Fix #709 Do not evaluate #elif after a taken group - #710

AI review

The change to the #if/#elif condition is correct. An #if is still evaluated only when the enclosing group is active (as before), and an #elif is now evaluated only when its state is ElseIsTrue. The state handling after the evaluation still gives the right result. I found no wrong preprocessor output, and the reported case now matches gcc and clang.

New findings:

1. Side effect on the macroUsage and ifCond outputs, which cppcheck uses. An #elif after a taken group is no longer evaluated, so macros in its condition are no longer reported in macroUsage, and the #elif no longer gets an IfCond entry. Take #define A 1 / #if 1 / #elif A / #endif. Before the PR, A was reported as used at line 4. With the PR, A is not reported at all. Tools built on macro usage (for example the unused-macro check / MISRA 2.5 through the dump) may now see such macros as unused. If this is intended, it should be a deliberate decision with a test. Otherwise, record maybeUsedMacros for the skipped #elif.
2. The fix is incomplete for an empty condition. The check !sameline(rawtok,rawtok->next) ("Syntax error in #elif") runs before the new skip logic. So #if 1 / #elif / #endif still aborts with an error and empty output, while gcc and clang accept it. This is not a regression; the base version behaves the same.

Testing

• I built /work/pr with make -j8 testrunner simplecpp CXXOPTS=-Werror without warnings. ./testrunner exits with 0. run-tests.py shows 0 failed (21 skipped, 16 todos). pytest integration_test.py shows 128 passed. ./selfcheck.sh exits with 0.
• I also built /work/base for comparison.
• Combined input (scratch/a.c) with these cases:
  • #elif 1/0 and #elif garbage( after a taken group
  • #elif 1/0 after a taken #elif
  • a nested #if/#elif 1/0 inside a skipped group
  • #elif defined( after a false #ifdef
  • Results: gcc and clang print a f i j (and report errors for the malformed defined(, which is in a group that must be evaluated). Base aborts with division/modulo by zero. The PR stops at an empty #elifdef with no name (line 7).
• Empty #elif after a taken group (not a regression; base is the same):

  #if 1
  a
  #elif
  b
  #endif

  /work/pr/simplecpp b.c prints b.c:3: syntax error: Syntax error in #elif and no code. gcc -E -P b.c and clang -E -P b.c print a without diagnostics. The same happens when the #if 1 / #elif is nested inside #if 0 (c.c); gcc and clang print x.
• Macro usage and IfCond through the API. I wrote a small driver, scratch/dm.cpp, that calls simplecpp::preprocess(out,raw,files,cache,dui,&ol,&mu,&ic) and prints mu and ic. Input d.c:

  #define A 1
  #define B 1
  #if 1
  #elif A
  #elifdef B
  #endif

  Base output:

  usage A line 4
  usage A line 4
  ifcond line 3 1=1
  ifcond line 4 1=1
  

  PR output:

  ifcond line 3 1=1
  

Comments

simplecpp.cpp:3829

This check runs before the new skip logic, so an empty #elif after a taken group still aborts preprocessing:

#if 1
a
#elif
b
#endif

simplecpp prints Syntax error in #elif and gives no output. gcc and clang print a. The same happens when the #if 1/#elif is inside an #if 0 group. Since the PR is about not evaluating such #elif lines, consider applying the check only when the condition will actually be evaluated.

Automated review of a3e956c by Claude. The pull request was built and tested locally.