Recover C/C++ functions inside preprocessor blocks and error nodes

[Tomer-PL]

Summary

• Tree-sitter's C grammar places function definitions inside preproc_if, preproc_ifdef, preproc_else, preproc_elif, and ERROR container nodes rather than at the translation unit's top level
• The C and C++ parsers only walked root.children, silently missing all functions nested inside these containers
• The fix recurses into these container nodes in _visit_top_level_node for both parsers

Why this happens

When tree-sitter encounters #if/#else blocks with alternative code paths (common in cross-platform C code), it nests all subsequent content inside the preprocessor node. For example:

#if defined(_WIN32)
unsigned int __stdcall worker(void * arg)
#else
void * worker(void * arg)
#endif
{
    return 0;
}

// This function ends up inside the preproc_else node:
void important_api(int x) { ... }

Similarly, #if/#elif/#else chains inside static initializer arrays produce ERROR nodes that swallow all subsequent function definitions.

Impact

Tested on libavif v1.4.1 (24K lines of C):

Metric	Before	After
Nodes	818	1,015 (+197)
Functions	680	798 (+118)
Call edges	6,339	7,558 (+1,219)

118 functions across 14 files were previously invisible. Key recoveries:

File	Recovered	Notable functions
src/reformat.c	12	avifImageYUVToRGB (public YUV→RGB API, 9 cross-module callers)
src/read.c	11	avifParseMinimizedImageBox (cc=100, highest complexity on decode path)
apps/shared/avifjpeg.c	24	JPEG I/O with platform-specific code
src/reformat_libyuv.c	18	Entire libyuv integration (all functions behind #ifdef)
src/codec_aom.c	14	AOM codec interface
src/obu.c	8	AV1 bitstream parsing

avifParseMinimizedImageBox at cc=100 is the new highest-complexity function on the decode path — it was entirely invisible before this fix because it sits inside #if defined(AVIF_ENABLE_EXPERIMENTAL_MINI).

Test plan

• 2 new tests in test_c_parser.py: #ifdef branch recovery and #if/#else split-signature recovery
• All 1,043 existing tests pass
• Validated on real-world C codebase (libavif)

🤖 Generated with Claude Code