C++: `struct Derived : Base` in `.h` files is misdetected as C (#1159 follow-up)

[Jaysenpeng]

Summary

A .h file whose only C++ syntax is struct Derived : Base is classified as C. The derived struct is missing from the index, and a phantom function Base with returnType=Derived is emitted at the derived declaration line.

This differs from #1159: the current heuristic handles struct ENGINE_API Derived : Base, but not the plain form without an export macro.

Minimal reproduction

Create src/min.h:

struct Base {};
struct Derived : Base {};

Run:

codegraph init .
codegraph query Derived --json
codegraph query Base --json

Actual:

Derived -> []
Base    -> struct Base (language=c, line 1)
           function Base (language=c, line 2, returnType=Derived)

Expected: Derived is indexed as a C++ struct, with no phantom function.

Renaming the file to .hpp, adding namespace marker {}, or mapping .h to cpp in codegraph.json makes Derived index correctly. This isolates the problem to .h language detection.

Root cause

looksLikeCpp() checks only the first 8192 characters and does not recognize a plain struct Name : Base clause. The file therefore falls back to the C grammar.

Suggested fix

Recognize class/struct base clauses as a C++ signal and scan the full source for that signal. Adding the pattern only to the existing 8192-character sample would still fail on large headers with a long C-compatible preamble.

Regression tests should cover the plain form, a base clause after character 8192, absence of the phantom function, and a genuine C bitfield control.

Environment

• @colbymchenry/codegraph 1.5.0
• Windows x64
• The same heuristic is present on current main at 81e1f4a92fdbd9413ba73cf72c8a5408102a7897

[colbymchenry]

Reproduced on current main exactly as described (Derived → nothing, phantom function Base with returnType=Derived, file recorded as C), and your root-cause read is right — the heuristic only knew the export-macro spelling of a base clause.

Fix in #1593: the .h check now also recognizes a plain class/struct … : Base base clause (with or without public/virtual, scoped or templated bases, final), and that check runs over the whole file with comments stripped rather than the first 8 KB, so a long C-compatible preamble can't hide it — your point about large headers. The C controls you asked for are in the tests (bit-fields, sizeof(struct foo) : 0 ternaries, struct_end: labels, doc-comment prose), plus an end-to-end check that Derived indexes as a C++ struct and no phantom Base function is minted. Thanks for the precise report — re-index after the next release to pick up affected headers.