fix(objc): do not report an @interface class extension as a function

[Eljees]

Refs #305 — this fixes the part of that report that is still reproducible today.

Where the report stands

Running the reporter's .mm file against master (0b97a9a) shows most of #305 has been
fixed since 2020: init, doSomethingWithCPPobj: and dealloc are found, and the bogus
if@20-23 entry is gone. One line of their output survives, though — the very first one:

$ python lizard.py OCCFile.mm
  NLOC    CCN   token  PARAM  length  location
     4      1      9      0       4 OCCFile@9-12@OCCFile.mm      <-- not a function
     8      2     22      0       8 init@18-25@OCCFile.mm
     5      1     25      0       5 doSomethingWithCPPobj:@27-31@OCCFile.mm
     5      1     11      0       6 dealloc@33-38@OCCFile.mm

Lines 9-12 are the class extension with its instance-variable block:

@interface OCCFile ()
{
    CppFile* mCppFile;
}
@end

Cause

ObjCStates has no notion of @interface / @protocol. The tokens are
['@', 'interface', 'OCCFile', '(', ')', '{', ...], which for the C-like states is
indistinguishable from a function definition OCCFile () { ... }, so a function is created.

The @interface Foo : NSObject { ... } form escapes by accident: the : knocks the machine
out of _state_function before the { arrives. That is why the bug is specific to class
extensions and categories — @interface Foo () and @interface Foo (Private).

Fix

A method body can only appear in @implementation; @interface and @protocol are pure
declarations. The reader now skips such a section up to its @end, so its contents never
reach the C-like states.

Tests

New Test_objc_interface in test/test_languages/testObjC.py:

• class extension with instance variables;
• category with instance variables;
• a trimmed version of the reporter's file — expects exactly ["init", "dealloc"]; before the
  change it returned ["OCCFile", "init", "dealloc"];
• a @protocol guard, which passed before the change too and is kept as a regression test.

Following .cursor/rules/issue.mdc, the whole suite was run before and after:
python -m pytest → 1313 passed, 7 skipped, 1 failed. The single failure,
test/testFilesFilter.py::TestFilesFilter::test_gitignore_filter, is pre-existing — it
fails on a clean checkout of 0b97a9a as well (checked with git stash) and is unrelated to
this change. pycodestyle lizard_languages/objc.py is clean.

I left the issue open rather than auto-closing it, in case you read the original report more
broadly than the output it quotes.

---

Prepared with AI assistance, following the workflow in .cursor/rules/. Every claim above was
verified by running it rather than assumed.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 617f13fee6

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[Eljees]

Following up on the 13 August change — the three @protocol forms you raised now behave identically on 241374a and on current master, so the skip-until-@end state no longer swallows anything:

source	master (3f9270d)	this branch (241374a)
@protocol Foo; then @implementation Bar	['doIt']	['doIt']
@protocol(NSObject) inside an expression	['f']	['f']
real @protocol definition then @implementation	['qux']	['qux']
@interface OCCFile () — the #305 case	['OCCFile', 'init']	['init']

Only the last row differs: the bogus OCCFile entry is gone and nothing else moved. Suite is 1316 passed / 7 skipped, CI is green 4/4, and the branch still merges cleanly.

Ready whenever you have time to look.