perf: use FxHashSet for O(1) name lookups in declaration instantiation#5024
Open
tkshsbcue wants to merge 5 commits intoboa-dev:mainfrom
Open
perf: use FxHashSet for O(1) name lookups in declaration instantiation#5024tkshsbcue wants to merge 5 commits intoboa-dev:mainfrom
tkshsbcue wants to merge 5 commits intoboa-dev:mainfrom
Conversation
Replace Vec with FxHashSet for name-tracking collections in bytecode compiler declaration instantiation. 22 .contains() calls go from O(n) to O(1), eliminating O(n^2) behavior when compiling scripts with many declarations. Affected: global/eval/function declaration instantiation functions.
Contributor
Author
|
@jedel1043 i think you might fight this interesting haha! |
Test262 conformance changes
Fixed tests (2):Broken tests (157):Tested main commit: |
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #5024 +/- ##
===========================================
+ Coverage 47.24% 59.76% +12.51%
===========================================
Files 476 589 +113
Lines 46892 63632 +16740
===========================================
+ Hits 22154 38027 +15873
- Misses 24738 25605 +867 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
Contributor
|
do you have benchmark results? 👀 |
Contributor
Author
|
hey hi @zhuzhu81998 these are the benchmarks i compiled
|
jedel1043
added
A-Performance
Member
|
Need to fix formatting first |
Test262 conformance changes
Tested main commit: |
Contributor
Author
|
hey @jedel1043 fixed formatting! |
Contributor
|
The names that are hashed here, they are not internal right? They come from JS? |
Contributor
Author
|
hey @zhuzhu81998 The keys here are Sym, which is just a NonZeroUsize it's an integer index into the string interner, not the raw JS string itself. FxHash on sequential/near-sequential |
zhuzhu81998
suggested changes
Apr 7, 2026
| @@ -718,6 +710,7 @@ impl ByteCompiler<'_> { | |||
|
|
|||
| // 12. Let declaredVarNames be a new empty List. | |||
| let mut declared_var_names = Vec::new(); | |||
Contributor
There was a problem hiding this comment.
do we still need this separately?
Contributor
Author
There was a problem hiding this comment.
Are you suggesting we drop the Vec and use only the FxHashSet here?
Contributor
There was a problem hiding this comment.
yea unless there is a reason to keep this that i am not seeing? you dropped the Vec elsewhere already no?
Member
There was a problem hiding this comment.
We could try to drop it, but IIRC the spec requires vars to be initialized in a certain order, so I don't know if using a hash set will break assumptions about this
Contributor
Author
There was a problem hiding this comment.
Hmm, looking at it the iteration at step 18 just calls CreateGlobalVarBinding for each name, and those bindings are independent of each other. Wouldn't the end result be the same regardless of order?
Contributor
Author
There was a problem hiding this comment.
uhhh what should i do then lmao?
Member
There was a problem hiding this comment.
Are you suggesting we drop the Vec and use only the FxHashSet here?
This, but use a separate commit just in case it breaks tests. Makes reversing easier
Contributor
Author
There was a problem hiding this comment.
alright i will implement this and push another follow up commit !
Remove the redundant Vec alongside the FxHashSet for declared_var_names in global_declaration_instantiation and eval_declaration_instantiation. The Vec was only iterated at the end, and iteration order is not required by the spec for these bindings.
zhuzhu81998
reviewed
Apr 17, 2026
Comment on lines
259
to
264
| // 2. Append fn to declaredFunctionNames. | ||
| if declared_function_names.insert(name.sym()) { | ||
| // SKIP: 1. If varEnv is a Global Environment Record, then | ||
|
|
||
| // 2. Append fn to declaredFunctionNames. | ||
| declared_function_names.push(name.sym()); | ||
|
|
||
| // SKIP: 3. Insert d as the first element of functionsToInitialize. | ||
| } |
Contributor
There was a problem hiding this comment.
do we still need this empty if block?
Contributor
Author
There was a problem hiding this comment.
Haha Good catch after the Vec→FxHashSet swap the body is just // SKIP: spec-comment markers, so the if wraps nothing. I'll drop it (same applies to the matching blocks at lines 93, 422, 668, and 944) and keep just declared_function_names.insert(name.sym());.
| if !strict { | ||
| let lexically_declared_names = lexically_declared_names(body); | ||
| let lexically_declared_names: FxHashSet<Sym> = | ||
| lexically_declared_names(body).into_iter().collect(); |
Contributor
There was a problem hiding this comment.
the doc of lexically_declared_names says:
/// Returns a list with the lexical bindings of a node, which may contain duplicates.
///
/// This is equivalent to the [`LexicallyDeclaredNames`][spec] syntax operation in the spec.
///
/// [spec]: https://tc39.es/ecma262/#sec-static-semantics-lexicallydeclarednames
what cases will result in duplicates here? 🤔
Contributor
Author
There was a problem hiding this comment.
Duplicates can occur in the !strict branch via Annex B e.g. sibling blocks each declaring the same function name, or repeated function bindings across nested CaseClauses. That said, dedup isn't really the goal of the change here; the loop below does .contains(&f) for every Annex B function name, so the win is O(1) lookup vs O(n) scan. I'll update the commit message / PR description to make that the stated motivation rather than "duplicates"
Contributor
There was a problem hiding this comment.
my concern is whether or not the preservation of duplicates is necessary here for correctness/spec-compliance
Contributor
Author
There was a problem hiding this comment.
No dedup is safe here. The only use of lexically_declared_names in this block is the .contains(&f) membership check in the Annex B loop below:
for f in annex_b_function_declarations_names(body) {
if !lexically_declared_names.contains(&f) {
...
}
}
Addresses review feedback: after the container swap, two sites had the
shape `if set.insert(x) { /* only SKIP comments */ }`, where the boolean
return of insert was no longer meaningful. Collapse those to a plain
`set.insert(x);` and move the SKIP spec-step comments above the call to
preserve the spec-structure reading order.
zhuzhu81998
approved these changes
Apr 21, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Description
This PR replaces
VecwithFxHashSetfor name-tracking collections in the bytecode compiler's declaration instantiation functions, reducing.contains()lookups from O(n) to O(1).The problem
During script/module/function compilation, the bytecode compiler tracks declared function names, variable names, and parameter bindings using
Vec<Sym>. These vectors are repeatedly checked with.contains()inside loops, creating O(n^2) behavior when a script has many declarations.There are 22
.contains()calls indeclarations.rs, many inside nested loops:The fix
Replace
Vec<Sym>withFxHashSet<Sym>for all name-tracking collections used only for membership checks. Where a collection is also iterated (e.g.,declared_var_namesfor emitting bindings), a parallelFxHashSetis maintained alongside theVec.FxHashSet(fromrustc_hash) is already used elsewhere in the codebase and provides faster hashing thanstd::collections::HashSetfor small integer keys likeSym.Affected functions
global_declaration_instantiation_contextdeclared_function_names,declared_var_namesprepare_eval_declaration_instantiationdeclared_function_names,lexically_declared_namesglobal_declaration_instantiationdeclared_function_names,declared_var_nameseval_declaration_instantiationdeclared_function_names,declared_var_namesfunction_declaration_instantiationfunction_names,parameter_bindings,instantiated_var_namesWhy this matters
Declaration instantiation runs on every script, eval, and function compilation. Real-world JavaScript bundles routinely have hundreds or thousands of declarations. The O(n^2) behavior is particularly visible when:
eval()with significant codeTest plan
cargo checkpassescargo clippy-- no warningscargo test -p boa_engine --lib-- all 958 tests pass