (View in Graphite)
• feat(ast): compute max value that can be mapped to AstType #12225
• main

---

How to use the Graphite Merge Queue

Add either label to this PR to merge it via the merge queue:

• 0-merge - adds this PR to the back of the merge queue
• hotfix - for urgent hot fixes, skip the queue and merge this PR next

You must have a Graphite account in order to use the merge queue. Sign up using this link.

An organization admin has enabled the Graphite Merge Queue in this repository.

Please do not merge from GitHub as this will restart CI on PRs being processed by the merge queue.

This stack of pull requests is managed by Graphite. Learn more about stacking.

[codspeed-hq]

CodSpeed Instrumentation Performance Report

Merging #12226 will not alter performance

_{Comparing 07-11-feat_semantic_add_set_of_asttype_that_exist_in_ast (891c8d1) with 07-11-feat_ast_compute_max_value_that_can_be_mapped_to_asttype_ (ea30a17)}

Summary

✅ 34 untouched benchmarks

[graphite-app]

The bounds check should be updated to use <= instead of < since the FixedBitSet was created with capacity AST_TYPE_MAX + 1, making AST_TYPE_MAX a valid index. Consider changing to:

debug_assert!((kind.ty() as usize) <= AST_TYPE_MAX as usize);

This ensures the assertion correctly validates that the index is within the allocated capacity.

Suggested change

	debug_assert!((kind.ty() as usize) < self.node_kinds_set.len());
	debug_assert!((kind.ty() as usize) <= AST_TYPE_MAX as usize);

Spotted by Diamond

Is this helpful? React 👍 or 👎 to let us know.

[overlookmotel]

I have a branch I never got around to polishing and pushing containing a fixed-size bitset (genuinely fixed size - size is const, and determined at compile time).

It's a generalization of this thing:

https://github.com/oxc-project/oxc/blob/242e7210aa37e6eef14457b28aebbc31035d3c28/tasks/ast_tools/src/schema/derives.rs

That might be useful here as it stores the data on stack instead of heap, so should be a bit cheaper, and it encapsulates the unsafe code.

Here is a cut-down version, specialized for AstType:

const USIZE_BITS: usize = usize::BITS as usize;

/// Number of bytes required for bit set which can represent all [`AstType`]s.
const NUM_USIZES: usize = (AST_TYPE_MAX + 1).div_ceil(USIZE_BITS);

/// Bit set with a bit for each [`AstType`].
pub struct AstTypesBitset([usize; NUM_USIZES]);

impl AstTypesBitset {
    /// Create empty [`AstTypesBitset`] with no bits set.
    pub const fn new() -> Self {
        Self([0; NUM_USIZES])
    }

    /// Returns `true` if bit is set for provided [`AstType`].
    pub const fn has(self, ty: AstType) -> bool {
        let (index, mask) = Self::index_and_mask(ty);
        (self.0[index] & mask) != 0
    }

    /// Set bit for provided [`AstType`].
    pub const fn set(&mut self, ty: AstType) {
        let (index, mask) = Self::index_and_mask(ty);
        self.0[index] |= mask;
    }

    /// Returns `true` if any bit is set in both `self` and `other`.
    pub fn intersects(&self, other: &Self) -> bool {
        let mut intersection = 0;
        for (&a, &b) in self.0.iter().zip(other.0.iter()) {
            intersection |= a & b;
        }
        intersection != 0
    }

    /// Get index and mask for an [`AstType`].
    /// Returned `index` is guaranteed not to be out of bounds of the array.
    const fn index_and_mask(ty: AstType) -> (usize, usize) {
        let n = ty as usize;
        let index = n / USIZE_BITS;
        let mask = 1usize << (n % USIZE_BITS);
        (index, mask)
    }
}

Note: The unsafe code may not even be necessary. Maybe compiler can prove from its knowledge of AstType's variants that the index operations cannot be out of bounds, so self.0[index] produces same code as self.0.get_unchecked(index). You could check that in Godbolt.

[overlookmotel]

Yes, it looks like compiler is smart enough to remove the bounds checks. No unsafe needed. https://godbolt.org/z/EjPnEde3G

I've edited the code above to remove the unsafe.