fix(cpp): keep symbols for nested C++ types and C++/CLI sources (#2876) by rajarshidattapy · Pull Request #2886 · Graphify-Labs/graphify

rajarshidattapy · 2026-08-20T04:33:17Z

Two independent gaps in the C++ extractor, both reproduced verbatim from the issue.

1. Nested C++ types were dropped silently

A nested type is a field_declaration whose type field is the class_specifier. The C++ member-variable branch in extractors/engine.py consumed that node and returned before the walk could descend, so the nested type and everything it declares vanished — no parse error, no warning.

namespace N { class Outer { public: class Inner { public: static void Method() { } }; }; }

	labels
before	`['nested.h', 'Outer']`
after	`['nested.h', 'Outer', 'Inner', '.Method()']`

The fix walks the class_specifier with parent_class_nid threaded through, which routes it into the engine's existing nested-type handling — Inner gets a contains edge from Outer rather than from the file, exactly as for the other languages. The declarator loop still runs afterwards, so class Inner { } inst; yields both the type and the member.

2. C++/CLI took the whole class body down with it

tree-sitter-cpp implements none of ref class, Type^, Type%, gcnew or [assembly:…]. Because the ERROR lands on the type header, everything inside dissolves and recovery invents symbols that were never written:

namespace N {
    public ref class Wrapper {
    public:
        static void Init() { }
        static System::String^ Name() { return gcnew System::String(""); }
    };
}

	labels
before	`['cli.h', 'Wrapper()', 'public']` — plus `parse_errors` on line 2
after	`['cli.h', 'Wrapper', '.Init()', '.Name()', 'String']`, no parse errors

_normalize_cpp_cli() maps each C++/CLI spelling onto the nearest standard C++ one before parsing, via _extract_generic's existing source_override hook (the same one Vue SFCs use). Three properties keep it safe:

Byte-length preserving. Dropped tokens are overwritten with spaces, never deleted, and gcnew becomes new plus padding. Every offset, line and column still points at the same place in the file on disk, so reported source locations stay accurate.
Gated on a C++/CLI marker (ref/value class, interface class, gcnew, [assembly:). Plain C/C++/CUDA is parsed byte-for-byte as before — ^/% alone are not markers, since they are the ordinary XOR and modulo operators.
^/% are rewritten only in their attached suffix spelling (String^ s, int% n), so a spaced a ^ b / a % b in the same file keeps its meaning. Covered by a test.

Tests

tests/test_cpp_nested_and_cli.py — both issue reproducers, the class Inner { } inst; case, Outer contains Inner, byte-offset preservation (newline positions unchanged), plain C++ left untouched, and operators surviving inside a CLI file. The existing 82 C++/ObjC tests pass unchanged.

Not included

The issue's closing "unrelated, one line" note about --force not invalidating graphify-out/cache/ast/<version>. Worth noting for anyone hitting it: that directory is namespaced by graphify version (cache_dir() → v{_EXTRACTOR_VERSION}-s{_AST_CACHE_SCHEMA}), so a released extractor change invalidates it automatically. The stale-cache problem is specific to running a locally patched extractor at an unchanged version. Happy to add a --force AST-cache bypass in a separate PR if that is wanted.

…hify-Labs#2876) Two independent gaps in the C++ extractor, both hit by .NET interop wrappers (a 141-method C++/CLI class extracted 12 symbols). Nested types: a nested `class Inner { … };` is a `field_declaration` whose `type` field IS the class_specifier. The C++ member-variable branch in engine.py consumed that node and returned before the walk could descend, so Inner and everything it declares vanished — silently, with no parse error. Walk the class_specifier with parent_class_nid threaded through, which routes it into the engine's existing nested-type handling and gives it a `contains` edge from the enclosing type. The declarator loop still runs, so `class Inner { } inst;` yields both the type and the member. C++/CLI: tree-sitter-cpp implements none of `ref class`, `Type^`, `Type%`, `gcnew` or `[assembly:…]`, and because the ERROR lands on the type header the whole class body dissolves and recovery invents junk symbols. Normalize those spellings to the nearest standard C++ ones before parsing. The rewrite is byte-length preserving — dropped tokens are overwritten with spaces, `gcnew` becomes `new` plus padding — so every offset, line and column still points at the same place in the file on disk. It engages only on files carrying a C++/CLI marker, so plain C/C++/CUDA parses byte-for-byte as before, and `^`/`%` are rewritten only in their attached suffix spelling so XOR and modulo keep their meaning. Both reproducers from the issue now give the expected labels: `['nested.h', 'Outer', 'Inner', '.Method()']` and `['cli.h', 'Wrapper', '.Init()', '.Name()', 'String']`.

graphify-labs

Graphify reviewed this change.

Worth a look — the grounded gate found no coupling regressions or blocking issues, but 1 advisory finding(s) below merit a look before merge.

Formal verification. No changes could be formally verified in this run.

Graphify review — findings

Adds C++/CLI support to extract_cpp via a new _normalize_cpp_cli that rewrites ref/value class, Type^/Type%, gcnew, and [assembly:…] spellings into standard C++ using byte-length-preserving substitutions (spaces/padding) so offsets and source locations stay intact, gated behind a marker regex so plain C/C++ is untouched. Fixes nested-type extraction in _extract_generic: a field_declaration whose type field is a class body is now walked as a class (getting its contains edge) instead of being consumed by the member-variable branch and silently dropped. Adds tests/test_cpp_nested_and_cli.py covering nested types, type+instance declarations, CLI body survival, offset preservation, and operator (^/%) safety.

