Fix #748 (Move the argument count check into Macro::getMacroParameters()) - #749

AI review

I found no functional problems. The refactoring keeps the old behaviour: I compared the PR with base on a set of edge cases, and the output and diagnostics were identical every time. My only finding is that the new doc comment of getMacroParameters() is wrong in one case. For a function-like macro whose name is not followed by (, the function now throws wrongNumberOfParameters. The comment says it returns an empty vector. That happens when the name is the last token. The only caller hides this by returning early when the next token is not (, but anyone who reads the comment and calls the function will get the wrong idea.

Testing

• make -j8 testrunner simplecpp CXXOPTS=-Werror in pr/ built cleanly.
• ./testrunner: 0 failed, including the new define_wrong_number_of_arguments.
• python3 run-tests.py: 176 tests, 0 failed, 15 todos. base gives the same numbers.
• python3 -m pytest integration_test.py: 128 passed.
• ./selfcheck.sh: exit code 0.

I built base/ and ran both simplecpp binaries on scratch files. Every case gave identical output and diagnostics in base and the PR:

• a function-like macro name as the last token of the file (#define f(a) a\nf, also with a variadic macro)
• a call whose arguments are completed outside the #define (#define g f(1 / g,2))
• f x with parenthesised x inside another macro
• arguments that span several lines
• an unterminated f(1
• __VA_OPT__ with empty variadic arguments
• __COUNTER__ in arguments, with too few and with too many arguments

One limitation is not caused by this PR, since base behaves the same. A function-like macro name as the very last token, as in #define f(a) a\nf or ... f\n at the end of the file, gives "Wrong number of parameters". gcc -E simply outputs f.

Automated review of c764bc8 by Claude. The pull request was built and tested locally.