perf(formatter): reduce AstNode size by 8 bytes using thread-local allocator

[Dunqing]

Memory Benchmark Results

Summary

This benchmark measures memory usage during formatting before and after removing the allocator field from AstNode (using thread-local storage instead).

Metric	Value
AstNode size	40 bytes → 32 bytes (-20%)
Total formatting overhead BEFORE	33,231,380 bytes (31.7 MB)
Total formatting overhead AFTER	31,221,021 bytes (29.8 MB)
Memory saved	2,010,359 bytes (1.9 MB)
Reduction	6.0%

Per-File Results

File	Source Size	BEFORE	AFTER	Saved	%
RadixUIAdoptionSection.jsx	2.5 KB	159.0 KB	151.1 KB	7.9 KB	5.0%
errors.ts	8.2 KB	336.6 KB	322.6 KB	14.0 KB	4.2%
Search.tsx	11.7 KB	789.8 KB	746.3 KB	43.5 KB	5.5%
core.js	12.3 KB	776.2 KB	731.1 KB	45.1 KB	5.8%
next.ts	17.3 KB	1.1 MB	1.0 MB	100.0 KB	9.1%
index.tsx	30.5 KB	1.8 MB	1.7 MB	100.0 KB	5.6%
handle-comments.js	27.2 KB	1.3 MB	1.3 MB	0.0 KB	0.0%
types.ts	76.1 KB	7.5 MB	7.1 MB	400.0 KB	5.3%
App.tsx	345.5 KB	21.3 MB	20.1 MB	1.2 MB	5.6%
TOTAL	531.3 KB	35.0 MB	33.1 MB	1.9 MB	5.4%

Benchmark Files

Files sourced from real-world projects:

• RadixUIAdoptionSection.jsx - Small JSX (61 lines)
• errors.ts - Vue.js compiler-core (191 lines)
• Search.tsx - Outline app (389 lines)
• core.js - Prettier main/core (426 lines)
• next.ts - Next.js server (619 lines)
• index.tsx - Next.js client (995 lines)
• handle-comments.js - Prettier comments handler (1092 lines)
• types.ts - Hono types (2370 lines)
• App.tsx - Excalidraw App component (11180 lines)

How to Reproduce

# Create measurement tool
cat > tasks/benchmark/examples/mem_measure.rs << 'EOF'
use oxc_allocator::Allocator;
use oxc_formatter::{FormatOptions, Formatter, SortImportsOptions, get_parse_options};
use oxc_parser::Parser;
use oxc_tasks_common::TestFiles;

fn main() {
    for file in TestFiles::formatter().files() {
        let allocator = Allocator::default();
        let program = Parser::new(&allocator, &file.source_text, file.source_type)
            .with_options(get_parse_options())
            .parse()
            .program;
        let after_parse = allocator.used_bytes();
        let format_options = FormatOptions {
            experimental_sort_imports: Some(SortImportsOptions::default()),
            ..Default::default()
        };
        let _ = Formatter::new(&allocator, format_options).build(&program);
        let after_format = allocator.used_bytes();
        println!("{}: {} -> {}", file.file_name, after_parse, after_format);
    }
}
EOF

# Run benchmark
cargo run -p oxc_benchmark --example mem_measure --release

[Dunqing]

• perf(formatter): reduce AstNode size by 8 bytes using following_span_start #18347
• perf(formatter): reduce AstNode size by 8 bytes using thread-local allocator #18340 👈 (View in Graphite)
• 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 Performance Report

Merging this PR will not alter performance

_{Comparing 01-21-perf_formatter_reduce_astnode_size_by_8_bytes_using_thread-local_allocator (c5a2ae8) with main (88e0896)¹}

Summary

✅ 38 untouched benchmarks
⏩ 7 skipped benchmarks²

Footnotes

1. No successful run was found on main (04a2871) during the generation of this report, so 88e0896 was used instead as the comparison base. There might be some changes unrelated to this pull request in this report. ↩

2. 7 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[Dunqing]

Not worth it as it has to store the allocator in the thread_local, but the benefits are negligible.