Worth a look

Multiline C++/CLI attributes destroy line offsets — graphify/extract.py:2008 · Escalate · medium
- agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review Execution auto-disposal is off for this run; enable it (with sandbox isolation) to have Graphify try to confirm or refute this automatically.

Analysis details — impact, health, verification

Impact & health

Graphify review

Impact — 1772 functions depend on the 418 functions this change touches.

Health — this change adds coupling hotspots:

new: extract() — 472 callers, 42 callees
new: _rebuild_code() — 98 callers, 51 callees
new: _extract_generic() — 18 callers, 24 callees
new: extract_xaml() — 19 callers, 17 callees
new: extract_objc() — 27 callers, 9 callees
new: extract_js() — 80 callers, 3 callees
new: dispatch_command() — 2 callers, 117 callees
new: _get_extractor() — 26 callers, 6 callees
…and 32 more — each is listed as a finding

Verification — 1772 functions in the blast radius were not formally verified this run (proofs are advisory here).

Gate & verification

graphify gate

PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.

Advisory (not blocking):

verification_scope: 1629 function(s) in the blast radius were not formally verified this run

Formal verification

Could not verify: Could not verify extract\_cpp.

The verifier did not have enough to check extract\_cpp, so it is saying so rather than guessing. No false assurance is the whole point.

Guarantee: No guarantee either way, this is an honest abstention, not a pass.

Note: Reason: parameter `path` is annotated `Path` — outside the synthesizable primitive/collection set

Could not verify: Could not verify \_extract\_generic.

The verifier did not have enough to check \_extract\_generic, so it is saying so rather than guessing. No false assurance is the whole point.

Guarantee: No guarantee either way, this is an honest abstention, not a pass.

Note: Reason: parameter `path` is annotated `Path` — outside the synthesizable primitive/collection set

· 1 grounded finding(s) anchored inline below; 39 more finding(s) on lines outside this diff (see the check run).

graphify-labs · 2026-08-20T04:40:04Z

+    return _CPP_CLI_ATTR_RE.sub(blank, out)
+
+
 def extract_cpp(path: Path) -> dict:


⚠️ Health regression — extract_cpp()

24 callers depend on it (afferent coupling).

Grounded coupling-delta finding (deterministic), not an LLM guess.

Review catch: byte length alone was not enough. `_CPP_CLI_ATTR_RE` matches with re.S, so `[assembly:AssemblyVersion(\n "1.0"\n)]` — ordinary formatting — was overwritten with spaces including its newlines. The `\s+` in the class-header pattern could swallow one the same way when the access specifier sits on its own line. The byte count stayed right and the file still parsed, but two source lines merged into one, so every symbol below the attribute reported a line number that was too low. That contradicts the property the rewrite is supposed to guarantee. Blank everything except CR and LF, so line and column stay stable along with the offset. Two tests: reported source_location for a class and method below a multiline attribute, and newline positions unchanged for a class header split across lines.

graphify-labs

Graphify reviewed this change.

Worth a look — the grounded gate found no coupling regressions or blocking issues, but 1 advisory finding(s) below merit a look before merge.

Formal verification. No changes could be formally verified in this run.

Graphify review — findings

Adds C++/CLI normalization to extract_cpp via a new _normalize_cpp_cli helper that rewrites ref class, Type^/%, gcnew, and [assembly:…] into standard C++ while preserving byte offsets and line breaks, so tree-sitter no longer dissolves the class body into junk nodes (#2876). Fixes _extract_generic to descend into nested type declarations (class Inner { … }; inside a class body) instead of consuming them as member variables, giving them a contains edge from the enclosing type. Adds tests/test_cpp_nested_and_cli.py covering nested types, CLI rewrites, operator preservation, and line-number stability.

Worth a look

_CPP_CLI_SUFFIX_RE rewrites '^'/'%' as XOR/mod inside a CLI file when attached to an identifier token — graphify/extract.py · Escalate · medium
- agreed by 2 of 2 members but NOT verified (no proof, no reproducing execution) — consensus is not a verdict; needs human review Execution auto-disposal is off for this run; enable it (with sandbox isolation) to have Graphify try to confirm or refute this automatically.

Analysis details — impact, health, verification

Impact & health

Graphify review

Impact — 1777 functions depend on the 423 functions this change touches.

Health — this change adds coupling hotspots:

new: extract() — 472 callers, 42 callees
new: _rebuild_code() — 98 callers, 51 callees
new: _extract_generic() — 18 callers, 24 callees
new: extract_xaml() — 19 callers, 17 callees
new: extract_objc() — 27 callers, 9 callees
new: extract_js() — 80 callers, 3 callees
new: dispatch_command() — 2 callers, 117 callees
new: _get_extractor() — 26 callers, 6 callees
…and 32 more — each is listed as a finding

Verification — 1777 functions in the blast radius were not formally verified this run (proofs are advisory here).

Gate & verification

graphify gate

PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.

Advisory (not blocking):

verification_scope: 1634 function(s) in the blast radius were not formally verified this run