Add Nix Language Support

[eyildirir]

Issue: #1049

This PR adds native Nix (.nix) parser and AST extractor support to graphify. This enables the tool to discover .nix files, parse them using tree-sitter-nix, extract code symbols (let-bindings, attributes, recursive attribute sets), and map relationships (such as relative imports and list-based module declarations) into the knowledge graph.

Key Changes

1. Parser & Extractor:

   • Added tree-sitter-nix to project dependencies in pyproject.toml.
   • Implemented AST traversal logic in extract_nix in graphify/extract.py to parse Nix expressions (bindings, import statements, recursive rec { ... } sets, and list attributes like imports = [ ... ]).

2. AST Build, Scoping, and Presentation:

   • Added Nix suffix checks and mapped them to the "nix" language family in detect.py, build.py, analyze.py, symbol_resolution.py, and callflow_html.py to prevent false-positive cross-language references and format Nix module labels and HTML descriptions correctly.

3. Verification:

   • Added automated tests in test_nix_extraction in tests/test_languages.py.

Manual Verification

Tested against a couple of nix based projects I already knew deep and wide.

To manually verify the changes against a Nix configuration project:

1. Install dependencies and the package:

   pip install -e .

2. Run extraction against a Nix project:

   graphify extract /path/to/nix/project

3. Inspect output:
   • Verify that communities are correctly grouped and the generated graphify-out/graph.html and GRAPH_REPORT.md show the extracted Nix attributes and import edges cleanly.

[safishamsi]

Thanks for the Nix support! The extractor logic is solid — the _get_import_argument traversal handles Nix's curried import ./dep.nix correctly and the imports = [ ... ] list case is a nice catch. A few things need addressing before merge:

Use _extract_generic + LanguageConfig

Every other language in graphify follows the _extract_generic(path, config: LanguageConfig) pattern — see extract_r, extract_lua, extract_zig for simple examples. The custom function/node/edge helper duplication inside extract_nix is ~60 lines that _extract_generic already provides. Please refactor to a _NIX_CONFIG = LanguageConfig(...) with the appropriate tree-sitter node type mappings and use _extract_generic to drive the walk. The custom import resolution logic (_add_import_edge, _handle_imports_list) can live as extra_walk on the config.

Add a fixture file

Tests use tmp_path inline. Add tests/fixtures/sample.nix instead — same pattern as sample.lua, sample.r, etc. The fixture should exercise bindings, an import, and at least one imports = [...] list so the tests are reproducible without constructing ad-hoc files.

Minor: duplicate branch in _add_import_edge

if path_str.startswith("./") or path_str.startswith("../"):
    resolved_path = (path.parent / path_str).resolve()
else:
    resolved_path = (path.parent / path_str).resolve()  # identical

Both arms do the same thing — the if/else can be collapsed to a single line.

tree-sitter-nix as a hard dependency

Adding it to dependencies means every pip install graphifyy downloads the package regardless of whether the user has any .nix files. Follow the existing soft-import pattern used by other optional grammars: catch ImportError and return {"nodes": [], "edges": [], "error": "tree-sitter-nix not installed"} (which the dispatcher already handles).

Once these are addressed this is ready to land.

[eyildirir]

Thanks for the Nix support! The extractor logic is solid — the _get_import_argument traversal handles Nix's curried import ./dep.nix correctly and the imports = [ ... ] list case is a nice catch. A few things need addressing before merge:

Use _extract_generic + LanguageConfig

Every other language in graphify follows the _extract_generic(path, config: LanguageConfig) pattern — see extract_r, extract_lua, extract_zig for simple examples. The custom function/node/edge helper duplication inside extract_nix is ~60 lines that _extract_generic already provides. Please refactor to a _NIX_CONFIG = LanguageConfig(...) with the appropriate tree-sitter node type mappings and use _extract_generic to drive the walk. The custom import resolution logic (_add_import_edge, _handle_imports_list) can live as extra_walk on the config.

Add a fixture file

Tests use tmp_path inline. Add tests/fixtures/sample.nix instead — same pattern as sample.lua, sample.r, etc. The fixture should exercise bindings, an import, and at least one imports = [...] list so the tests are reproducible without constructing ad-hoc files.

Minor: duplicate branch in _add_import_edge

if path_str.startswith("./") or path_str.startswith("../"):
    resolved_path = (path.parent / path_str).resolve()
else:
    resolved_path = (path.parent / path_str).resolve()  # identical

Both arms do the same thing — the if/else can be collapsed to a single line.

tree-sitter-nix as a hard dependency

Adding it to dependencies means every pip install graphifyy downloads the package regardless of whether the user has any .nix files. Follow the existing soft-import pattern used by other optional grammars: catch ImportError and return {"nodes": [], "edges": [], "error": "tree-sitter-nix not installed"} (which the dispatcher already handles).

Once these are addressed this is ready to land.

Addressed all 4 issues raised, following the example of how Lua, Swift and C# is handled. Deconflicted with HEAD as well. All tests pass again.

Ready for a second pass @safishamsi.

Thank you for your time!