fix: generate React component declarations

[chrisbbreuer]

What changed

• infer portable JSX.Element returns for function, arrow, nullable, fragment, member-tag, generic, and class components
• preserve forwardRef, memo, lazy, namespace, named, and type-only React imports when emitted declarations reference them
• scan JSX initializer nesting correctly so default exports and following declarations are not swallowed
• support named and nested higher-order component wrappers without broadening ordinary call inference
• cover both semantic inference and isolated-declarations paths, including the Zig emitter

Why

JSX brackets were previously interpreted as comparisons, arithmetic, or generic arguments. This produced incorrect number/boolean component returns, dropped wrapper imports, and could consume declarations following a JSX initializer.

Validation

• 20 focused React declaration tests pass across the TypeScript and Zig engines
• 313 Zig/Bun integration tests pass
• 36 native type-inference tests pass
• 26 native root scanner/emitter tests pass
• 1,259 of 1,260 dtsx tests pass; the remaining pre-existing Vue test references the deleted packages/dtsx/test/fixtures/vue/NFTBurnButton.vue fixture
• typecheck and build pass
• lint reports zero errors and one pre-existing Markdown warning

[netlify]

❌ Deploy Preview for dtsx failed. Why did it fail? →

Name	Link
🔨 Latest commit	aee2894
🔍 Latest deploy log	https://app.netlify.com/projects/dtsx/deploys/6a5ecd6a2c10b5000803a16b

[github-actions]

Benchmark Results

Platform: linux x64 | Runtime: Bun 1.3.14 | Date: 2026-07-21

In-Process API — Cached

Smart caching (hash check + cache hit) for watch mode, incremental builds, and CI.

Tool	Small (~50 lines)	Medium (~100 lines)	Large (~330 lines)	XLarge (~1050 lines)	XXLarge (~2000 lines)	Huge (~5000 lines)
dtsx (cached)	3.7 µs	1.8 µs	1.5 µs	2.6 µs	5.9 µs	11.7 µs
zig-dtsx	14.4 µs (3.9x)	19.2 µs (10.8x)	62.7 µs (42.2x)	341.1 µs (132.8x)	357.6 µs (61.1x)	928.8 µs (79.2x)
oxc-transform	20.1 µs (5.4x)	64.2 µs (36.1x)	213.7 µs (143.9x)	1.10 ms (429.2x)	891.8 µs (152.4x)	2.22 ms (189.1x)
TypeScript 7 (CLI)	95.47 ms (25571.5x)	100.95 ms (56780.1x)	105.64 ms (71158.5x)	128.24 ms (49910.3x)	119.18 ms (20371.9x)	152.08 ms (12975.4x)

In-Process API — No Cache

Raw single-transform comparison (cache cleared every iteration).

Tool	Small (~50 lines)	Medium (~100 lines)	Large (~330 lines)	XLarge (~1050 lines)	XXLarge (~2000 lines)	Huge (~5000 lines)
zig-dtsx	14.4 µs (3.7x)	19.2 µs (7.4x)	62.7 µs (38.1x)	341.1 µs (112.6x)	357.6 µs (60.4x)	928.8 µs (72.2x)
oxc-transform	20.1 µs (5.1x)	64.2 µs (24.6x)	213.7 µs (129.9x)	1.10 ms (364.0x)	891.8 µs (150.5x)	2.22 ms (172.4x)
dtsx (no-cache)	3.9 µs	2.6 µs	1.6 µs	3.0 µs	5.9 µs	12.9 µs
TypeScript 7 (CLI)	95.47 ms (24460.6x)	100.95 ms (38754.8x)	105.64 ms (64226.4x)	128.24 ms (42332.4x)	119.18 ms (20118.7x)	152.08 ms (11829.5x)

Note: TypeScript 7 is CLI-only. Each measurement includes ~40ms process spawn overhead, so it is not directly comparable to the in-process tools above. It is reported separately from the in-process tools.

Multi-File Project

All tools processing files in-process sequentially.

