Inline GetCombinedNodeFlags and GetCombinedModifierFlags bodies

[Marius Schulz (mds-ant)]

This PR replaces the generic getCombinedFlags[T] helper with direct field reads in GetCombinedNodeFlags (node.Flags) and GetCombinedModifierFlags (node.ModifierFlags()). The parent-walk and OR-accumulation are unchanged; this just removes the indirect call through a function pointer at each step.

getCombinedFlags and the trivial getNodeFlags wrapper had no other callers and have been removed. Adds BenchmarkGetCombinedFlags over the standard fixture set.

This PR was assisted by Claude Code.

Performance

goos: darwin
goarch: arm64
pkg: github.com/microsoft/typescript-go/internal/ast
cpu: Apple M1 Pro
                                                                 │ /tmp/baseline.txt │           /tmp/patched.txt           │
                                                                 │      sec/op       │    sec/op     vs base                │
GetCombinedFlags/empty.ts-10                                            1.460n ± 14%   1.458n ± 15%        ~ (p=0.827 n=30)
GetCombinedFlags/checker.ts-10                                          666.9µ ±  0%   504.2µ ±  0%  -24.40% (p=0.000 n=30)
GetCombinedFlags/dom.generated.d.ts-10                                  246.0µ ±  1%   171.8µ ±  0%  -30.16% (p=0.000 n=30)
GetCombinedFlags/Herebyfile.mjs-10                                     11.435µ ±  0%   7.089µ ±  1%  -38.00% (p=0.000 n=30)
GetCombinedFlags/jsxComplexSignatureHasApplicabilityError.tsx-10        2.927µ ±  0%   1.913µ ±  1%  -34.61% (p=0.000 n=30)
geomean                                                                 6.037µ         4.434µ        -26.56%

[Copilot]

Pull request overview

Note

Copilot was unable to run its full agentic suite in this review.

This PR adds a benchmark for combined flag calculation and refactors GetCombined*Flags helpers to inline their logic (likely to reduce overhead and improve performance).

Changes:

• Added BenchmarkGetCombinedFlags to measure GetCombinedNodeFlags / GetCombinedModifierFlags over real fixture ASTs.
• Inlined combined-flag traversal logic into GetCombinedModifierFlags and GetCombinedNodeFlags, removing the generic helper and function-pointer indirection.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated 2 comments.

File	Description
internal/ast/utilities_bench_test.go	Adds a benchmark that parses fixture files, collects declarations, and measures combined flag computations.
internal/ast/utilities.go	Refactors combined flag computation by inlining logic for node vs modifier flags.

[Jake Bailey (jakebailey)]

This is a microoptimization; do you have anything that says this is hot code? The old times are already microseconds.

[Marius Schulz (mds-ant)]

This came up during a performance audit of the codebase. I agree the win is small, but I think it might still be worth landing given that (1) the relative win is measurable and (2) we're not introducing additional complexity or a maintenance burden.

[Jake Bailey (jakebailey)]

Hm, but as the copilot comment notes, this involves a copy paste

[Marius Schulz (mds-ant)]

The duplication is what it takes to unlock this optimization though. Any shared helper either reintroduces indirect calls or grows GetCombinedNodeFlags past the point where its own callers (IsVarConst, IsVarLet etc.) inline. The two functions are sitting right next to each other, so if the traversal logic ever changes, it should be easy to spot that it should change in two places.

But as you said, this is a microoptimization. If you'd rather not have this duplication (however small), we can close this PR.

[Marius Schulz (mds-ant)]

Jake Bailey (@jakebailey) Are you happy merging this, or would you rather not? The win is measurable in relative terms, but small in absolute terms.

[Jake Bailey (jakebailey)]

I'm still not sure; obviously it's measurable, but there's a balance between code duplication and how much this benefits real-world code.

[Marius Schulz (mds-ant)]

Yeah, that's fair. I'm happy to close this one out if you'd rather not take on the duplication.

[Jake Bailey (jakebailey)]

With optimizations I'm trying in #4903, this turns out to actually help!