Windows: Stack overflow (0xc00000fd) in tree-sitter SQL parser on large .sql files during mode=full index

[Sajid11194]

Environment

• OS: Windows 11 Pro
• Version: 0.4.10
• Go runtime: 1.26.1 (from stack trace)

Description

index_repository with mode=full crashes with a Windows stack overflow exception (0xc00000fd = STATUS_STACK_OVERFLOW) when the project contains a large .sql file (~4.5 MB). The crash occurs in the CGo tree-sitter parser (cbm_extract_file) during pass2.definitions.

This also means the MCP server always crashes on startup, since the server triggers a mode=full pipeline automatically after the notifications/initialized handshake. Claude Code marks the server as "failed" every session, making the MCP completely unusable without a workaround.

Steps to Reproduce

1. Have a project containing a .sql file ≥ ~4 MB (e.g. a database dump with large INSERT statements)
2. Run:

   codebase-memory-mcp cli index_repository '{"repo_path": "/path/to/project", "mode": "full"}'
   

3. Observe segfault / exit code 139

Confirmed reproducible — removing the file from indexing via .cbmignore makes the crash disappear immediately.

Stack Trace

Exception 0xc00000fd 0x1 0xcd87803f18 0x7ff7c2ff7f5d
PC=0x7ff7c2ff7f5d
signal arrived during external code execution

runtime.cgocall(0x7ff7c2feb1e0, 0x2dd73843fbc8)
	C:/hostedtoolcache/windows/go/1.26.1/x64/src/runtime/cgocall.go:167 +0x3e
github.com/DeusData/codebase-memory-mcp/internal/cbm._Cfunc_cbm_extract_file(0x2dd738882000, 0x4533a3, 0x21, ...)
	_cgo_gotypes.go:467 +0x51
github.com/DeusData/codebase-memory-mcp/internal/cbm.ExtractFile(...)
	cbm.go:254 +0x18b
github.com/DeusData/codebase-memory-mcp/internal/pipeline.cbmParseFileFromSource(...)
	pipeline_cbm.go:46 +0x1bc
github.com/DeusData/codebase-memory-mcp/internal/pipeline.cbmParseFile(...)
	pipeline_cbm.go:30 +0x109
github.com/DeusData/codebase-memory-mcp/internal/pipeline.(*Pipeline).passDefinitions.func1()
	pipeline.go:1093 +0x10b

Root Cause (hypothesis)

The tree-sitter SQL grammar recurses deeply on large files (bulk INSERT dumps, etc.), exhausting the C stack. Windows has a smaller default thread stack than Linux/macOS, making this Windows-specific. mode=fast is unaffected because it skips large files.

Workaround

Add a .cbmignore file in the repo root to exclude large SQL files:

path/to/large-dump.sql

Expected Behavior

The parser should either:

• Detect and skip files above a configurable size threshold before passing to tree-sitter
• Increase the C stack size for the parser goroutine (runtime/debug.SetMaxStack or spawning a thread with a larger stack)
• Gracefully recover from C-level stack overflows instead of crashing the whole process

[kingchenc]

Fixed in PR #66.

Added per-language file size guard in cbmParseFile():

• SQL files > 1 MB → skipped (logged as cbm.skip.large_sql)
• Any file > 4 MB → skipped (logged as cbm.skip.large_file)

This prevents the tree-sitter SQL grammar from deeply recursing on bulk INSERT dumps and exhausting the C stack. The .cbmignore workaround is no longer needed for this case.