Tool	50 files	100 files
zig-dtsx	5.78 ms (51.3x)	12.05 ms (20.1x)
dtsx	112.7 µs	598.4 µs
oxc-transform	17.27 ms (153.2x)	36.31 ms (60.7x)
tsc	537.79 ms (4771.6x)	1.03 s (1726.9x)

No previous benchmark found for regression comparison

zig-dtsx vs oxc-transform

Input Size	zig-dtsx	oxc-transform	Speedup
Small (~50 lines)	14.4 µs	20.1 µs	🟢 1.39x
Medium (~100 lines)	19.2 µs	64.2 µs	🟢 3.33x
Large (~330 lines)	62.7 µs	213.7 µs	🟢 3.41x
XLarge (~1050 lines)	341.1 µs	1.10 ms	🟢 3.23x
XXLarge (~2000 lines)	357.6 µs	891.8 µs	🟢 2.49x
Huge (~5000 lines)	928.8 µs	2.22 ms	🟢 2.39x

Internal Benchmark Details

Extraction

Benchmark	Avg	Min	Max	Throughput	Memory
Simple (0001.ts)	1.9 µs	942 ns	4.4 µs	380.7 M chars/s	0.0 MB
Medium (0002.ts)	6.3 µs	2.2 µs	13.0 µs	155.5 M chars/s	0.0 MB
Complex (0003.ts) 🏆	1.6 µs	911 ns	7.5 µs	821.6 M chars/s	0.0 MB
Very Complex (0005.ts)	7.6 µs	6.7 µs	11.4 µs	11.0 G chars/s	0.0 MB
Lodash-like (real-world)	4.4 µs	2.1 µs	20.5 µs	5.9 G chars/s	0.0 MB
React-like (real-world)	3.9 µs	3.4 µs	6.8 µs	8.5 G chars/s	0.0 MB

Synthetic

Benchmark	Avg	Min	Max	Throughput	Memory
100 lines 🏆	1.6 µs	872 ns	19.2 µs	4.5 G chars/s	0.0 MB
500 lines	3.9 µs	2.6 µs	18.2 µs	9.3 G chars/s	0.0 MB
1000 lines	5.0 µs	4.5 µs	17.9 µs	14.3 G chars/s	0.0 MB
5000 lines	28.2 µs	22.2 µs	56.2 µs	13.2 G chars/s	0.0 MB
10000 lines	55.8 µs	47.3 µs	80.2 µs	13.4 G chars/s	0.0 MB

Memory

Benchmark	Avg	Min	Max	Throughput	Memory
Large File Memory 🏆	18.62 ms	18.62 ms	18.62 ms	40.3 M chars/s	0.0 MB

Real-World

Benchmark	Avg	Min	Max	Throughput	Memory
Lodash-like 🏆	3.3 µs	2.9 µs	7.4 µs	7.7 G chars/s	0.0 MB
React-like	4.6 µs	3.6 µs	12.7 µs	7.3 G chars/s	0.0 MB

Generation

Benchmark	Avg	Min	Max	Throughput	Memory
Single File (100 lines) 🏆	1.14 ms	843.9 µs	1.69 ms	6.2 M chars/s	0.6 MB
Medium File (1000 lines)	3.72 ms	3.26 ms	4.20 ms	19.4 M chars/s	0.1 MB
Large File (5000 lines)	16.89 ms	14.14 ms	19.92 ms	22.1 M chars/s	15.1 MB

Phase Timing

File	Phase	Avg Time	% of Total
500 lines	Processing	1.38 ms	████████████████████ 99.5%
500 lines	Extraction	5.3 µs	█ 0.4%
500 lines	Formatting	211 ns	█ 0.0%
500 lines	File Read	88 ns	█ 0.0%
2000 lines	Processing	3.49 ms	████████████████████ 99.6%
2000 lines	Extraction	12.2 µs	█ 0.3%
2000 lines	Formatting	230 ns	█ 0.0%
2000 lines	File Read	131 ns	█ 0.0%
5000 lines	Processing	8.69 ms	████████████████████ 99.7%
5000 lines	Extraction	28.7 µs	█ 0.3%
5000 lines	Formatting	237 ns	█ 0.0%
5000 lines	File Read	78 ns	█ 0.0%

Metric	Value
Total benchmarks	17
Avg time	2.38 ms
Total time	0.3 s