[danmar]

No description provided.

[dmcppcheck]

AI review

The PR only changes test.cpp. All tests pass, and I checked every TODO comment's claim against gcc and clang (the R/X name at the end of the file, the empty token from r(,)/t(,,), ==0 after the comment, a"s" should be an error). Each claim is correct, and every "wanted" value matches the reference preprocessors.

There is one new finding. tokenMacro10 uses ASSERT_EQUALS to lock in empty Token::macro for the results of # and ##, and its comment presents that as a general rule. It only holds for function-like macros. Object-like macros give the pasted token the macro name (#define XY x##y gives xy@XY). So does a function-like macro whose pasted token is built from body text, as in cat2(x) → xb with an empty name. cppcheck uses this field, so this looks like a known gap that should be a TODO, not a rule. I found no other problems.

Testing

Build and test suites

• make -j8 testrunner simplecpp CXXOPTS=-Werror built without warnings.
• ./testrunner passed with exit code 0 and no failures.
• python3 run-tests.py passed: 176 tests, 21 skipped, 15 todos with 0 fixed, 0 failed.
• python3 -m pytest integration_test.py passed: 128 tests.
• ./selfcheck.sh passed with exit code 0.

The TODO claims, checked against gcc and clang

• #define R() / R: gcc -E -P outputs R, so the wanted value "\nR" with no error is right.
• cat(a,"s"): gcc (in both C and C++11 mode) reports pasting "a" and ""s"" does not give a valid preprocessing token, and clang reports a similar error. simplecpp outputs a"s" with no error.
• { r(,) }: gcc outputs { }. simplecpp outputs { }, with an empty token between the braces.
• The str(strncmp(...) // this goes away\n == 0) example: gcc gives ... '\\4') == 0". simplecpp gives '\\4') ==0".

Token::macro for # and ## results

I linked a small program against simplecpp.o and printed str@macro for each output token:

#define str(x) #x
#define XY x##y
#define cat(a,b) a##b
#define cat2(a) a##b
XY cat(x,y) cat2(x) str(z)

Output: xy@XY xy@ xb@ "z"@. The object-like macro keeps its macro name on the pasted token. In the function-like macros the name is empty, even for xb, which is pasted with the body token b.

---

Automated review of 7ac04ea by Claude. The pull request was built and tested locally.

[dmcppcheck]

AI review

This comment presents a general rule, but it only holds for function-like macros. In an object-like macro the pasted token keeps the macro name: #define XY x##y / XY gives xy@XY, while cat(x,y) gives xy with an empty macro. A paste that takes part of its text from the body also loses the name: #define cat2(a) a##b / cat2(x) gives xb with macro empty.

cppcheck uses Token::macro to recognise code that comes from a macro. This looks like an inconsistency rather than intended behaviour, so ASSERT_EQUALS would lock it in. I'd use TODO_ASSERT_EQUALS here (or at least a comment that says it is the current behaviour), and add an object-like ## case so that the difference is covered.